[144] 3rd Gen Intel® Xeon® Scalable processor outperforms Graviton2 by up to 1.31x in Server-side Java (critical) (Intel Xeon based AWS c6i instance outperforms Graviton2 c6g for 16vCPU, 32vCPU, and 64vCPU).

1.27x more operations per second in Server-side Java workload (critical) for16vCPU, 1.31x for 32vCPU, and 1.04x for 64vCPU, Workload Server-Side Java, Other SW:OpenJDK17.0.1​,Ubuntu 20.04.4 LTS, Kernel 5.19.0-1025-aws.

New :c6i.4xlarge 32GB memory capacity/instance, c6i.8xlarge 64 GB memory capacity/instance, c6i.16xlarge128 GB memory capacity/instance (Xeon).

Baseline :c6g.4xlarge 32 GB memory capacity/instance, c6g.8xlarge 64 GB memory capacity/instance, c6g.16xlarge128GB memory capacity/instance (Graviton2).

Test by Intel.

New:May19, 2023

Baseline:May 19, 2023

New : c6i.xlarge 7 GB memory capacity/instance, c6i.2xlarge 16 GB memory capacity/instance, c6i.4xlarge 32 GB memory capacity/instance, c6i.8xlarge 64 GB memory capacity/instance, c6i.16xlarge 129 GB memory capacity/instance (Xeon).

Baseline : c6g.xlarge 7 GB memory capacity/instance, c6g.2xlarge 16 GB memory capacity/instance, c6g.4xlarge 32 GB memory capacity/instance, c6g.8xlarge 64 GB memory capacity/instance, c6g.16xlarge 129 GB memory capacity/instance (Graviton2).

Test by Intel on Apr 12, 2022.

New: April 12, 2022

Baseline: April 12, 2022

New : c6i.xlarge 7 GB memory capacity/instance, c6i.2xlarge 16 GB memory capacity/instance, c6i.4xlarge 32 GB memory capacity/instance, c6i.8xlarge 64 GB memory capacity/instance, c6i.16xlarge 129 GB memory capacity/instance (Xeon).

Baseline : c7g.xlarge 7 GB memory capacity/instance, c7g.2xlarge 16 GB memory capacity/instance, c7g.4xlarge 32 GB memory capacity/instance, c7g.8xlarge 64 GB memory capacity/instance, c7g.16xlarge 129 GB memory capacity/instance (Graviton3).

Test by Intel on Apr 12, May 25, 2022.

New: April 12, 2022

Baseline: May 25, 2022

New : m6i.2xlarge 32 GB memory capacity/instance, m6i.4xlarge 64 GB memory capacity/instance, m6i.8xlarge 128 GB memory capacity/instance, m6i.12xlarge 192 GB memory capacity/instance, m6i.16xlarge 256 memory capacity/instance, m6i.24xlarge 384 GB memory capacity/instance, m6i.32xlarge 512 GB memory capacity/instance (Xeon).

Baseline : m6g.2xlarge 32 GB memory capacity/instance, m6g.4xlarge 64 GB memory capacity/instance, m6g.8xlarge 128 GB memory capacity/instance, m6g.12xlarge 192 GB memory capacity/instance, m6g.16xlarge 256 GB memory capacity/instance (Graviton2). M6g does not offer Graviton2 in 96vCPU or 128vCPU sizes.

Test by Intel on Oct 16, 2021.

New : c6i.xlarge 8 GB memory capacity/instance, c6i.2xlarge 16 GB memory capacity/instance, c6i.4xlarge 32 GB memory capacity/instance, c6i.8xlarge 64 GB memory capacity/instance, c6i.16xlarge 128 GB memory capacity/instance (Xeon).

Baseline : c6g.xlarge 8 GB memory capacity/instance, c6g.2xlarge 16 GB memory capacity/instance, c6g.4xlarge 32 GB memory capacity/instance, c6g.8xlarge 64 GB memory capacity/instance, c6g.16xlarge 128 GB memory capacity/instance (Graviton2).

Test by Intel on Mar 25-29, Apr 18, 2022.

New: March 25, March 29, and April 18, 2022

Baseline: March 25, 2022

New : c6i.xlarge 8 GB memory capacity/instance, c6i.2xlarge 16 GB memory capacity/instance, c6i.4xlarge 32 GB memory capacity/instance, c6i.8xlarge 64 GB memory capacity/instance, c6i.16xlarge 128 GB memory capacity/instance (Xeon).

Baseline : c6g.xlarge 8 GB memory capacity/instance, c6g.2xlarge 16 GB memory capacity/instance, c6g.4xlarge 32 GB memory capacity/instance, c6g.8xlarge 64 GB memory capacity/instance, c6g.16xlarge 128 GB memory capacity/instance (Graviton2).

Test by Intel on Mar 25, Apr 18, 2022.

New: March 25 and April 18, 2022

Baseline: March 25, 2022

New : c6i.xlarge 8 GB memory capacity/instance, c6i.2xlarge 16 GB memory capacity/instance, c6i.4xlarge 32 GB memory capacity/instance, c6i.8xlarge 64 GB memory capacity/instance, c6i.16xlarge 128 GB memory capacity/instance (Xeon).

Baseline : c7g.xlarge 8 GB memory capacity/instance, c7g.2xlarge 16 GB memory capacity/instance, c7g.4xlarge 32 GB memory capacity/instance, c7g.8xlarge 64 GB memory capacity/instance, c7g.16xlarge 128 GB memory capacity/instance (Graviton3).

Test by Intel on Mar 25, Apr 18, May 23, 2022.

New: March 25 and April 18, 2022

Baseline: May 23, 2022

New : r6i.xlarge 32 GB memory capacity/instance, r6i.2xlarge 64 GB memory capacity/instance, r6i.4xlarge 128 GB memory capacity/instance, r6i.8xlarge 256 GB memory capacity/instance, r6i.16xlarge 512 GB memory capacity/instance (Xeon).

Baseline : r6g.xlarge 32 GB memory capacity/instance, r6g.2xlarge 65 GB memory capacity/instance, r6g.4xlarge 130 GB memory capacity/instance, r.6g.8xlarge 260 GB memory capacity/instance, r6g.16xlarge 521 GB memory capacity/instance (Graviton2).

Test by Intel on Mar 25-29, Apr 18, 2022.

New: March 29 and April 18, 2022

Baseline: March 25, 2022

New : r6i.xlarge 32 GB memory capacity/instance, r6i.2xlarge 64 GB memory capacity/instance, r6i.4xlarge 128 GB memory capacity/instance, r6i.8xlarge 256 GB memory capacity/instance, r6i.16xlarge 512 GB memory capacity/instance (Xeon).

Baseline : r6g.xlarge 32 GB memory capacity/instance, r6g.2xlarge 65 GB memory capacity/instance, r6g.4xlarge 130 GB memory capacity/instance, r.6g.8xlarge 260 GB memory capacity/instance, r6g.16xlarge 521 GB memory capacity/instance (Graviton2).

Test by Intel on Mar 25-29, Apr 18, 2022.

New: March 29 and April 18, 2022

Baseline: March 25, 2022

New : c6i.xlarge 8 GB memory capacity/instance, c6i.2xlarge 16 GB memory capacity/instance, c6i.4xlarge 32 GB memory capacity/instance, c6i.8xlarge 64 GB memory capacity/instance, c6i.16xlarge 128 GB memory capacity/instance (Xeon).

Baseline : c6g.xlarge 8 GB memory capacity/instance, c6g.2xlarge 16 GB memory capacity/instance, c6g.4xlarge 32 GB memory capacity/instance, c6g.8xlarge 64 GB memory capacity/instance, c6g.16xlarge 128 GB memory capacity/instance (Graviton2).

Test by Intel on Nov 30, 2021, Mar 14-22, 2022.

New: November 30, 2021 and March 22, 2022

Baseline: March 14 and 15, 2022

New : c6i.xlarge 8 GB memory capacity/instance, c6i.2xlarge 16 GB memory capacity/instance, c6i.4xlarge 32 GB memory capacity/instance, c6i.8xlarge 64 GB memory capacity/instance, c6i.16xlarge 128 GB memory capacity/instance (Xeon).

Baseline : c6g.xlarge 8 GB memory capacity/instance, c6g.2xlarge 16 GB memory capacity/instance, c6g.4xlarge 32 GB memory capacity/instance, c6g.8xlarge 64 GB memory capacity/instance, c6g.16xlarge 128 GB memory capacity/instance (Graviton2).

Test by Intel on Nov 30, 2021, Mar 14-22, 2022.

New: November 30, 2021 and March 22, 2022

Baseline: March 14 and 15, 2022

New : c6i.8xlarge 64 GB memory capacity/instance, c6i.16xlarge 128 GB memory capacity/instance (Xeon).

Baseline : c7g.8xlarge 64 GB memory capacity/instance, c7g.16xlarge 129 GB memory capacity/instance (Graviton3).

Test by Intel on Nov 30, 2021, Mar 22, May 24-25, 2022.

New: November 30, 2021 and March 22, 2022

Baseline: May 24 and 25, 2022

New : c6i.xlarge 8 GB memory capacity/instance, c6i.2xlarge 16 GB memory capacity/instance, c6i.4xlarge 32 GB memory capacity/instance, c6i.8xlarge 64 GB memory capacity/instance, c6i.16xlarge 129 GB memory capacity/instance (Xeon).

Baseline : c6g.xlarge 8 GB memory capacity/instance, c6g.2xlarge 16 GB memory capacity/instance, c6g.4xlarge 32 GB memory capacity/instance, c6g.8xlarge 64 GB memory capacity/instance, c6g.16xlarge 129 GB memory capacity/instance (Graviton2).

Test by Intel on May 27-28, 2022.

New: May 27 and 28, 2022

Baseline: May 27 and 28, 2022

New : c6i.xlarge 8 GB memory capacity/instance, c6i.2xlarge 16 GB memory capacity/instance, c6i.4xlarge 32 GB memory capacity/instance, c6i.8xlarge 64 GB memory capacity/instance, c6i.16xlarge 129 GB memory capacity/instance (Xeon).

Baseline : c6g.xlarge 8 GB memory capacity/instance, c6g.2xlarge 16 GB memory capacity/instance, c6g.4xlarge 32 GB memory capacity/instance, c6g.8xlarge 64 GB memory capacity/instance, c6g.16xlarge 129 GB memory capacity/instance (Graviton2).

Test by Intel on May 26-28, 2022.

New: May 26 and 28, 2022

Baseline: May 27 and 28, 2022

Up to 26x better performance per watt vs. AMD Milan for object detection (SSD-ResNet34)

New: 1-node Supermicro 220U-TNR, 2x Intel® Xeon® Platinum 8358 Scalable processor (32C), HT on, Turbo on, SNC off, Total Memory: 1 TB (16 slots/ 64GB/ 3200), ucode: x2a0, Intel X540-T2 , 1 SATA SSD, 1x P5510 NVMe, SSD-RESTNET34, Intel optimized Tensor Flow v2.7, Ubuntu 20.04LTS (5.4.0-120-generic).

Baseline: 1-node Supermicro AS-2124US-TNRP, 2x AMD EPYC Processor (32C 7543), SMT On, Boost ON, NPS=1, Total Memory: 1 TB (16 slots/ 64GB/ 3200), ucode: 0xa00111d, Intel X540-T2 , 1 SATA SSD, 1x P5510 NVMe, SSD-RESNET34, TF_​v2.7_​ZenDNN_​v3.2, Ubuntu 20.04LTS (5.4.0-120-generic). Test by Intel on April 2022.

Up to 1.4x better performance per watt vs. AMD Milan for life science applications (geomean of LINPACK, NAMD, LAMMPS with Intel® AVX-512)

LINPACK:

New: 1-node Supermicro 220U-TNR, 2x Intel® Xeon® Platinum 8358 Scalable processor (32C), HT on (1T/core), Turbo on, SNC on, Total Memory: 1 TB (16 slots/ 64GB/ 3200), ucode: x2a0, Intel X540-T2 , 1 SATA SSD, 1x P5510 NVMe, App Version: The Intel Distribution for LINPACK Benchmark; Build notes: Tools: Intel MPI 2019u7; threads/core: 1; Turbo: used; Build: build script from Intel Distribution for LINPACK package; 1 rank per NUMA node: 1 rank per socket, Ubuntu 20.04.3 (5.4.0-92-generic).

Baseline: 1-node Supermicro AS-2124US-TNRP, 2x AMD EPYC Processor (24C 7443, 32C 7543), SMT On (1T/core), Boost ON, NPS=4, Total Memory: 1 TB (16 slots/ 64GB/ 3200), ucode: 0xa00111d, Intel X540-T2 , 1 SATA SSD, 1x P5510 NVMe, App Version: AMD official HPL 2.3 MT version with BLIS 2.1; Build notes: Tools: hpc-x 2.7.0; threads/core: 1; Turbo: used; Build: pre-built binary (gcc built) from https://developer.amd.com/amd-aocl/blas-library/; 1 rank per L3 cache, 4 threads per rank, Ubuntu 20.04.3 (5.4.0-92generic).

NAMD:

New: 1-node Supermicro 220U-TNR, 2x Intel® Xeon® Platinum 8358 Scalable processor (32C), HT on, Turbo on, SNC on, Total Memory: 1 TB (16 slots/ 64GB/ 3200), ucode: x2a0, Intel X540-T2 , 1 SATA SSD, 1x P5510 NVMe, App Version: 2.15-Alpha1 (includes AVX tiles algorithm); Build notes: Tools: Intel MKL , Intel C Compiler 2020u4, Intel MPI 2019u8, Intel Threading Building Blocks 2020u4; Build knobs: -ip -fp-model fast=2 -no-prec-div -qoverride-limits -qopenmp-simd -O3 -xCORE-AVX512 -qopt-zmm-usage=high, Red Hat Enterprise Linux 8.5 (4.18.0-348.12.2.el8_​5.x86_​64).

Baseline: 1-node Supermicro AS-2124US-TNRP, 2x AMD EPYC Processor (24C 7443, 32C 7543), SMT On, Boost ON, NPS=4, Total Memory: 1 TB (16 slots/ 64GB/ 3200), ucode: 0xa00111d, Intel X540-T2 , 1 SATA SSD, 2x P5510 NVMe, App Version: 2.15-Alpha1 (includes AVX tiles algorithm); Build notes: Tools: Intel MKL , AOCC 2.2.0, gcc 9.3.0, Intel MPI 2019u8; Build knobs: -O3 -fomit-frame-pointer -march=znver1 -ffast-math, Red Hat Enterprise Linux 8.5 (4.18.0-348.12.2.el8_​5.x86_​64).

LAMMPS:

New: 1-node Supermicro 220U-TNR, 2x Intel® Xeon® Platinum 8358 Scalable processor (32C), HT on, Turbo on, SNC on, Total Memory: 1 TB (16 slots/ 64GB/ 3200), ucode: x2a0, Intel X540-T2 , 1 SATA SSD, 1x P5510 NVMe, App Version: v2020-10-29; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4, Intel MPI 2019u8; Build knobs: -O3 -ip -xCORE-AVX512 -qopt-zmm-usage=high, Red Hat Enterprise Linux 8.5 (4.18.0-348.12.2.el8_​5.x86_​64)

Baseline: 1-node Supermicro AS-2124US-TNRP, 2x AMD EPYC Processor (32C 7543), SMT On, Boost ON, NPS=4, Total Memory: 1 TB (16 slots/ 64GB/ 3200), ucode: 0xa00111d, Intel X540-T2 , 1 SATA SSD, 2x P5510 NVMe, App Version: v2020-10-29; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4, Intel MPI 2019u8; Build knobs: -O3 -ip -march=core-avx2, Red Hat Enterprise Linux 8.5 (4.18.0-348.12.2.el8_​5.x86_​64).

Tested by Intel on January-February 2022.

Up to 2x better performance per watt vs. AMD Milan for networks (key exchange NGINX RSA2K)

New: 1-node Supermicro X12DPG-QT6, 1x Intel® Xeon® Gold 6338N Scalable processor (32C), HT off, Turbo off, SNC off, Total Memory: 128 GB (8 slots/ 16GB/ 3200), ucode: x332, 1x Intel E810 2x100G, 1 SATA SSD, Ubuntu 20.04LTS (5.4.0-67-generic), NGINX 1.20.1, OpenSSL 1.1.1f, RSA2K, TLS Handshake

Baseline: 1-node Supermicro H12DSi-N6, 1x AMD EPYC Processor (32C 7513), SMT off, Boost off, NPS=2 , Total Memory: 128 GB (8 slots/ 16GB/ 3200), ucode:0xa001143, 1x Intel E810 2x100G, 1 SATA SSD, Ubuntu 20.04LTS (5.4.0-67-generic), NGINX 1.20.1, OpenSSL 1.1.1f, RSA2K, TLS Handshake. Tested by Intel on May 2022.

Up to 1.07x better performance per watt vs. AMD Milan for new orders per minute (MySQL)

New: 1-node Supermicro 220U-TNR, 2x Intel® Xeon® Platinum 8358 Scalable processor (32C), HT on, Turbo on, SNC off, Total Memory: 1 TB (16 slots/ 64GB/ 3200), ucode: x2a0, Intel X540-T2 , 1 SATA SSD, 2x P5510 NVMe, HammerDB 4.3, MySQL 8.0.27, Ubuntu 20.04LTS (5.4.0-120-generic).

Baseline: 1-node Supermicro AS-2124US-TNRP, 2x AMD EPYC Processor (32C 7543), SMT On, Boost ON, NPS=1, Total Memory: 1 TB (16 slots/ 64GB/ 3200), ucode: 0xa00111d, Intel X540-T2 , 1 SATA SSD, 2x P5510 NVMe, HammerDB 4.3, MySQL 8.0.27, Ubuntu 20.04LTS (5.4.0-120-generic). Tested by Intel Feb-Mar 2022.

Baseline: 1-node Supermicro H12DSi-N6, 1x AMD EPYC Processor (32C 7513), SMT off, Boost off, NPS=2 , Total Memory: 128 GB (8 slots/ 16GB/ 3200), ucode: 0xa001143, 1x Intel E810 2x100G, 2x E810 1x100G, 1 SATA SSD, Ubuntu 20.04LTS (5.4.0-67-generic), VPP IPSec 21.01. Tested by Intel May 2022.

Integer throughput/Floating Point throughput/STREAM/LINPACK

Baseline: Feb 04, 2021

Baseline: 5-node, 2x Intel Xeon Platinum 8260L processor on Wolf Pass with 384 GB (12 slots/ 32GB/ 2933[2933]) total DDR4 memory, ucode 0x4003006, HT on, Turbo on, CentOS 7.9.2009, 1x Intel S4510 SSD 1.92TB, Intel P4510 SSD 2TB, 1x Intel 10GbE integrated X722 , Splunk Perf Kit 0.4.21, Splunk 8.13, Splunk Operator 1.0.0, Minio Operator v2.0.9, Splunk Cluster - 3 Indexers, 2 search heads, test by Intel on 04/16/2021.

1.74x higher INT8 batch inference throughput on BERT-Large SQuAD with 3rd Gen Intel® Xeon® Scalable processor supporting Intel® DL Boost vs. prior generation

Baseline: Feb 17, 2021

1.66x higher INT8 batch inference throughput on MobileNet-v1 with 3rd Gen Intel® Xeon® Scalable processor supporting Intel® DL Boost vs. prior generation

Baseline: Feb 17, 2021

1.56x higher INT8 batch inference throughput on ResNet50 with 3rd Gen Intel® Xeon® Scalable processor supporting Intel® DL Boost vs. prior generation

Baseline: Feb 17, 2021

Baseline: Feb 17, 2021

1.42x higher INT8 batch inference throughput on SSD-MobileNet-v1 with 3rd Gen Intel® Xeon® Scalable processor supporting Intel® DL Boost vs. prior generation

Baseline: Feb 17, 2021

Baseline: Feb 17, 2021

Baseline: March 15, 2021

OpenVINO FP32 model running on Intel® Xeon® Platinum 8380 CPU @ 2.30GHz gives 2.2X latency improvement over baseline on Intel® Xeon® Platinum 8280 CPU @ 2.70GHz.

OpenVINO FP32 model running on Intel® Xeon® Platinum 8380 CPU @ 2.30GHz gives 1.6X throughput improvement over baseline on Intel® Xeon® Platinum 8280 CPU @ 2.70GHz.

Custom Deep Learning based Encoder Decoder model: Optimized:New: Tested by Intel as of 03/25/2021. 2 socket Intel® Xeon® Platinum 8380 Processor, 40 cores per socket, Ucode 0xd000270, HT On, Turbo On, OS Ubuntu 18.04.5 LTS, Kernel 5.4.0-65-generic, Total Memory 256GB, BIOS SE5C6200.86B.0022.D08.2103221623, Framework: Intel OpenVINO toolkit 2021.2.185, Python 3.6.13, Intel-openmp 2021.1.2, Numpy 1.19.5, GCC 7.5.0, model – custom Autoencoder

Baseline: Tested by Intel as of 03/25/2021. 2 socket Intel® Xeon® Platinum 8280 Processor, 28 cores per socket, Ucode 0x5003003, HT On, Turbo On, OS Ubuntu 18.04.5 LTS, Kernel 5.4.0-65-generic, Total Memory 384GB, BIOS SE5C620.86B.02.01.0011.032620200659, Framework: Intel OpenVINO toolkit 2021.2.185, Python 3.6.13, Intel-openmp 2021.1.2, Numpy 1.19.5, GCC 7.5.0, model – custom Autoencoder

Baseline: March 25, 2021

Predictive Analytics using XGBoost

Optimized:New: Tested by Intel as of 02/24/2021. 2 socket Intel® Xeon® Platinum 8380 Processor, 40 cores per socket, Ucode 0x8d05a260, HT On, Turbo On, OS Ubuntu 18.04.5 LTS, Kernel 5.4.0-65-generic, Total Memory 256GB, BIOS SE5C6200.86B.3021.D40.2103160200, Framework: XGBoost 1.3.3, Intel-openmp 2020.2, Intel MKL 2020.2, Numpy 1.19.2 (Intel), Pandas 1.2.1 (Intel), scikit-learn 0.23.2 (Intel), Anaconda Python 3.7.9, GCC 7.5.0, model trained – GBT Classifier, custom train data

Baseline: Tested by Intel as of 02/24/2021. 2 socket Intel® Xeon® Platinum 8280 Processor, 28 cores per socket, Ucode 0x5003003, HT On, Turbo On, OS Ubuntu 18.04.5 LTS, Kernel 5.4.0-65-generic, Total Memory 384GB, BIOS SE5C620.86B.02.01.0011.032620200659, Framework: XGBoost 1.3.3, Intel-openmp 2020.2, Intel MKL 2020.2, Numpy 1.19.2 (Intel), Pandas 1.2.1 (Intel), scikit-learn 0.23.2 (Intel), Anaconda Python 3.7.9, GCC 7.5.0, model trained – GBT Classifier, custom train data

Baseline: Feb 25, 2021

Baseline: 5-node, 2x Intel Xeon Platinum 8260L processor on Wolf Pass with 384 GB (12 slots/ 32GB/ 2933[2933]) total DDR4 memory, ucode 0x4003006, HT on, Turbo on, CentOS 7.9.2009, 1x Intel S4510 SSD 1.92TB, Intel P4510 SSD 2TB, 1x Intel 10GbE integrated X722 , Splunk Perf Kit 0.4.21, Splunk 8.13, Splunk Operator 1.0.0, Minio Operator v2.0.9, Splunk Cluster - 3 Indexers, 2 search heads, test by Intel on 04/16/2021.

Baseline: 5-node, 2x Intel Xeon Platinum 8260L processor on Wolf Pass with 384 GB (12 slots/ 32GB/ 2933[2933]) total DDR4 memory, ucode 0x4003006, HT on, Turbo on, CentOS 7.9.2009, 1x Intel S4510 SSD 1.92TB, Intel P4510 SSD 2TB, 1x Intel 10GbE integrated X722 , Splunk Perf Kit 0.4.21, Splunk 8.13, Splunk Operator 1.0.0, Minio Operator v2.0.9, Splunk Cluster - 3 Indexers, 2 search heads, test by Intel on 04/16/2021.

1.48X average performance gains on Splunk Parallel Search workload - Ice Lake vs. Cascade Lake: New Config: 5-node, 2x Intel Xeon Platinum 8380 processor on Coyote Pass with 512 GB (16 slots/ 32GB/ 3200[3200]) total DDR4 memory, ucode 0x8d05a260, HT on, Turbo on, Ubuntu 18.04.5, 5.4.0-65-generic, 1x Intel S4510 SSD 1.92TB, Intel P4510 SSD 2TB, 1x Intel X540-T2 10Gb, Dogfood 2007 dataset 103gb, Splunk 8.1.3, batch_​search_​max_​pipeline=16, test by Intel on 04/16/2021.

Baseline: 5-node, 2x Intel Xeon Platinum 6258R processor on Wolf Pass with 384 GB (12 slots/ 32GB/ 2933[2933]) total DDR4 memory, ucode 0x5003003, HT on, Turbo on, Ubuntu 18.04.5, 5.4.0-65-generic, 1x Intel S4510 SSD 1.92TB, Intel P4510 SSD 2TB, 1x Intel 10GbE integrated X722 , Dogfood 2007 dataset 103gb, Splunk 8.1.3, batch_​search_​max_​pipeline=16, test by Intel on 04/22/2021.

Up to 1.53x higher FSI Kernel performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

Up to 1.60x higher Life and Material Science performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

Up to 1.41x higher HPCG performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

Up to 1.38x higher HPL performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

Up to 1.47x higher STREAM Triad Performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

Up to 1.58x higher WRF performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

Up to 1.28x higher Binomial Options performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

Up to 1.67x higher Black Scholes performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

Up to 1.70x higher Monte Carlo performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

Up to 1.51x higher OpenFOAM performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

Up to 1.64x higher GROMACS performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

Up to 1.60x higher LAMMPS performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

Up to 1.57x higher NAMD performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

Up to 1.61x higher RELION Plasmodium Ribosome performance on 3rd Gen Intel Xeon Scalable platform vs. prior gen

New: 8380: 1-node, 2x Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x055261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96 . Tested by Intel between March 12, 2021 and March 29, 2021.

Baseline: 8280: 1-node, 2x Intel Xeon Platinum 8280 (28C/2.7GHz, 205W TDP) processor on Intel Software Development Platform with 192GB (12 slots/ 16GB/ 2933) total DDR4 memory, ucode 0x4002f01, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG48 . Tested by Intel between February 1, 2021 to February 20, 2021.

1.53x higher HPC performance (geomean HPL, HPCG, STREAM Triad, WRF, Binomial Options, Black Scholes, Monte Carlo, OpenFOAM, GROMACS, LAMMPS, NAMD, RELION)

1.53x higher FSI Kernel performance (geomean Binomial Options, Black Scholes, Monte Carlo)

1.60x higher Life and Material Science performance (geomean GROMACS, LAMMPS, NAMD, RELION)

1.41x higher HPCG performance App Version: 2019u5 MKL; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel MPI 2019u8; threads/core: 1; Turbo: used; Build knobs: -O3 -ip -xCORE-AVX512

1.38x higher HPL performance App Version: The Intel Distribution for LINPACK Benchmark 2019u5; Build notes: threads/core: 1; Turbo: used; Build: build script from Intel Distribution for LINPACK package; 1 rank per NUMA node: 1 rank per socket

1.47x higher STREAM Triad Performance App Version: McCalpin_​STREAM_​OMP-version; Build notes: Tools: Intel C Compiler 2019u5; threads/core: 1; Turbo: used; BIOS settings: HT=off Turbo=On SNC=On

1.58x higher WRF performance (geomean Conus-12km, Conus-2.5km, NWSC-3-NA-3km) App Version: 4.2.2; Build notes: Intel Fortran Compiler 2020u4, Intel MPI 2020u4; threads/core: 1; Turbo: used; Build knobs:-ip -w -O3 -xCORE-AVX2 -vec-threshold0 -ftz -align array64byte -qno-opt-dynamic-align -fno-alias $(FORMAT_​FREE) $(BYTESWAPIO) -fp-model fast=2 -fimf-use-svml=true -inline-max-size=12000 -inline-max-total-size=30000

1.28x higher Binomial Options performance App Version: v1.0; Build notes: Tools: Intel C Compiler 2020u4, Intel Threading Building Blocks ; threads/core: 2; Turbo: used; Build knobs: -O3 -xCORE-AVX512 -qopt-zmm-usage=high -fimf-domain-exclusion=31 -fimf-accuracy-bits=11 -no-prec-div -no-prec-sqrt

1.67x higher Black Scholes performance App Version: v1.3; Build notes: Tools: Intel MKL, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4; threads/core: 1; Turbo: used; Build knobs: -O3 -xCORE-AVX512 -qopt-zmm-usage=high -fimf-precision=low -fimf-domain-exclusion=31 -no-prec-div -no-prec-sqrt -fimf-domain-exclusion=31

1.70x higher Monte Carlo performance App Version: v1.1; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4; threads/core: 1; Turbo: used; Build knobs: -O3 -xCORE-AVX512 -qopt-zmm-usage=high -fimf-precision=low -fimf-domain-exclusion=31 -no-prec-div -no-prec-sqrt

1.51x higher OpenFOAM performance (geomean 20M_​cell_​motorbike, 42M_​cell_​motorbike) App Version: v8; Build notes: Tools: Intel FORTRAN Compiler 2020u4, Intel C Compiler 2020u4, Intel MPI 2019u8; threads/core: 1; Turbo: used; Build knobs: -O3 -ip -xCORE-AVX512

OpenFOAM Disclaimer: This offering is not approved or endorsed by OpenCFD Limited, producer and distributor of the OpenFOAM software via www.openfoam.com, and owner of the OPENFOAM® and OpenCFD® trademark

1.64x higher GROMACS performance (geomean ion_​channel_​pme, lignocellulose_​rf, water_​pme, water_​rf) App Version: v2020.5_​SP; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel MPI 2019u8; threads/core: 2; Turbo: used; Build knobs: -O3 -ip -xCORE-AVX512

1.60x higher LAMMPS performance (geomean Polyethylene, Stillinger-Weber, Tersoff, Water) App Version: v2020-10-29; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4, Intel MPI 2019u8; threads/core: 2; Turbo: used; Build knobs: -O3 -ip -xCORE-AVX512 -qopt-zmm-usage=high

1.57x higher NAMD performance (geomean Apoa1, f1atpase, STMV) App Version: 2.15-Alpha1 (includes AVX tiles algorithm); Build notes: Tools: Intel MKL, Intel C Compiler 2020u4, Intel MPI 2019u8, Intel Threading Building Blocks 2020u4; threads/core: 2; Turbo: used; Build knobs: -ip -fp-model fast=2 -no-prec-div -qoverride-limits -qopenmp-simd -O3 -xCORE-AVX512 -qopt-zmm-usage=high

1.61x higher RELION Plasmodium Ribosome performance App Version: 3_​1_​1; Build notes: Tools: Intel C Compiler 2020u4, Intel MPI 2019u9; threads/core: 2; Turbo: used; Build knobs: -O3 -ip -g -debug inline-debug-info -xCOMMON-AVX512 -qopt-report=5 –restrict

Baseline: February 2021

1.54x higher NAMD STMV performance using the AVX Tile Algorithm on Platinum 8380 vs. without AVX Tiles

2.43x higher NAMD STMV performance using the AVX Tiles Alogrithm on Platinum 8380 vs. prior gen without AVX Tiles

1.57x higher NAMD STMV performance on Platinum 8380 vs. prior gen without AVX Tiles

New: 8380: 1-node, 2x Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, Tested by Intel between March 12, 2021 and March 29, 2021.

Baseline: 8280: 1-node, 2x Intel Xeon Platinum 8280 (28C/2.7GHz, 205W TDP) processor on Intel Software Development Platform with 192GB (12 slots/ 16GB/ 2933) total DDR4 memory, ucode 0x4002f01, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG48 .Tested by Intel between February 1, 2021 to February 20, 2021

1.54x higher performance on NAMD STMV from using the AVX Tiles Algorithm vs. without AVX Tiles

2.43x higher performance on NAMD STMV with AVX Tiles Algorithm vs. prior gen without AVX Tile

1.57x higher performance on NAMD STMV without AVX Tiles Algorithm vs. prior gen

NAMD with Tiles: App Version: 2.15-Alpha1 (includes AVX tiles algorithm); Build notes: Tools: Intel MKL, Intel C Compiler 2020u4, Intel MPI 2019u8, Intel Threading Building Blocks 2020u4; threads/core: 2; Turbo: used; Build knobs: -ip -fp-model fast=2 -no-prec-div -qoverride-limits -qopenmp-simd -O3 -xCORE-AVX512 -qopt-zmm-usage=high

NAMD without Tiles: App Version: 2.15-Alpha1 (built without AVX tiles algorithm); Build notes: Tools: Intel MKL, Intel C Compiler 2020u4, Intel MPI 2019u8, Intel Threading Building Blocks 2020u4; threads/core: 2; Turbo: used; Build knobs: -ip -fp-model fast=2 -no-prec-div -qoverride-limits -qopenmp-simd -O3 -xCORE-AVX512 -qopt-zmm-usage=high -DNAMD_​KNL

tested by Intel and results as of March 2021

Baseline: February 2021

1.42x higher performance on OpenFOAM Motorbike 42M

New: 6354: 1-node, 2x Intel Xeon Gold 6354 (18C/3.0GHz, 205W TDP) processor on Intel Software Development Platform with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3, 4.18.0-240.10.1.el8_​3.x86_​64, 1x Intel_​SSDSC2KG96 .Tested by Intel between March 12, 2021 and March 29, 2021.

Baseline: 6154: 1-node, 2x Intel Xeon Gold 6154 (18C/3.0GHz, 200W TDP) processor on Intel Software Development Platform with 192GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x2006a0a, HT on, Turbo on, CentOS Linux 8.3, 4.18.0-240.10.1.el8_​3.x86_​64, 1x Intel_​SSDSC2KG96 . Tested by Intel between February 1, 2021 to February 20, 2021

App Version: v8; Build notes: Tools: Intel FORTRAN Compiler 2020u4, Intel C Compiler 2020u4, Intel MPI 2019u8; threads/core: 1; Turbo: used; Build knobs: -O3 -ip -xCORE-AVX512

tested by Intel and results as of March 2021

OpenFOAM Disclaimer: This offering is not approved or endorsed by OpenCFD Limited, producer and distributor of the OpenFOAM

Baseline: February 2021

1.52x higher manufacturing performance (geomean Altair RADIOSS, Ansys Fluent, Ansys LS-DYNA, Converge, Numeca, OpenFOAM)

New: 8380: 1-node, 2x Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96 .Tested by Intel between March 12, 2021 and March 29, 2021.

Baseline: 8280: 1-node, 2x Intel Xeon Platinum 8280 (28C/2.7GHz, 205W TDP) processor on Intel Software Development Platform with 192GB (12 slots/ 16GB/ 2933) total DDR4 memory, ucode 0x4002f01, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG48. Tested by Intel between February 1, 2021 to February 20, 2021.

1.47x higher Altair RADIOSS performance (geomean Neon1M/80ms, T10M/8ms) App Version: 2020; Build notes: Tools: Intel FORTRAN Compiler 2021u1, Intel C Compiler 2021u1, Intel MPI 2021u1; threads/core: 1; Turbo: used;

1.54x higher Ansys Fluent performance (geomean aircraft_​wing_​14m. aircraft_​wing_​2m, combustor_​12m, combustor_​16m, combustor_​71m, exhaust_​system_​33m, fluidized_​bed_​2m, ice_​2m, landing_​gear_​15m, oil_​rig_​7m, pump_​2m, rotor_​3m, sedan_​4m) App Version: 2021 R1; Build notes: One thread per core; Multi-threading Enabled; Turbo Boost Enabled; Intel FORTRAN Compiler 19.5.0; Intel C/C++ Compiler 19.5.0; Intel Math Kernel Library 2020.0.0; Intel MPI Library 2019 Update 8

1.48x higher Ansys LS-DYNA performance (geomean 3cars-150ms, car2car-120ms, ODB_​10M-30ms) App Version: R11; Build notes: Tools: Intel Compiler 2019u5 (AVX512), Intel MPI 2019u9; threads/core: 1; Turbo: used

1.52x higher Converge SI8_​engine_​PFI_​SAGE_​transient_​RAN performance App Version: 3.0.17; Build notes: Tools: Intel MPI 2019u9; threads/core: 1; Turbo: used; 3.0.17 Converge official converge-intelmp binary

1.61x higher Numeca performance (geomean FO_​hpcc_​single_​passage, FT_​hpcc_​single_​passage) FineOpen App Version: v10.1; Build notes: Tools: Customer pre-built binaries (Intel Fortran Compiler 2019, Intel C Compiler 2015), Intel MPI 2019u9; threads/core:1 ; Turbo: used; Build knobs: Fortran = -O2 -fp-model precise, C = -O2 -fPIC -pipe -Wno-deprecated -Wreturn-type -fp-model precise -std=c++11 FineTurbo App Version: v15.1; Build notes: Tools: Customer pre-built binaries (Intel Fortran Compiler 2015, Intel C Compiler 2015), Intel MPI 2019u4; threads/core:1 ; Turbo: used; Build knobs: Fortran = -O2 -fp-model precise, C = -O2 -fPIC -pipe -Wno-deprecated -Wreturn-type -fp-model precise -std=c++11

1.51x higher OpenFOAM performance (geomean 20M_​cell_​motorbike, 42M_​cell_​motorbike) App Version: v8; Build notes: Tools: Intel FORTRAN Compiler 2020u4, Intel C Compiler 2020u4, Intel MPI 2019u8; threads/core: 1; Turbo: used; Build knobs: -O3 -ip -xCORE-AVX512

OpenFOAM Disclaimer: This offering is not approved or endorsed by OpenCFD Limited, producer and distributor of the OpenFOAM

tested by Intel and results as of March 2021

Baseline: February 2021

HPCG: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: 2019u5 MKL; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel MPI 2019u8; threads/core: 1; Turbo: used; Build knobs: -O3 -ip -xCORE-AVX512. EPYC 7543: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: 2019u5 MKL; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel MPI 2019u8; threads/core: 1; Turbo: used; Build knobs: -O3 -ip -march=core-avx2, tested by Intel and results as of April 2021

HPL: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: The Intel Distribution for LINPACK Benchmark; Build notes: Tools: Intel MPI 2019u7; threads/core: 1; Turbo: used; Build: build script from Intel Distribution for LINPACK package; 1 rank per NUMA node: 1 rank per socket, EPYC 7543: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: AMD official HPL 2.3 MT version with BLIS 2.1; Build notes: Tools: hpc-x 2.7.0; threads/core: 1; Turbo: used; Build: pre-built binary (gcc built) from https://developer.amd.com/amd-aocl/blas-library/; 1 rank per L3 cache, 4 threads per rank, tested by Intel and results as of April 2021

STREAM Triad: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: McCalpin_​STREAM_​OMP-version; Build notes: Tools: Intel C Compiler 2019u5; threads/core: 1; Turbo: used; BIOS settings: HT=on Turbo=On SNC=On. EPYC 7543: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: McCalpin_​STREAM_​OMP-version; Build notes: Tools: Intel C Compiler 2019u5; threads/core: 1; Turbo: used; BIOS settings: HT=on Turbo=On SNC=On, tested by Intel and results as of April 2021

WRF Geomean of Conus-12km, Conus-2.5km, NWSC-3 NA-3km: Platinum 8358: 1-node 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: 4.2.2; Build notes: Intel Fortran Compiler 2020u4, Intel MPI 2020u4; threads/core: 1; Turbo: used; Build knobs:-ip -w -O3 -xCORE-AVX2 -vec-threshold0 -ftz -align array64byte -qno-opt-dynamic-align -fno-alias $(FORMAT_​FREE) $(BYTESWAPIO) -fp-model fast=2 -fimf-use-svml=true -inline-max-size=12000 -inline-max-total-size=30000. EPYC 7543: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: 4.2.2; Build notes: Intel Fortran Compiler 2020u4, Intel MPI 2020u4; threads/core: 1; Turbo: used; Build knobs: -ip -w -O3 -march=core-avx2 -ftz -align all -fno-alias $(FORMAT_​FREE) $(BYTESWAPIO) -fp-model fast=2 -inline-max-size=12000 -inline-max-total-size=30000, tested by Intel and results as of April 2021

Binomial Options: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: v1.0; Build notes: Tools: Intel C Compiler 2020u4, Intel Threading Building Blocks ; threads/core: 2; Turbo: used; Build knobs: -O3 -xCORE-AVX512 -qopt-zmm-usage=high -fimf-domain-exclusion=31 -fimf-accuracy-bits=11 -no-prec-div -no-prec-sqrt. EPYC 7543: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: v1.0; Build notes: Tools: Intel C Compiler 2020u4, Intel Threading Building Blocks ; threads/core: 2; Turbo: used; Build knobs: -O3 -march=core-avx2 -fimf-domain-exclusion=31 -fimf-accuracy-bits=11 -no-prec-div -no-prec-sqrt, tested by Intel and results as of April 2021

Monte Carlo: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: v1.1; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4; threads/core: 1; Turbo: used; Build knobs: -O3 -xCORE-AVX512 -qopt-zmm-usage=high -fimf-precision=low -fimf-domain-exclusion=31 -no-prec-div -no-prec-sqrt. EPYC 7543: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: v1.1; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4; threads/core: 2; Turbo: used; Build knobs: -O3 -march=core-avx2 -fimf-precision=low -fimf-domain-exclusion=31 -no-prec-div -no-prec-sqrt, tested by Intel and results as of April 2021

Ansys Fluent Geomean of aircraft_​wing_​14m, aircraft_​wing_​2m, combustor_​12m, combustor_​16m, combustor_​71m, exhaust_​system_​33m, fluidized_​bed_​2m, ice_​2m, landing_​gear_​15m, oil_​rig_​7m, pump_​2m, rotor_​3m, sedan_​4m: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: 2021 R1; Build notes: One thread per core; Multi-threading Enabled; Turbo Boost Enabled; Intel FORTRAN Compiler 19.5.0; Intel C/C++ Compiler 19.5.0; Intel Math Kernel Library 2020.0.0; Intel MPI Library 2019 Update 8. EPYC 7543: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: 2021 R1; Build notes: One thread per core; Multi-threading Enabled; Turbo Boost Enabled; Intel FORTRAN Compiler 19.5.0; Intel C/C++ Compiler 19.5.0; Intel Math Kernel Library 2020.0.0; Intel MPI Library 2019 Update 8, tested by Intel and results as of April 2021

Ansys LS-DYNA Geomean of car2car-120ms, ODB_​10M-30ms: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: R11; Build notes: Tools: Intel Compiler 2019u5 (AVX512), Intel MPI 2019u9; threads/core: 1; Turbo: used. EPYC 7543: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: R11; Build notes: Tools: Intel Compiler 2019u5 (AMDAVX2), Intel MPI 2019u9; threads/core: 1; Turbo: used, tested by Intel and results as of April 2021

OpenFOAM 42M_​cell_​motorbike: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: v8; Build notes: Tools: Intel FORTRAN Compiler 2020u4, Intel C Compiler 2020u4, Intel MPI 2019u8; threads/core: 1; Turbo: used; Build knobs: -O3 -ip -xCORE-AVX512. EPYC 7543: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: v8; Build notes: Tools: Intel FORTRAN Compiler 2020u4, Intel C Compiler 2020u4, Intel MPI 2019u8; threads/core: 1; Turbo: used; Build knobs: -O3 -ip -march=core-avx2, tested by Intel and results as of April 2021

LAMMPS Geomean of Polyethylene, Stillinger-Weber, Tersoff, Water: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: v2020-10-29; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4, Intel MPI 2019u8; threads/core: 2; Turbo: used; Build knobs: -O3 -ip -xCORE-AVX512 -qopt-zmm-usage=high. EPYC 7543: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: v2020-10-29; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4, Intel MPI 2019u8; threads/core: 2; Turbo: used; Build knobs: -O3 -ip -march=core-avx2, tested by Intel and results as of April 2021

NAMD Geomean of Apoa1, STMV: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: 2.15-Alpha1 (includes AVX tiles algorithm); Build notes: Tools: Intel MKL , Intel C Compiler 2020u4, Intel MPI 2019u8, Intel Threading Building Blocks 2020u4; threads/core: 2; Turbo: used; Build knobs: -ip -fp-model fast=2 -no-prec-div -qoverride-limits -qopenmp-simd -O3 -xCORE-AVX512 -qopt-zmm-usage=high. EPYC 7543: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: 2.15-Alpha1 (includes AVX tiles algorithm); Build notes: Tools: Intel MKL , AOCC 2.2.0, gcc 9.3.0, Intel MPI 2019u8; threads/core: 2; Turbo: used; Build knobs: -O3 -fomit-frame-pointer -march=znver1 -ffast-math, tested by Intel and results as of April 2021

RELION Plasmodium Ribosome: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: 3_​1_​1; Build notes: Tools: Intel C Compiler 2020u4, Intel MPI 2019u9; threads/core: 2; Turbo: used; Build knobs: -O3 -ip -g -debug inline-debug-info -xCOMMON-AVX512 -qopt-report=5 –restrict. EPYC 7543: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: 3_​1_​1; Build notes: Tools: Intel C Compiler 2020u4, Intel MPI 2019u9; threads/core: 2; Turbo: used; Build knobs: -O3 -ip -g -debug inline-debug-info -march=core-avx2 -qopt-report=5 -restrict, tested by Intel and results as of April 2021

Baseline: April 2021

New: m6i.xlarge 16 GB memory capacity/instance, m6i.2xlarge 32 GB memory capacity/instance, m6i.4xlarge 64 GB memory capacity/instance, m6i.8xlarge 128 GB memory capacity/instance, m6i.16xlarge 256 memory capacity/instance (Xeon).

Baseline: m6g.xlarge 16 GB memory capacity/instance, m6g.2xlarge 32 GB memory capacity/instance, m6g.4xlarge 64 GB memory capacity/instance, m6g.8xlarge 128 GB memory capacity/instance, m6g.16xlarge 256 GB memory capacity/instance (Graviton2).

Baseline: October 2021

New: m6i.xlarge 16 GB memory capacity/instance, m6i.2xlarge 32 GB memory capacity/instance, m6i.4xlarge 64 GB memory capacity/instance, m6i.8xlarge 128 GB memory capacity/instance (Xeon).

Baseline: m6g.xlarge 16 GB memory capacity/instance, m6g.2xlarge 32 GB memory capacity/instance, m6g.4xlarge 64 GB memory capacity/instance, m6g.8xlarge 128 GB memory capacity/instance (Graviton2).

Baseline: November 2021

New: m6i.xlarge 16 GB memory capacity/instance, m6i.2xlarge 32 GB memory capacity/instance, m6i.4xlarge 64 GB memory capacity/instance, m6i.8xlarge 128 GB memory capacity/instance (Xeon).

Baseline: m6g.xlarge 16 GB memory capacity/instance, m6g.2xlarge 32 GB memory capacity/instance, m6g.4xlarge 64 GB memory capacity/instance, m6g.8xlarge 128 GB memory capacity/instance (Graviton2).

Baseline: November 2021

Baseline: Feb 04, 2021

1.6x higher throughput under SLA and 1.4x higher throughput for Server Side Java: New: Platinum 8380: 1-node, 2x Intel Xeon Platinum 8380 processor on Coyote Pass with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, Ubuntu 20.04.1 LTS, 5.4.0-64-generic, 1x SSDSC2BA40, Java workload, JDK 1.15.0.1, test by Intel on 3/15/2021. Baseline: Platinum 8280: 1-node, 2x Intel Xeon Platinum 8280 processor on Wolf Pass with 384 GB (12 slots/ 32GB/ 2933) total DDR4 memory, ucode 0x5003003, HT on, Turbo on, Ubuntu 20.04.1 LTS, 5.4.0-64-generic, 1x INTEL_​SSDSC2KG01, Java workload, JDK 1.15.0.1, test by Intel on 2/18/2021.

1.48x higher responses on WordPress with HTTPS: New: Platinum 8380: 1-node, 2x Intel Xeon Platinum 8380 processor on Coyote Pass with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, Ubuntu 20.04, 5.4.0-65-generic, 1x Intel 895GB SSDSC2KG96, 1x XL710-Q2, WordPress 4.2 with HTTPS, gcc 9.3.0, GLIBC 2.31-0ubuntu9.1, mysqld Ver 10.3.25-MariaDB-0ubuntu0.20.04.1, PHP 7.4.9-dev (fpm-fcgi), Zend Engine v3.4.0, test by Intel on 3/15/2021. Baseline: Platinum 8280: 1-node, 2x Intel Xeon Platinum 8280 processor on Wolf Pass with 384 GB (12 slots/ 32GB/ 2933) total DDR4 memory, ucode 0x5003003, HT on, Turbo on, Ubuntu 20.04, 5.4.0-65-generic, 1x Intel 1.8T SSDSC2KG01, 1x Intel X722, test by Intel on 2/5/2021.

Baseline: Feb 05, 2021

Baseline: Feb 04, 2021

Baseline: Feb 05, 2021

Up to 1.4x higher Server Side Java throughput performance with 3rd Gen Intel® Xeon® Platinum 8380 processor vs. prior generation platform

Baseline: Feb 18, 2021

Up to 21% more PostgreSQL database transactions vs. AMD EPYC.

New: 1-node, Intel Software Development Platform, 2x Intel® Xeon Platinum 8358 (32C, 2.6GHz, 250W TDP), HT On, Turbo ON, Total Memory: 512 GB (16 slots/ 32GB/ 3200 MHz), ucode: x260, 1x SSD boot, 1xP5800x 1.6TB, Red Hat 8.3, kernel: 4.18.0-240.el8.x86_​64, HammerDB v4.1, PostgreSQL 13.0. Tested by Intel as of April 2021.

AMD EPYC: 1-node, Dell PowerEdge R7525, 2x AMD 7543 (32C, 2.8GHz, 240W cTDP), SMT On, Boost ON, Total Memory: 512 GB (16 slots/ 32GB/ 3200 MHz), ucode: 0xa001119, 1x NVMe boot, 1x PM1733 3.84TB, Red Hat 8.3, kernel: 4.18.0-240.el8.x86_​64, HammerDB v4.1, PostgreSQL 13.0. Tested by Intel as of April 2021.

Baseline: April 2021

Up to 12% more MySQL transactions and 34% lower average new transaction database latency vs. AMD EPYC.

New: 1-node, Intel Software Development Platform, 2x Intel® Xeon Platinum 8358 (32C, 2.6GHz, 250W TDP), HT On, Turbo ON, Total Memory: 512 GB (16 slots/ 32GB/ 3200 MHz), ucode: x260, 1x SSD boot, 1xP5800x 1.6TB, Red Hat 8.3, kernel: 4.18.0-240.el8.x86_​64, HammerDB v4, MySQL 8.0.22. Average P95 new order (NEWORD) latency as measured by HammerDB. Tested by Intel as of April 2021.

AMD /EPYC: 1-node, Dell PowerEdge R7525, 2x AMD 7543 (32C, 2.8GHz, 240W cTDP), SMT On, Boost ON, Total Memory: 512 GB (16 slots/ 32GB/ 3200 MHz), ucode: 0xa001119, 1x NVMe boot, 1x PM1733 3.84TB, Red Hat 8.3, kernel: 4.18.0-240.el8.x86_​64, HammerDB v4, MySQL 8.0.22. Average P95 new order (NEWORD) latency as measured by HammerDB. Tested by Intel as of April 2021.

Baseline: April 2021

With Intel® 3rd Gen Xeon® Scalable processors, CoSP's can achieve up to 21% boost in vBNG performance, while enabling increased flexibility for fixed and mobile convergence, manageability and scalability to expand use cases to address both mobile and broadband workloads.

With Intel® 3rd Gen Xeon® Scalable processors, CoSP's can increase 5G UPF performance by 42%. Combined with Intel Ethernet 800 series adapters, they can deliver the performance, efficiency and trust for use cases that require low latency, including augmented reality, cloud-based gaming, discrete automation and even robotic-aided surgery.

With Intel® 3rd Gen Xeon® Scalable processors and the latest Intel® Optane™ Persistent Memory you can get up to 63% higher throughput and 33% more memory capacity, enabling you to serve the same number of subscribers at higher resolution or a greater number of subscribers at the same resolution.

With Intel® 3rd Gen Xeon® Scalable processors, you can support up to 94% more secure networking connections and achieve significantly faster speeds to support cloud, edge and work-from-home use cases.

With the higher core performance and new crypto acceleration of Intel® 3rd Gen Xeon® Scalable processors, CoSP's can achieve 72% better CMTS platform performance. Additional QAT offload can add up to another 10% boost.

With Intel® 3rd Gen Xeon® Scalable processors, enhance Vector Packet Processing - Forward Information Base performance by 66% vs. the prior generation.

With Intel® 3rd Gen Xeon® Scalable processors, enhance DPDK L3 Forwarding performance by 88% vs. the prior generation.

With Intel® 3rd Gen Intel® Xeon® Scalable processors, Ethernet 800 series and vRAN dedicated accelerators, CoSP's can get to 2x Massive MIMO throughput in a similar power envelope for a best-in-class 3x100mhz 64T64R configuration.

1.62x average network performance gains: geomean of Virtual Broadband Network Gateway, 5G User Plane Function, Virtual Cable Modem Termination System, Vector Packet Processing - Forward Information Base 512B, DPDK L3 Forward 512B, CDN-Live, Vector Packet Processing - IP Security 1420B.

1.2x  Virtual Broadband Network Gateway: New: Gold 6338N: 1-node, 2(1 socket used)x Intel Xeon Gold 6338N on Intel* Whitley with 256 GB (16 slots/ 16GB/ 2666) total DDR4 memory, ucode 0x261, HT on, Turbo off, Ubuntu 20.04 LTS (Focal Fossa)​, 5.4.0-40-generic, 1x INTEL* 240G SSD , 3x E810-CQDA2 (Tacoma Rapids), vBNG 20.07, Gcc 9.3.0​, test by Intel on 3/11/2021. Baseline: Gold 6252N: 1-node, 2(1 socket used)x Intel Xeon Gold 6252N on SuperMicro* X11DPG-QT with 192 GB (12 slots/ 16GB/ 2933)  total DDR4 memory, ucode 0x5002f01, HT on, Turbo off, Ubuntu 20.04 LTS (Focal Fossa)​, 5.4.0-40-generic, 1x INTEL* 240G SSD , 3x E810-CQDA2 (Tacoma Rapids), vBNG 20.07, Gcc 9.3.0​,  test by Intel on 2/2/2021.

1.42x 5G User Plane Function: New: 1-node, 2(1 socket used)x Intel Xeon Gold 6338N on Whitley Coyote Pass 2U  with 128 GB (8 slots/ 16GB/ 2666)  total DDR4 memory, ucode 0x261, HT on, Turbo off, Ubuntu 18.04.5 LTS, 4.15.0-134-generic, 1x Intel 810 (Columbiaville), FlexCore 5G UPF, Jan’ 2021​ MD5 checksum: c4ad7f8422298ceb69d01e67419ff1c1, GCC 7.5.0, 5G UPF228 Gbps / 294 Gbps,  test by Intel on 3/16/2021. Baseline: 1-node, 2(1 socket used)x Intel Xeon Gold 6252N on SuperMicro* X11DPG-QT with 96 GB (6 slots/ 16GB/ 2934)  total DDR4 memory, ucode 0x5003003, HT on, Turbo off, Ubuntu 18.04.5 LTS, 4.15.0-132-generic, 1x Intel 810 (Columbiaville), FlexCore 5G UPF, Jan’ 2021  MD5 checksum: c4ad7f8422298ceb69d01e67419ff1c1, GCC 7.5.0, 5G UPF161 Gbps / 213 Gbps,  test by Intel on 2/12/2021.

1.63x CDN Live: New: 1 node, 2x Intel® Xeon® Gold 6338N Processor, 32 core HT ON Turbo ON, Total DRAM 256GB (16 slots/16GB/2666MT/s), Total Optane Persistent Memory 200 Series 2048GB (16 slots/128GB/2666MT/s), BIOS SE5C6200.86B.2021.D40.2103100308 (ucode: 0x261), 4x Intel® E810, Ubuntu 20.04, kernel 5.4.0-65-generic, gcc 9.3.0 compiler, openssl 1.1.1h, varnish-plus 6.0.7r2. 2 clients, Test by Intel as of 3/11/2021. Baseline: Gold 6252N: 2x Intel® Xeon® Gold 6252N Processor, 24 core HT ON Turbo ON, Total DRAM 192GB (12 slots/16GB/2666MT/s), Total Optane Persistent Memory 100 Series 1536GB(12 slots/128GB/2666MT/s), 1x Mellanox MCX516A-CCAT, BIOS: SE5C620.86B.02.01.0013.121520200651 (ucode: 0x5003003), Ubuntu 20.04, kernel 5.4.0-65-generic, wrk master 4/17/2019. Test by Intel as of 2/15/2021. Throughput measured with 100% Transport Layer Security (TLS) traffic with 93.3% target cache hit ratio and keep alive on, 512 total connections.

1.94x Vector Packet Processing - IP Security 1420B: New: 1-node, 2(1 socket used)x Intel Xeon Gold 6338N on Intel* Whitley with 128 GB (8 slots/ 16GB/ 2666)  total DDR4 memory, ucode 0x261, HT on, Turbo off, Ubuntu 20.04 LTS (Focal Fossa)​, 5.4.0-40-generic, 1x INTEL* 240G SSD , 1x E810-2CQDA2 (Chapman Beach), v21.01-release, Gcc 9.3.0​, VPPIPSEC(24c24t) test by Intel on 3/17/2021 .Baseline: 1-node, 2(1 socket used)x Intel Xeon Gold 6252N on SuperMicro* X11DPG-QT with 96 GB (6 slots/ 16GB/ 2933)  total DDR4 memory, ucode 0x5002f01, HT off, Turbo off, Ubuntu 20.04 LTS (Focal Fossa)​, 5.4.0-40-generic, 1x INTEL* 240G SSD , 1x E810-CQDA2 (Tacoma Rapids), v21.01-release, Gcc 9.3.0​,  VPPIPSEC(18c18t) test by Intel on 2/2/2021.

1.72x Virtual Cable Modem Termination System: New: Gold 6338N: 1-node, 2(1 socket used)x Intel Xeon Gold 6338N on Coyote Pass with 256 GB (16 slots/ 16GB/ 2666) total DDR4 memory, ucode 0x261, HT on, Turbo off(no SST-BF)/on(SST-BF), Ubuntu 20.04 LTS (Focal Fossa)​, 5.4.0-40-generic, 1x INTEL* 240G SSD , 3x E810-CQDA2 (Tacoma Rapids), vCMTS 20.10​, Gcc 9.3.0​, SST-BF (2.4 Ghz,1.9 Ghz frequencies for the priority cores and the other cores respectively ), test by Intel on 3/11/2021. Baseline: Gold 6252N: 1-node, 2(1 socket used)x Intel Xeon Gold 6252N on SuperMicro* X11DPG-QT with 192 GB (12 slots/ 16GB/ 2933)  total DDR4 memory, ucode 0x5002f01, HT on, Turbo off, Ubuntu 20.04 LTS (Focal Fossa)​, 5.4.0-40-generic, 1x INTEL* 240G SSD , 2x E810-CQDA2 (Tacoma Rapids), vCMTS 20.10​, Gcc 9.3.0​, vCMTS90 (14 instances),  test by Intel on 2/2/2021.

1.66 Vector Packet Processing - Forward Information Base 512B: New: 1-node, 2(1 socket used)x Intel Xeon Gold 6338N on Intel* Whitley with 128 GB (8 slots/ 16GB/ 2666)  total DDR4 memory, ucode 0x261, HT on, Turbo off, Ubuntu 20.04 LTS (Focal Fossa)​, 5.4.0-40-generic, 1x INTEL* 240G SSD , 1x E810-2CQDA2 (Chapman Beach), v20.05.1-release, Gcc 9.3.0​, VPPFIB(24c24t)​,  test by Intel on 3/17/2021. Baseline: 1-node, 2(1 socket used)x Intel Xeon Gold 6252N on SuperMicro* X11DPG-QT with 96 GB (6 slots/ 16GB/ 2933)  total DDR4 memory, ucode 0x5002f01, HT off, Turbo off, Ubuntu 20.04 LTS (Focal Fossa)​, 5.4.0-40-generic, 1x INTEL* 240G SSD , 1x E810-CQDA2 (Tacoma Rapids), v20.05.1-release, Gcc 9.3.0​, VPPFIB (18c18t)​,  test by Intel on 2/2/2021.

1.88x DPDK L3 Forward 512B: New: 1-node, 2(1 socket used)x Intel Xeon Gold 6338N on Intel* Whitley with 128 GB (8 slots/ 16GB/ 2666)  total DDR4 memory, ucode 0x261, HT on, Turbo off, Ubuntu 20.04 LTS (Focal Fossa)​, 5.4.0-40-generic, 1x INTEL* 240G SSD , 1x E810-2CQDA2 (Chapman Beach), v20.08.0, Gcc 9.3.0​, DPDKL3FWD (24c24t), test by Intel on 3/17/2021, Baseline: 2(1 socket used)x Intel Xeon Gold 6252N on SuperMicro* X11DPG-QT with 96 GB (6 slots/ 16GB/ 2933)  total DDR4 memory, ucode 0x5002f01, HT off, Turbo off, Ubuntu 20.04 LTS (Focal Fossa)​, 5.4.0-40-generic, 1x INTEL* 240G SSD , 1x E810-CQDA2 (Tacoma Rapids), v20.08.0, Gcc 9.3.0​​, DPDKL3FWD (12c12t),  test by Intel on 2/2/2021.

FlexRAN: 2x MIMO Throughput: Results have been estimated or simulated. Based on 2x estimated throughput from 32Tx32R (5Gbps) on 2nd Gen Intel® Xeon® Gold 6212U processor to 64Tx64R (10Gbps) on 3rd Gen Intel Xeon Gold 6338N processor at similar power ~185W.

FlexRAN

Baseline: Feb 02, 2021

[91] Up to 1.74x higher performance with the new 3rd Gen Intel Xeon Scalable Platform supporting Intel® Optane™ PMem 200 Series for CDN Live use case vs. prior generation Intel Xeon® Scalable platform supporting Intel® Total Optane Persistent Memory 100 Series

Baseline: May 10, 2021

Up to 383.33Gbps network throughput with 2 socket 3rd Gen Intel Xeon Scalable platform featuring gen 4 Intel DC P5510 SSD for Video-On-Demand CDN use case using Varnish

191.58Gbps CDN Video On Demand with Intel NVMe: 1 node, 1x Intel® Xeon® Gold 6338 Processor, 32 core HT ON Turbo ON, Total Memory 256GB (8 slots/32GB/3200MT/s), BIOS 1.1 (ucode: 0xd000280), 8x Intel® P5510, 1x Intel® E810, Ubuntu 20.04, kernel 5.4.0-65-generic, gcc 9.3.0 compiler, openssl 1.1.1h, varnish-plus-6.0.8r1 revision 96a565db42d1ba89c21a3caa1b06a42d296581f3. 2 clients, wrk master 4/17/2019 (keep alive on, 3000 total connections). Test by Intel as of 5/10/2021. Throughput measured with 100% Transport Layer Security (TLS) traffic with 100% target cache hit ratio.

383.33Gbps CDN Video On Demand with Intel NVMe: 1 node, 2x Intel® Xeon® Gold 6338 Processor, 32 core HT ON Turbo ON, Total Memory 256GB (16 slots/16GB/3200MT/s), BIOS 1.1 (ucode: 0xd000280), 12x Intel® P5510, 4x Intel® E810, Ubuntu 20.04, kernel 5.4.0-65-generic, gcc 9.3.0 compiler, openssl 1.1.1h, varnish-plus-6.0.8r1 revision 9ec6a3b587aa2500a6f7dd3cc7510d617ebb0c4e. 4 clients, wrk master 4/17/2019 (keep alive on, 3000 total connections). Test by Intel as of 5/10/2021. Throughput measured with 100% Transport Layer Security (TLS) traffic with 100% target cache hit ratio.

Baseline: May 10, 2021

Baseline: January 17, 2021

Up to 2.4x better performance on NGINX Web Secure Key Exchange vs. AMD EPYC.

New: 1-node, 1x Intel® Xeon® Gold 6338N (32C/2.2GHz) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x260, HT on, Turbo off, 1xSSD, 3 x QUAD E810-XXVDA4 (12 x 25 GbE), Ubuntu* 20.04, 5.4.0.40-generic, Glib 2.31, Async NGINX* v0.4.3, QAT Engine v0.6.4, Crypto MB ippcp_​2020u3, Intel® Multi-Buffer Crypto for IPsec v0.54, GCC 9.3.0, OpenSSL* 1.1.1h, tested by Intel as of March 2021.

AMD EPYC: 1-node,1-socket AMD EPYC 7513 (32C/2.6GHz) on Supermicro H12SSL-CT server with 128 GB (8 slots/ 16GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost off, 1xSSD, 6 x 100GbE Controller E810-C for QSFP, Ubuntu* 20.04, 5.4.0.40-generic, Glib 2.31, nginx/1.18.0 async v0.4.5, GCC 9.3.0, OpenSSL 1.1.1j, tested by Intel as of July 2021.

Baseline: July 2021

4.96x more sentences per second in BERT-Large with INT8 precision for 32vCPU and 3.07x for 64vCPU, Workload BERT, Other SW (Graviton with FP32 precision): Ubuntu 20.04.2 LTS, Kernel 5.4.0-1060-aws, GCC 9.3.0, library TensorFlow 2.5.0, Docker 20.10.7, containerd 1.3.7, other SW (Intel with INT8 precision): Ubuntu 20.04 LTS, Kernel 5.11.0-1022-aws, GCC 8.4.0, libraries Python 3.6.9, TensorFlow 2.5.0, Docker 20.10.7, containerd 1.5.5.

New: m6i.8xlarge 128 GB memory capacity/instance, m6i.16xlarge 256 GB memory capacity/instance (Xeon).

Baseline: m6g.8xlarge 128 GB memory capacity/instance, m6g.16xlarge 256 GB memory capacity/instance (Graviton2).

Test by Intel. New: December 1, 2021; Baseline: November 10, 2021

BERT-Large inference - Natural language processing speedup (sentences per second)

Baseline: November 10, 2021

2.04x more sentences per second in BERT-Large with FP32 precision for 32vCPU and 1.38x for 64vCPU, Workload BERT, Other SW (Graviton): Ubuntu 20.04.2 LTS, Kernel 5.4.0-1060-aws, GCC 9.3.0, library TensorFlow 2.5.0, Docker 20.10.7, containerd 1.3.7, other SW (Intel): Ubuntu 20.04 LTS, Kernel 5.11.0-1022-aws, GCC 8.4.0, libraries Python 3.6.9, TensorFlow 2.5.0, Docker 20.10.7, containerd 1.5.5.

New: m6i.8xlarge 128 GB memory capacity/instance, m6i.16xlarge 256 GB memory capacity/instance (Xeon).

Baseline: m6g.8xlarge 128 GB memory capacity/instance, m6g.16xlarge 256 GB memory capacity/instance (Graviton2).

Test by Intel. New: December 1, 2021; Baseline: November 10, 2021

BERT-Large inference - Natural language processing speedup (sentences per second)

Baseline: November 10, 2021

Virtualization workload

Baseline: Dec 22, 2020

Baseline: Nov 30, 2020

Baseline: Feb 06, 2021

Baseline: Feb 05, 2021

Up to 1.43x higher transactions on the new 3rd Gen Intel Xeon Scalable processor with Intel Optane PMem 200, Intel P5510 SSD and Intel Ethernet E810 Network Adaptor running Aerospike with index in PMem and data in SSD vs. prior generation platform

Baseline: March 16, 2021

Baseline: Jan 31, 2021

New: m6i.2xlarge 32 GB memory capacity/instance, m6i.4xlarge 64 GB memory capacity/instance, m6i.16xlarge 256 memory capacity/instance (Xeon).

Baseline: m6g.2xlarge 32 GB memory capacity/instance, m6g.4xlarge 64 GB memory capacity/instance, m6g.16xlarge 256 GB memory capacity/instance (Graviton2).

Baseline: October 2021

New: M6i.4xlarge 64 GB memory capacity/instance (Xeon).

Baseline: M6g.4xlarge 64 GB memory capacity/instance (Graviton2).

Baseline: November 2021

Baseline: March 08, 2021

Baseline: March 05, 2021

Baseline: February 09, 2021

Up to 3x higher 4KB Rand Read/Write 70/30 IOPS performance with 3rd Gen Intel Xeon® scalable platform supporting PCIe Gen4 Intel Optane™ SSDs vs. prior generation Intel Xeon® Scalable platform supporting Intel DC P4610 SSDs

Up to 1.37x higher 4KB Rand Read/Write 70/30 IOPS performance with 3rd Gen Intel Xeon Scalable platform with Intel® SSD DC 5510 series vs. prior generation Intel Xeon® Scalable platform supporting Intel DC P4610 SSDs

Up to 2.6x higher 4KB Random Read IOPS performance with 3rd Gen Intel Xeon® scalable platform supporting PCIe Gen4 Intel Optane™ SSDs vs. prior generation Intel Xeon® Scalable platform supporting Intel DC P4610 SSDs

Up to 1.5x higher 4KB Random Read IOPS performance with 3rd Gen Intel Xeon® scalable platform and Intel DC P5510 SSDs vs. prior generation Intel Xeon® Scalable platform supporting Intel DC P4610 SSDs

Upgrade to the latest 3rd Gen Intel Xeon Scalable family and Latest Intel® SSDs and benefit from significantly lower latency and enhanced performance Up to 15% Lower latency with Intel DC P5510 SSDs and up to 94% lower latency with Intel Optane™ SSDs

Baseline: February 10, 2021

Up to 1.5x more 4KB Random Read IOPS/VM performance with 3rd Gen Intel Xeon® scalable platform supporting PCIe Gen4 Intel Optane™ SSDs vs. prior generation Intel Xeon® Scalable platform supporting Intel DC P4610 SSDs

Up to 1.3x more 4KB Random Read IOPS/VM performance with 3rd Gen Intel Xeon® scalable platform and Intel DC P5510 SSDs vs. prior generation Intel Xeon® Scalable platform supporting Intel DC P4610 SSDs

Up to 286K IOPS/VM on 3rd Gen Intel Xeon® scalable platform and PCIe Gen4 Intel Optane™ SSDs for 4KB Random Read vs. prior generation Intel Xeon® Scalable platform supporting Intel DC P4610 SSDs

Up to 192K IOPS/VM on 3rd Gen Intel Xeon® scalable platform and Intel DC P5510 SSDs for 4KB Random Read vs. prior generation Intel Xeon® Scalable platform supporting Intel DC P4610 SSDs

Baseline: February 10, 2021

Up to 2.7x higher IOPS throughput (4K random 70R/30W) for NVMe-over-TCP with the 3rd Gen Intel Xeon Scalable platform with Intel® Optane™ SSD P5800X Series vs. prior generation Intel Xeon® Scalable platform supporting Intel DC P4610 SSDs

Up to 2.7x higher IOPS throughput (4K random 70R/30W) for NVMe-over-TCP with the 3rd Gen Intel Xeon Scalable platform with Intel® SSD DC 5510 series vs. prior generation Intel Xeon® Scalable platform supporting Intel DC P4610 SSDs

Baseline: February 10, 2021

Baseline: March 26, 2021

Baseline: February 15, 2021

Baseline: February 15, 2021

Baseline: March 15, 2021

Baseline: March 19, 2021

Up to 5.13x increase in connections per second - Tencent TGW NGINX TLS1.2 Webserver Connection-Per-Second Performance on 3nd Generation Intel® Xeon® Processor Scalable Family

New: Test by Intel as of 03/19/2021. 2-node, 2x 3rd Gen Intel® Xeon® Scalable Processor, 36 cores HT On Turbo ON Total Memory 256 GB (16 slots/ 16GB/ 3200 MHz), BIOS: SE5C6200.86B.3020.P19.2103170131 (ucode: 0x8d05a260), CentOS 8.3, 4.18.0-240.1.1.el8_​3.x86_​64, gcc 8.3.1 compiler, PRNet Model, Deep Learning Framework: Intel-Tensorflow 2.4.0, https://github.com/Intel-tensorflow/tensorflow/releases/tag/v2.4.0, BS=1, Dummy Data, 18 instances/2 sockets, Datatype: FP32/INT8

Baseline: Test by Intel as of 03/19/2021. 2-node, 2x 2nd Gen Intel® Xeon® Scalable Processor, 24 cores HT On Turbo ON Total Memory 192 GB (12 slots/ 16GB/ 2933 MHz), BIOS: SE5C620.86B.0D.01.0438.032620191658(ucode:0x5003003), CentOS 8.3, 4.18.0-240.10.1.el8_​3.x86_​64, gcc 8.3.1 compiler, PRNet Model, Deep Learning Framework: Intel-Tensorflow 2.4.0, https://github.com/Intel-tensorflow/tensorflow/releases/tag/v2.4.0, BS=1, Dummy Data, 12 instances/2 sockets, Datatype: FP32/INT8

Tencent TGW:

New: Test by Intel as of 3/19/2021. 1-node, 2x 3rd Gen Intel® Xeon® Scalable Processor, 36 cores HT On Turbo ON Total Memory 256 GB (16 slots/ 16GB/ 3200 MHz), BIOS: SE5C6200.86B.3020.P19.2103170131 (ucode: 0x8d05a260), CentOS 8.3, 4.18.0-240.1.1.el8_​3.x86_​64, gcc 8.3.1 compiler, NGINX 1.18, OpenSSL 1.1.1f, QAT Engine 0.6.4, Ipp Crypto MB 2020 update3

Baseline: Test by Intel as of 3/13/2021. 1-node, 2x 2nd Gen Intel® Xeon® Scalable Processor, 24 cores HT On Turbo ON Total Memory 192 GB (12 slots/ 16GB/ 2933 MHz), BIOS: SE5C620.86B.02.01.0013.121520200651 (ucode:0x5003003), CentOS 8.3, 4.18.0-240.10.1.el8_​3.x86_​64, gcc 8.3.1 compiler, NGINX 1.18, OpenSSL 1.1.1f

Baseline: March 19, 2021

Baseline: March 19, 2021

Baseline: February 19, 2021

Intel® Optane™ persistent memory pricing & DRAM pricing referenced in TCO calculations is provided for guidance and planning purposes only and does not constitute a final offer. Pricing guidance is subject to change and may revise up or down based on market dynamics. Please contact your OEM/distributor for actual pricing. Pricing guidance as of March, 2021.

Baseline: March 23, 2021

Up to 18% better performance and 20% lower average latency on Cloud Microservices vs. AMD EPYC.

New: Super Micro SYS-220U-TNR, 2x Intel® Xeon Platinum 8358 (32C, 2.6GHz, 250W TDP), HT On, Turbo ON, SNC OFF, Total Memory: 512 GB (16 slots/ 32GB/ 3200 MHz), ucode: x280, 100GbE Network, RHEL 8.4, Kernel: 4.18.0-305.el8.x86_​64. Load generator: wrk, DeathStarBench: https://github.com/delimitrou/DeathStarBench, B509c933faca3e5b4789c6707d3b3976537411a9 (base-commit+ Intel Optimizations that are not released as of July 2021). Geomean of Hotel Resv, Social Net, and Media workloads. Tested by Intel as of July 2021.

AMD EPYC: Dell PowerEdge R7525, 2x AMD 7543 (32C, 2.8GHz, 240W cTDP), SMT On, Boost ON, NPS=1, Total Memory: 1 TB (16 slots/ 64GB/ 3200 MHz), ucode: 0xa00111d, 100GbE Network, RHEL 8.4, Kernel: 4.18.0-305.el8.x86_​64. Load generator: wrk, DeathStarBench: https://github.com/delimitrou/DeathStarBench, B509c933faca3e5b4789c6707d3b3976537411a9 (base-commit+ Intel Optimizations that are not released as of July 2021). Geomean of Hotel Resv, Social Net, and Media workloads. Tested by Intel as of July 2021.

Baseline: July 2021

Up to 6% lower cost per user with Intel® Optane™ PMem on VDI vs. AMD EPYC at equal software VMware licensing.

New: 4-node, Each node: Intel Software Development Platform 1TB PMem, 2x Intel® Xeon Platinum 8358 (32C, 2.6GHz, 250W TDP), HT On, Turbo ON, SNC OFF, Total Memory: 1 TB (16 slots/ 16GB/ 3200 MHz + 8 slots/128GB/PMem), ucode: x280, 2x 25GbE Intel Ethernet E810-XXVDA2, Gen4 x8, Per node cache tier: 2x 400GB Optane P5800x, Gen4, Per node capacity tier: 6x Intel P5510 3.84TB , Gen4, ESXi 7.0u2 17630552, vCenter 7.0u2 17694817, Citrix Virtual Apps and Desktops 7 2103, LoginVSI 4.1.40. Knowledge worker profile 2vCPU/4GB. 4-node solution cost $379K (Per node: base system @ $13757,https://www.thinkmate.com/system/rax-xt10-21s3-aiom, 1TB of Optane: 16x16GB DDR4 + 8x128GB Optane PMem 200 $12136, Cache tier: 2x Optane P5800X drives $1162.79/each, https://www.cdw.com/product/intel-optane-ssd-dc-p5800x-series-solid-state-drive-400-gb-pci-expres/6457148?pfm=srh, Capacity tier: 6x P5510 drives $758.09/each, https://www.cdw.com/product/intel-solid-state-drive-d7-p5510-series-solid-state-drive-3.84-tb-u.2/6509262?pfm=srh, 3 year of VMware S/W: ESXi License, https://www.cdw.com/product/vmware-vsan-enterprise-plus-v.-7-license-1-processor/6030177?pfm=srh, vSAN License, https://www.cdw.com/product/vmware-support-and-subscription-production-technical-support-for-vmware/6030176?pfm=srh, ESXi Support, https://www.cdw.com/product/vmware-support-and-subscription-production-technical-support-for-vmware/6030336?pfm=srh, vSAN support, https://www.cdw.com/product/vmware-vsphere-enterprise-plus-v.-7-license-1-processor/6030328?pfm=srh, $28938.50, Citrix per user: $144.39, https://www.cdw.com/product/citrix-virtual-apps-and-desktops-advanced-edition-license-1-user-device/3625316?pfm=srh). Tested by Evaluator Group as of July 2021 and prices as of July 2021.

AMD EPYC: 4-node, Each node: Dell PowerEdge R7525, 2x AMD 7543 (32C, 2.8GHz, 240W cTDP), SMT On, Boost ON, NPS=1, Total Memory: 1 TB (16 slots/ 64GB/ 3200 MHz), ucode: 0x1911000A00, 2x25GbE Mellanox Connectx-5, Gen4 x8, Per node cache tier: 2x 1.92TB Samsung PM1733 Gen4, Per node capacity tier: 6x Intel P5510 3.84TB, Gen4, ESXi 7.0u2 17630552, vCenter 7.0u2 17694817, Citrix Virtual Apps and Desktops 7 2103, LoginVSI 4.1.40. Knowledge worker profile 2vCPU/4GB. 4-node solution cost $410K (Per node: base system @ $13414, https://www.thinkmate.com/system/rax-qn10-21e2, 1TB DDR memory: 16x64GB DDR4 $21200, Cache tier: 2x Samsung PM1733 $478.79/each, https://www.cdw.com/product/samsung-pm1733-mzwlj1t9hbjr-solid-state-drive-1.92-tb-pci-express-4.0/6247494?pfm=srh, Capacity tier: 6x P5510 drives $758.09/each, https://www.cdw.com/product/intel-solid-state-drive-d7-p5510-series-solid-state-drive-3.84-tb-u.2/6509262?pfm=srh, 3 year of VMware S/W: ESXi License, https://www.cdw.com/product/vmware-vsan-enterprise-plus-v.-7-license-1-processor/6030177?pfm=srh, vSAN License, https://www.cdw.com/product/vmware-support-and-subscription-production-technical-support-for-vmware/6030176?pfm=srh, ESXi Support, https://www.cdw.com/product/vmware-support-and-subscription-production-technical-support-for-vmware/6030336?pfm=srh, vSAN support, https://www.cdw.com/product/vmware-vsphere-enterprise-plus-v.-7-license-1-processor/6030328?pfm=srh, $28938.50, Citrix per user: $144.39, https://www.cdw.com/product/citrix-virtual-apps-and-desktops-advanced-edition-license-1-user-device/3625316?pfm=srh). Tested by Evaluator Group as of July 2021 and prices as of July 2021.

Baseline: July 2021

Up to 39% higher throughput and 26% lower latency on VMware vSAN hyper-converged infrastructure performance vs. AMD EPYC.

New: 4-node, Each node: Intel Software Development Platform, 2x Intel® Xeon Platinum 8358 (32C, 2.6GHz, 250W TDP), HT On, Turbo ON, SNC OFF, Total Memory: 1 TB (16 slots/ 64GB/ 3200 MHz), ucode: x280, 2x 25GbE Intel Ethernet E810-XXVDA2, Gen4 x8, Per node cache tier: 2x 400GB Optane P5800x, Gen4, Per node capacity tier: 6x Intel P5510 3.84TB , Gen4, ESXi 7.0u2 17630552, vCenter 7.0u2 17694817, HCIBench 2.5.3, Vdbench 50407. Geomean of 7 test cases (R, R/W, W, varying block sizes, random and sequential). Tested by Evaluator Group as of May 2021.

AMD EPYC: 4-node, Each node: Dell PowerEdge R7525, 2x AMD 7543 (32C, 2.8GHz, 240W cTDP), SMT On, Boost ON, NPS=1, Total Memory: 1 TB (16 slots/ 64GB/ 3200 MHz), ucode: 0x1911000A00, 2x25GbE Mellanox Connectx-5, Gen4 x8, Per node cache tier: 2x 1.92TB Samsung PM1733 Gen4, Per node capacity tier: 6x Intel P5510 3.84TB , Gen4, ESXi 7.0u2 17630552, vCenter 7.0u2 17694817,1, HCIBench 2.5.3, Vdbench 50407, Geomean of 7 test cases (R, R/W, W, varying block sizes, random and sequential). Tested by Evaluator Group as of May 2021.

Baseline: May 2021

1.57x better performance on LAMMPS Geomean of Polyethylene, Stillinger-Weber, Tersoff, Water: New: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: v2020-10-29; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4, Intel MPI 2019u8; threads/core: 2; Turbo: used; Build knobs: -O3 -ip -xCORE-AVX512 -qopt-zmm-usage=high. AMD EPYC: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: v2020-10-29; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4, Intel MPI 2019u8; threads/core: 2; Turbo: used; Build knobs: -O3 -ip -march=core-avx2, tested by Intel and results as of April 2021.

1.62x better performance on NAMD Geomean of Apoa1, STMV: New: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: 2.15-Alpha1 (includes AVX tiles algorithm); Build notes: Tools: Intel MKL, Intel C Compiler 2020u4, Intel MPI 2019u8, Intel Threading Building Blocks 2020u4; threads/core: 2; Turbo: used; Build knobs: -ip -fp-model fast=2 -no-prec-div -qoverride-limits -qopenmp-simd -O3 -xCORE-AVX512 -qopt-zmm-usage=high. AMD EPYC: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: 2.15-Alpha1 (includes AVX tiles algorithm); Build notes: Tools: Intel MKL, AOCC 2.2.0, gcc 9.3.0, Intel MPI 2019u8; threads/core: 2; Turbo: used; Build knobs: -O3 -fomit-frame-pointer -march=znver1 -ffast-math, tested by Intel and results as of April 2021.

1.68x better performance on RELION Plasmodium Ribosome: New:Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: 3_​1_​1; Build notes: Tools: Intel C Compiler 2020u4, Intel MPI 2019u9; threads/core: 2; Turbo: used; Build knobs: -O3 -ip -g -debug inline-debug-info -xCOMMON-AVX512 -qopt-report=5 -restrict. AMD EPYC: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: 3_​1_​1; Build notes: Tools: Intel C Compiler 2020u4, Intel MPI 2019u9; threads/core: 2; Turbo: used; Build knobs: -O3 -ip -g -debug inline-debug-info -march=core-avx2 -qopt-report=5 -restrict, tested by Intel and results as of April 2021.

1.37x better performance on Binomial Options: New: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: v1.0; Build notes: Tools: Intel C Compiler 2020u4, Intel Threading Building Blocks ; threads/core: 2; Turbo: used; Build knobs: -O3 -xCORE-AVX512 -qopt-zmm-usage=high -fimf-domain-exclusion=31 -fimf-accuracy-bits=11 -no-prec-div -no-prec-sqrt. AMD EPYC: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: v1.0; Build notes: Tools: Intel C Compiler 2020u4, Intel Threading Building Blocks ; threads/core: 2; Turbo: used; Build knobs: -O3 -march=core-avx2 -fimf-domain-exclusion=31 -fimf-accuracy-bits=11 -no-prec-div -no-prec-sqrt, tested by Intel and results as of April 2021.

2.05x better performance on Monte Carlo: New: Platinum 8358: 1-node, 2x Intel® Xeon® Platinum 8358 (32C/2.6GHz, 250W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: v1.1; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4; threads/core: 1; Turbo: used; Build knobs: -O3 -xCORE-AVX512 -qopt-zmm-usage=high -fimf-precision=low -fimf-domain-exclusion=31 -no-prec-div -no-prec-sqrt. AMD EPYC: 1-node, 2-socket AMD EPYC 7543 (32C/2.8GHz, 240W cTDP) on Dell PowerEdge R7525 server with 1024 GB (16 slots/ 64GB/3200) total DDR4 memory, ucode 0xa001119, SMT on, Boost on, Power deterministic mode, NPS=4, Red Hat Enterprise Linux 8.3, 4.18, 2x Micron 5300 Pro, App Version: v1.1; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4; threads/core: 2; Turbo: used; Build knobs: -O3 -march=core-avx2 -fimf-precision=low -fimf-domain-exclusion=31 -no-prec-div -no-prec-sqrt, tested by Intel and results as of April 2021.

Baseline: April 2021

Baseline: February 2021

New: 1-node, 2x Intel Xeon Platinum 8380 processor on Coyote Pass with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode X55260, HT on, Turbo on, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Intel_​SSDSC2KG96, Intel SSDPE2KX010T8, MobileNet-v1, gcc-9.3.0, oneDNN 1.6.4, BS=1,56, INT8, TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow- 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart/, test by Intel on March 2021.

Baseline:1-node, 2x AMD EPYC 7763 on GigaByte with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode 0xa001114, HT on, Turbo on, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Samsung_​MZ7LH3T8, MobileNet-v1, gcc-9.3.0, oneDNN 1.6.4, BS=1,56, FP32, TensorFlow- 2.4.1, Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/benchmarks/image_recognition/tensorflow/mobilenet_v1, tested by Intel and results as of March 2021.

Baseline: March 2021

3rd Gen Intel Xeon Platinum 8380: 1-node, 2x Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode X55260, HT on, Turbo on, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Intel_​SSDSC2KG96, Intel SSDPE2KX010T8, tested by Intel, and results as of March 2021.

DL Measurements on A100: 1-node, 2-socket AMD EPYC 7742 (64C) with 256GB (8 slots/ 32GB/ 3200) total DDR4 memory, ucode 0x8301038, HT on, Turbo on, Ubuntu 20.04 LTS, 5.4.0-42-generic, INTEL SSDSC2KB01, NVIDIA A100-PCIe-40GB, HBM2-40GB, Accelerator per node =1, tested by Intel, and results as of March 2021.

ML Measurements on A100: 1-node, 2-socket AMD EPYC 7742 (64C) with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode 0x8301034, HT on, Turbo on, Ubuntu 18.04.5 LTS, 5.4.0-42-generic,NVIDIA A100 (DGX-1), 1.92TB M.2 NVMe, 1.92TB M.2 NVMe RAID tested by Intel, and results as of March 2021.

ResNet50-v1.5 Intel: gcc-9.3.0, oneDNN 1.6.4, BS=1, INT8, TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow- 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart

ResNet50-v1.5 NVIDIA:A100 (7 instance/GPU), BS=1,TensorFlow - 1.5.5 (NGC: tensorflow:21.02-tf1-py3), https://github.com/NVIDIA/DeepLearningExamples/tree/master/TensorFlow/Classification/ConvNets/resnet50v1.5, TF AMP (FP16+TF32);

ResNet50-v1.5 Training Intel: gcc-9.3.0, oneDNN 1.6.4, BS=256, FP32, TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow- 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart.

ResNet50-v1.5 Training NVIDIA:A100, BS=256,TensorFlow - 1.5.5 (NGC: tensorflow:21.02-tf1-py3),https://github.com/NVIDIA/DeepLearningExamples/tree/master/TensorFlow/Classification/ConvNets/resnet50v1.5, TF32; BERT-Large SQuAD Intel: gcc-9.3.0, oneDNN 1.6.4, BS=1, INT8,

TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow- 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart/ A100: BERT-Large SQuAD, BS=1, A100 (7 instance/GPU), TensorFlow - 1.5.5 (NGC: tensorflow:20.11-tf1-py3), https://github.com/NVIDIA/DeepLearningExamples/tree/master/TensorFlow/LanguageModeling/BERT,TF AMP (FP16+TF32) ; SSD-ResNet34 Intel: gcc-9.3.0, oneDNN 1.6.4, BS=1, INT8,

TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow- 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart/, SSD-ResNet34 INVIDIA: A100 (7 instance/GPU), BS=1,Pytorch - 1.8.0a0 (NGC Container, latest supported): A100: SSD-ResNet34 (NGC: pytorch:20.11-py3), https://github.com/NVIDIA/DeepLearningExamples/tree/master/PyTorch/Detection/SSD, AMP (FP16 +TF32) ; Python: Intel: Python 3.7.9, Scikit-Learn: Sklearn 0.24.1, OneDAL: Daal4py 2021.2, XGBoost: XGBoost 1.3.3 Python: NVIDIA A100: Python 3.7.9, Scikit-Learn: Sklearn 0.24.1, CuML 0.17, XGBoost 1.3.0dev.rapidsai0.17, Nvidia RAPIDS: RAPIDS 0.17, CUDA Toolkit: CUDA 11.0.221 Benchmarks: https://github.com/IntelPython/scikit-learn_bench

Baseline: March 2021

3rd Gen Intel Xeon: 8380: 1-node, 2x Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode X55260, HT on, Turbo on, Ubuntu 20.04 LTS, 5.4.0-65-generic/5.4.0-64-generic, 1x Intel_​SSDSC2KG96, Intel SSDPE2KX010T8, tested by Intel, and results as of March 2021.

AMD: EPYC 7763: 1-node, 2-socket AMD EPYC 7763 (64C/2.45GHz, 280W cTDP) on GIGABYTE R282-Z92 server with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode 0xa001114, SMT on, Boost on, Power deterministic mode, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Samsung_​MZ7LH3T8/INTEL SSDSC2KG019T8, tested by Intel, and results as of March 2021.

ResNet50-v1.5 Intel: gcc-9.3.0, oneDNN 1.6.4, BS=128, INT8, TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow- 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart, ResNet50-v1.5 AMD: gcc-9.3.0, oneDNN 1.6.4, BS=128, FP32, TensorFlow- 2.4.1, Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/benchmarks/image_recognition/tensorflow/resnet50v1_5

ResNet50-v1.5 Training Intel: gcc-9.3.0, oneDNN 1.6.4, BS=256, FP32, TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow- 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart, ResNet50-v1.5 Training AMD: gcc-9.3.0, oneDNN 1.6.4, BS=256, FP32, TensorFlow- 2.4.1, Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/benchmarks/image_recognition/tensorflow/resnet50v1_5 SSD-ResNet34 Intel: gcc-9.3.0, oneDNN 1.6.4, BS=1, INT8,

TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow- 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart/, AMD: SSD-ResNet34, gcc-9.3.0, oneDNN 1.6.4, BS=1, FP32, TensorFlow- 2.4, Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/benchmarks/object_detection/tensorflow/ssd-resnet34 BERT-Large SQuAD Intel: gcc-9.3.0, oneDNN 1.6.4, BS=1, INT8,

TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow- 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart/, AMD: BERT-Large SQuAD, gcc-9.3.0, oneDNN 1.6.4, BS=1, FP32, TensorFlow- 2.4.1, Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/benchmarks/language_modeling/tensorflow/bert_large Python: Python 3.7.9, SciKit-Learn: Sklearn 0.24.1, oneDAL: Daal4py 2021.2, XGBoost: XGBoost 1.3.3: Benchmarks: https://github.com/IntelPython/scikit-learn_bench

Baseline: March 2021

With Intel® 3rd Gen Xeon® Scalable processors, CoSP's can increase 5G UPF performance by 42%. Combined with Intel Ethernet 800 series adapters, they can deliver the performance, efficiency and trust for use cases that require low latency, including augmented reality, cloud-based gaming, discrete automation and even robotic-aided surgery.

With Intel® 3rd Gen Intel® Xeon® Scalable processors, Ethernet 800 series and vRAN dedicated accelerators, CoSP's can get up to 1.81x MIMO Midhaul Throughput in a similar power envelope for a best-in-class 3x100mhz 64T64R configuration.

FleXRAN: New: 1 node, 1 socket Intel Xeon Gold 6338N Processor, 32 core HT ON Turbo ON, Total DRAM 128 GB (8 slots/16GB/2666), BIOS WLYDCRB.SYS.WR.64.2021.09.4.04.0636_​0020.P86_​P80260_​LBG_​SPS_​8d055260_​EARLYG (ucode 0x261), Intel Mount Bryce (ACC100), CentOS 7.8.2003, 3.10.0-1127.19.1.rt56.1116.el7.x86_​64, FlexRAN L1 Massive MIMO, tested by Intel on 3/18/2021. Baseline: 1 node, 1 socket Intel Xeon Gold 6212U Processor, 24 core HT ON Turbo ON, Total Dram 96 GB (6 slots/16GB/2993), BIOS SE5C620.86B.02.01.0012.070720200218, Intel Mount Bryce (ACC100), CentOS 7.8.2003, 3.10.0-1127.19.1.rt56.1116.el7.x86_​64, FlexRAN L1 Massive MIMO, tested by Intel on 2/23/2021.

Baseline: February 15, 2021

Up to 53% better performance, 66% higher VM density, and 46% better performance per dollar on VMware ESXi/vSAN virtualization vs. AMD EPYC.

New: 4-node, Each node: Intel Software Development Platform 2TB PMem , 2x Intel® Xeon Platinum 8358 (32C, 2.6GHz, 250W TDP), HT On, Turbo ON, SNC OFF, Total Memory: 2 TB (16 slots/ 32GB/ 3200 MHz + 16 slots/128GB/PMem), ucode: x280, 2x 25GbE Intel Ethernet E810-XXVDA2, Gen4 x8, Per node cache tier: 2x 400GB Optane P5800x, Gen4, Per node capacity tier: 6x Intel P5510 3.84TB , Gen4, ESXi 7.0u2 17630552, vCenter 7.0u2 17694817, 4-node solution cost $293K (Per node: base system @ $13757, https://www.thinkmate.com/system/rax-xt10-21s3-aiom, 2TB of Optane: 16x32GB DDR4 + 16x128GB Optane PMem 200 $23728, Cache tier: 2x Optane P5800X drives $1162.79/each, https://www.cdw.com/product/intel-optane-ssd-dc-p5800x-series-solid-state-drive-400-gb-pci-expres/6457148?pfm=srh, Capacity tier: 6x P5510 drives $758.09/each, https://www.cdw.com/product/intel-solid-state-drive-d7-p5510-series-solid-state-drive-3.84-tb-u.2/6509262?pfm=srh, 3 year of VMware S/W: ESXi License https://www.cdw.com/product/vmware-vsan-enterprise-plus-v.-7-license-1-processor/6030177?pfm=srh, vSAN License https://www.cdw.com/product/vmware-support-and-subscription-production-technical-support-for-vmware/6030176?pfm=srh, ESXi Support, https://www.cdw.com/product/vmware-support-and-subscription-production-technical-support-for-vmware/6030336?pfm=srh, vSAN support, https://www.cdw.com/product/vmware-vsphere-enterprise-plus-v.-7-license-1-processor/6030328?pfm=srh, $28938.50). Tested by Evaluator Group as of July 2021 and prices as of July 2021.

AMD EPYC: 4-node, Each node: Dell PowerEdge R7525, 2x AMD 7543 (32C, 2.8GHz, 240W cTDP), SMT On, Boost ON, NPS=1, Total Memory: 1 TB (16 slots/ 64GB/ 3200 MHz), ucode: 0x1911000A00, 2x25GbE Mellanox Connectx-5, Gen4 x8, Per node cache tier: 2x 1.92TB Samsung PM1733 Gen4, Per node capacity tier: 6x Intel P5510 3.84TB , Gen4, ESXi 7.0u2 17630552, vCenter 7.0u2 17694817, 4-node solution cost $276K (Per node: base system @ $13414, https://www.thinkmate.com/system/rax-qn10-21e2, 1TB DDR memory: 16x64GB DDR4 $21200, Cache tier: 2x Samsung PM1733 $478.79/each, https://www.cdw.com/product/samsung-pm1733-mzwlj1t9hbjr-solid-state-drive-1.92-tb-pci-express-4.0/6247494?pfm=srh, Capacity tier: 6x P5510 drives $758.09/each, https://www.cdw.com/product/intel-solid-state-drive-d7-p5510-series-solid-state-drive-3.84-tb-u.2/6509262?pfm=srh, 3 year of VMware S/W: ESXi License, https://www.cdw.com/product/vmware-vsan-enterprise-plus-v.-7-license-1-processor/6030177?pfm=srh, vSAN License, https://www.cdw.com/product/vmware-support-and-subscription-production-technical-support-for-vmware/6030176?pfm=srh, ESXi Support, https://www.cdw.com/product/vmware-support-and-subscription-production-technical-support-for-vmware/6030336?pfm=srh, vSAN support, https://www.cdw.com/product/vmware-vsphere-enterprise-plus-v.-7-license-1-processor/6030328?pfm=srh, $28938.50). Tested by Evaluator Group as of July 2021 and prices as of July 2021.

Baseline: July 2021

2.03x higher OpenSSL ECDHE x25519 performance 3rd Gen Intel® Xeon® Scalable vs. AMD Milan

2.03x higher OpenSSL ECDHE x25519 performance

3rd Gen Intel Xeon: 8380: 1-node, 2x Intel® Xeon® Platinum 8380 CPU on M50CYP2SB2U with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode 0xd000270, HT On, Turbo Off, Ubuntu 20.04.1 LTS, 5.4.0-65-generic, 1x INTEL_​SSDSC2KG01, OpenSSL 1.1.1j, GCC 9.3.0, QAT Engine v0.6.4, Tested by Intel and results as of March 2021.

AMD: 7763: 1-node, 2x AMD EPYC 7763 64-Core Processor on R282-Z92-00 with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode 0xa001114, HT On, Turbo Off, Ubuntu 20.04.1 LTS, 5.4.0-65-generic, 1x SAMSUNG_​MZ7LH3T8, OpenSSL 1.1.1j, GCC 9.3.0, Tested by Intel and results as of March 2021.

Baseline: March 2021

3rd Gen Intel Xeon: 8380: 1-node, 2x Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: The Intel Distribution for LINPACK Benchmark; Build notes: Tools: Intel MPI 2019u7; threads/core: 1; Turbo: used; Build: build script from Intel Distribution for LINPACK package; 1 rank per NUMA node: 1 rank per socket

AMD: 7763: 1-node, 2-socket AMD EPYC 7763 (64C/2.45GHz, 280W cTDP) on GIGABYTE R282-Z92 server with 512 GB (16 slots/ 32GB/3200) total DDR4 memory, ucode 0xa001114, SMT on, Boost on, Power deterministic mode, NPS=4, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Samsung_​MZ7LH3T8, App Version: AMD official HPL 2.3 MT version with BLIS 2.1; Build notes: Tools: hpc-x 2.7.0; threads/core: 1; Turbo: used; Build: pre-built binary (gcc built) from https://developer.amd.com/amd-aocl/blas-library/; 1 rank per L3 cache, 4 threads per rank Tested by Intel and results as of March 2021

Baseline: March 2021

3rd Gen Intel Xeon: 8380: 1-node, 2x 3rd Gen Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: 3_​1_​1; Build notes: Tools: Intel C Compiler 2020u4, Intel MPI 2019u9; threads/core: 2; Turbo: used; Build knobs: -O3 -ip -g -debug inline-debug-info -xCOMMON-AVX512 -qopt-report=5 -restrict

AMD: 7763: 1-node, 2-socket AMD EPYC 7763 (64C/2.45GHz, 280W cTDP) on GIGABYTE R282-Z92 server with 512 GB (16 slots/ 32GB/3200) total DDR4 memory, ucode 0xa001114, SMT on, Boost on, Power deterministic mode, NPS=4, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Samsung_​MZ7LH3T8, App Version: 3_​1_​1; Build notes: Tools: Intel C Compiler 2020u4, Intel MPI 2019u9; threads/core: 2; Turbo: used; Build knobs: -O3 -ip -g -debug inline-debug-info -march=core-avx2 -qopt-report=5 -restrict Tested by Intel and results as of March 2021

Baseline: March 2021

3rd Gen Intel Xeon:8380: 1-node, 2x 3rd Gen Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: v1.1; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4; threads/core: 1; Turbo: used; Build knobs: -O3 -xCORE-AVX512 -qopt-zmm-usage=high -fimf-precision=low -fimf-domain-exclusion=31 -no-prec-div -no-prec-sqrt

AMD: 7763: 1-node, 2-socket AMD EPYC 7763 (64C/2.45GHz, 280W cTDP) on GIGABYTE R282-Z92 server with 512 GB (16 slots/ 32GB/3200) total DDR4 memory, ucode 0xa001114, SMT on, Boost on, Power deterministic mode, NPS=4, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Samsung_​MZ7LH3T8, App Version: v1.1; Build notes: Tools: Intel MKL 2020u4, Intel C Compiler 2020u4, Intel Threading Building Blocks 2020u4; threads/core: 2; Turbo: used; Build knobs: -O3 -march=core-avx2 -fimf-precision=low -fimf-domain-exclusion=31 -no-prec-div -no-prec-sqrt Tested by Intel and results as of March 2021

Baseline: March 2021

3rd Gen Intel Xeon: 8380: 1-node, 2x 3rd Gen Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 256 GB (16 slots/ 16GB/ 3200) total DDR4 memory, ucode 0x261, HT on, Turbo on, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Intel_​SSDSC2KG96, App Version: 2.15-Alpha1 (includes AVX tiles algorithm); Build notes: Tools: Intel MKL, Intel C Compiler 2020u4, Intel MPI 2019u8, Intel Threading Building Blocks 2020u4; threads/core: 2; Turbo: used; Build knobs: -ip -fp-model fast=2 -no-prec-div -qoverride-limits -qopenmp-simd -O3 -xCORE-AVX512 -qopt-zmm-usage=high

AMD: 7763: 1-node, 2x AMD EPYC 7763 (64C/2.45GHz, 280W cTDP) on GIGABYTE R282-Z92 server with 512 GB (16 slots/ 32GB/3200) total DDR4 memory, ucode 0xa001114, SMT on, Boost on, Power deterministic mode, NPS=4, CentOS Linux 8.3.2011, 4.18.0-240.1.1.el8_​3.crt1.x86_​64, 1x Samsung_​MZ7LH3T8, App Version: 2.15-Alpha1 (includes AVX tiles algorithm); Build notes: Tools: Intel MKL, AOCC 2.2.0, gcc 9.3.0, Intel MPI 2019u8; threads/core: 2; Turbo: used; Build knobs: -O3 -fomit-frame-pointer -march=znver1 -ffast-math Tested by Intel and results as of March 2021

Baseline: March 2021

3rd Gen Intel Xeon:8380: 1-node, 2x 3rd Gen Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode X55260, HT on, Turbo on, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Intel_​SSDSC2KG96, Intel SSDPE2KX010T8, SSD-ResNet34, gcc-9.3.0, oneDNN 1.6.4, BS=1 INT8, TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow- 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart/, tested by Intel, and results as of March 2021.

AMD: 7763: 1-node, 2-socket AMD EPYC 7763 (64C/2.45GHz, 280W cTDP) on GIGABYTE R282-Z92 server with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode 0xa001114, SMT on, Boost on, Power deterministic mode, NPS=1, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Samsung_​MZ7LH3T8, SSD-ResNet34, gcc-9.3.0, oneDNN 1.6.4, BS=1 FP32, TensorFlow- 2.4.1, Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/benchmarks/object_detection/tensorflow/ssd-resnet34, tested by Intel, and results as of March 2021.

Baseline: March 2021

3rd Gen Intel Xeon: 8380: 1-node, 2x Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode X55260, HT on, Turbo on, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Intel_​SSDSC2KG96, Intel SSDPE2KX010T8, BERT Large SQuAD, gcc-9.3.0, oneDNN 1.6.4, BS=1,128, INT8, TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow- 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart/, tested by Intel, and results as of March 2021.

AMD: 7763: 1-node, 2-socket AMD EPYC 7763 (64C/2.45GHz, 280W cTDP) on GIGABYTE R282-Z92 server with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode 0xa001114, SMT on, Boost on, Power deterministic mode, NPS=1, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Samsung_​MZ7LH3T8, BERT Large SQuAD, gcc-9.3.0, oneDNN 1.6.4, BS=1,128, FP32, TensorFlow- 2.4.1, Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/benchmarks/language_modeling/tensorflow/bert_large, tested by Intel, and results as of March 2021.

Baseline: March 2021

3rd Gen Intel Xeon: 8380: 1-node, 2x Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode X55260, HT on, Turbo on, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Intel_​SSDSC2KG96, Intel SSDPE2KX010T8, MobileNet-v1, gcc-9.3.0, oneDNN 1.6.4, BS=1,56, INT8, TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow- 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart/, tested by Intel, and results as of March 2021.

AMD: 7763: 1-node, 2-socket AMD EPYC 7763 (64C/2.45GHz, 280W cTDP) on GIGABYTE R282-Z92 server with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode 0xa001114, SMT on, Boost on, Power deterministic mode, NPS=1, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Samsung_​MZ7LH3T8, MobileNet-v1, gcc-9.3.0, oneDNN 1.6.4, BS=1, 56, FP32, TensorFlow- 2.4.1, Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/benchmarks/image_recognition/tensorflow/mobilenet_v1, tested by Intel, and results as of March 2021.

Baseline: March 2021

3rd Gen Intel Xeon: 8380: 1-node, 2x 3rd Gen Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode X55260, HT on, Turbo on, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Intel_​SSDSC2KG96, Intel SSDPE2KX010T8, SSD-MobileNet-v1, gcc-9.3.0, oneDNN 1.6.4, BS=1,448, INT8, TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart/, tested by Intel, and results as of March 2021.

AMD: 7763: 1-node, 2-socket AMD EPYC 7763 (64C/2.45GHz, 280W cTDP) on GIGABYTE R282-Z92 server with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode 0xa001114, SMT on, Boost on, Power deterministic mode, NPS=1, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Samsung_​MZ7LH3T8, SSD-MobileNet-v1, gcc-9.3.0, oneDNN 1.6.4, BS=1,448, FP32, TensorFlow- 2.4.1, Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/benchmarks/object_detection/tensorflow/ssd-mobilenet, tested by Intel, and results as of March 2021.

Baseline: March 2021

3rd Gen Intel Xeon: 8380: 1-node, 2x 3rd Gen Intel Xeon Platinum 8380 (40C/2.3GHz, 270W TDP) processor on Intel Software Development Platform with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode X55260, HT on, Turbo on, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Intel_​SSDSC2KG96, Intel SSDPE2KX010T8, ResNet50-v1.5, gcc-9.3.0, oneDNN 1.6.4, BS=1,128, INT8, TensorFlow 2.4.1 with Intel optimizations for 3rd Gen Intel Xeon Scalable processor, upstreamed to TensorFlow 2.5 (container- intel/intel-optimized-tensorflow:tf-r2.5-icx-b631821f), Model zoo: https://github.com/IntelAI/models/tree/icx-launch-public/quickstart/, tested by Intel, and results as of March 2021.

AMD:7763: 1-node, 2-socket AMD EPYC 7763 (64C/2.45GHz, 280W cTDP) on GIGABYTE R282-Z92 server with 512 GB (16 s/lots/ 32GB/ 3200) total DDR4 memory, ucode 0xa001114, SMT on, Boost on, Power deterministic mode, NPS=1, Ubuntu 20.04 LTS, 5.4.0-65-generic, 1x Samsung_​MZ7LH3T8, ResNet50-v1.5, gcc-9.3.0, oneDNN 1.6.4, BS=1,128, FP32, TensorFlow- 2.4.1, Model: https://github.com/IntelAI/models/tree/icx-launch-public/benchmarks/image_recognition/tensorflow/resnet50v1_5, tested by Intel, and results as of March 2021.

Baseline: March 2021

New: M6i.xlarge 4 vCPU, 16 GB memory capacity/instance, M6i.4xlarge 16 vCPU, 64 GB memory capacity/instance, M6i.16xlarge 64 vCPU, 256 GB memory capacity/instance (Xeon) with Crypto NI acceleration, WordPress v5.2, Storage/instance: Amazon Elastic Block Store 8GB, Compiler: GCC 9.3.0, Other SW: MariaDB v10.3.31, Nginx v1.18.0, PHP v7.3.30-1, Ubuntu 20.04.3 LTS, Kernel 5.11.0-1017-aws.

Baseline: September-October 2021

New: M6i.xlarge 4 vCPU, 16 GB memory capacity/instance, M6i.4xlarge 16 vCPU, 64 GB memory capacity/instance, M6i.16xlarge 64 vCPU, 256 GB memory capacity/instance (Xeon), Ubuntu 20.04.3 LTS, Kernel 5.11.0-1016-aws.

Baseline: M6g.xlarge 4 vCPU, 16 GB memory capacity/instance, m6g.4xlarge 16 vCPU, 64 GB memory capacity/instance, m6g.16xlarge 64 vCPU, 256 GB memory capacity/instance (Graviton2), Ubuntu 20.04.3 LTS, Kernel 5.11.0-1016-aws.

Baseline: September 2021

New: M6i.xlarge 4 vCPU, 16 GB memory capacity/instance, M6i.4xlarge 16 vCPU, 64 GB memory capacity/instance, M6i.16xlarge 64 vCPU, 256 GB memory capacity/instance (Xeon), Ubuntu 20.04.3 LTS, Kernel 5.11.0-1016-aws.

Baseline: M6g.xlarge 4 vCPU, 16 GB memory capacity/instance, m6g.4xlarge 16 vCPU, 64 GB memory capacity/instance, m6g.16xlarge 64 vCPU, 256 GB memory capacity/instance (Graviton2), Ubuntu 20.04.3 LTS, Kernel 5.11.0-1016-aws.

Baseline: October 2021

New: M6i.4xlarge 16 vCPU 64 GB memory capacity/instance, M6i.16xlarge 64 vCPU, 256 GB memory capacity/instance (Xeon), Redis 6.0.8, memtier 1.2.15, Network BW/Instance 12.5Gbps for M6i, Ubuntu 20.04 LTS, Kernel 5.11.0-1019-aws.

Baseline: M6g.4xlarge 16 vCPU 64 GB memory capacity/instance, M6g.16xlarge 64 vCPU, 256 GB memory capacity/instance (Graviton2), Redis 6.0.8, memtier 1.2.15, Network BW/Instance 10Gbps for M6g, Ubuntu 20.04 LTS, Kernel 5.11.0-1019-aws.

Baseline: September 2021

2.65x average performance gains with 3rd Gen Intel Xeon Platinum 8380 processor vs legacy E5-v4 server

3.1x average performance gains with 3rd Gen Intel Xeon Platinum 8380 processor vs legacy E5-v3 server

2.65x average performance gain - Ice Lake vs Broadwell: Geomean of 2.34x SPECrate2017_​int_​base (est), 2.6x SPECrate2017_​fp_​base (est), 2.55x Stream Triad, 3.18x Intel distribution of LINPACK.

3.1x average performance gain - Ice Lake vs Haswell: Geomean of 2.85x SPECrate2017_​int_​base (est), 3.08x SPECrate2017_​fp_​base (est), 2.8x Stream Triad, 3.97x Intel distribution of LINPACK.

New: 1-node, 2x Intel Xeon Platinum 8380 processor on Coyote Pass with 512 GB (16 slots/ 32GB/ 3200) total DDR4 memory, ucode 0x261, HT on (SPECcpu2017), off (others), Turbo on, Ubuntu 20.04, 5.4.0-66-generic, 1x S4610 SSD 960G, SPECcpu2017 (est) v1.1.0, Stream Triad, Linpack, ic19.1u2, MPI: Version 2019u9; MKL:2020.4.17, test by Intel on 3/15/2021.

Skylake Baseline: 1-node, 2x Intel Xeon Platinum 8180 processor on Wolf Pass with 192 GB (12 slots/ 16GB/ 2933[2666]) total DDR4 memory, ucode 0x2006a08, HT on (SPECcpu2017), off (others), Turbo on, Ubuntu 20.04, 5.4.0-62-generic, SPECcpu2017 (est) v1.1.0, Stream Triad, Intel distribution of LINPACK, ic19.1u2, MPI: Version 2019 Update 9 Build 20200923; MKL: psxe_​runtime_​2020.4.17, test by Intel on 1/27/21.

Broadwell Baseline: 1-node, 2x Intel Xeon processor E5-2699v4 on Wildcat Pass with 256 GB (8 slots/ 32GB/ 2400) total DDR4 memory, ucode 0x038, HT on (SPECcpu2017), off (others), Turbo on, Ubuntu 20.04, 5.4.0-62-generic, 1x S3700 400GB SSD, SPECcpu2017 (est) v1.1.0, Stream Triad, Intel distribution of LINPACK, ic19.1u2, MPI: Version 2019 Update 9 Build 20200923; MKL: psxe_​runtime_​2020.4.17, test by Intel on 1/17/21.

Haswell Baseline: 1-node, 2x Intel Xeon processor E5-2699v3 on Wildcat Pass with 256 GB (8 slots/ 32GB/ 2666[2133]) total DDR4 memory, ucode 0x44, HT on (SPECcpu2017), off (others), Turbo on, Ubuntu 20.04, 5.4.0-62-generic, 1x S3700 400GB SSD, SPECcpu2017 (est) v1.1.0, Stream Triad, Intel distribution of LINPACK, ic19.1u2, MPI: Version 2019 Update 9 Build 20200923; MKL: psxe_​runtime_​2020.4.17, test by Intel on 2/3/21.

Integer throughput/Floating Point throughput/STREAM/LINPACK

Baseline: Jan 17, 2021

Baseline: Jan 17, 2021

Baseline: Jan 17, 2021

Virtualization workload

Baseline: Jan 14, 2021

Baseline: Jan 27, 2021

Baseline: April 27, 2020

New- 3rd Gen Intel Xeon Scalable Processor: 1-node, 4x 3rd Gen Intel® Xeon® Platinum 8380H processor (pre-production 28C, 250W) on Intel Reference Platform (Cooper City) with 384 GB (24 slots / 16GB / 3200) total memory, ucode 0x700001b, HT on, Turbo on, with Ubuntu 20.04 LTS, Linux 5.4.0-29-generic, Intel SSD 800GB OS Drive, ResNet-50 v1.5 Throughput, https://github.com/Intel-tensorflow/tensorflow -b bf16/base, commit#6ef2116e6a09, Modelzoo: https://github.com/IntelAI/models/ -b v1.6.1, Imagenet dataset, oneDNN 1.4, FP32, BF16, global BS=1024, 4 instances, 28-cores/instance, test by Intel on 06/01//2020.

2nd Gen Intel Xeon Scalable Processor: 1-node, 4x Intel® Xeon® Platinum 8280 processor (28C, 205W) on Intel Reference Platform (Lightning Ridge) with 768 GB (24 slots / 32 GB / 2933 ) total memory, ucode 0x4002f00, HT on, Turbo on, with Ubuntu 20.04 LTS, Linux 5.4.0-29-generic, Intel SSD 800GB OS Drive, ResNet-50 v1.5 Throughput, https://github.com/Intel-tensorflow/tensorflow -b bf16/base, commit# 6ef2116e6a09, Modelzoo: https://github.com/IntelAI/models/ -b v1.6.1, Imagenet dataset, oneDNN 1.4, FP32, global BS=1024, 4 instances, 28-cores/instance, test by Intel on 06/01/2020.

Intel Xeon Scalable Processor: 1-node, 4x Intel® Xeon® Platinum 8180 processor (28C, 205W) on Intel Reference Platform (Lightning Ridge) with 768 GB (24 slots / 32 GB / 2666 ) total memory, ucode 0x2000069, HT on, Turbo on, with Ubuntu 20.04 LTS, 5.4.0-26-generic, Intel SSD 800GB OS Drive, Training: ResNet-50-v1.5,Inference: ResNet-50-v1.5 Throughput, https://github.com/Intel-tensorflow/tensorflow -b bf16/base, commit#6ef2116e6a09, Modelzoo: https://github.com/IntelAI/models/ -b v1.6.1, Imagenet dataset, oneDNN 1.4, FP32, global BS=1024, 4 instances, 28-cores/instance, test by Intel on 6/02/2020.

Baseline- Intel Xeon processor E7 v4: 1-node, 4x Intel® Xeon® processor E7-8890 v4 (24C, 165W) on Intel Reference Platform (Brickland) with 512 GB (32 slots /16GB/ 1600) total memory, ucode 0xb000038, HT on, Turbo on, with Ubuntu 20.04 LTS, Linux 5.4.0-29-generic, Intel SSD 800GB OS Drive, Training: ResNet-50-v1.5,Inference: ResNet-50-v1.5 Throughput, https://github.com/Intel-tensorflow/tensorflow -b bf16/base, commit#6ef2116e6a09, Modelzoo: https://github.com/IntelAI/models/ -b v1.6.1, Imagenet dataset, oneDNN 1.4, FP32, global BS=1024, 4 instances, 24-cores/instance, test by Intel on 6/08/2020

2nd Gen: June 01, 2020

1st Gen: June 02, 2020

Baseline: June 08, 2020

New 3rd Gen Intel Xeon Scalable Processor (Cooper Lake): 1-node, 4x 3rd Gen Intel® Xeon® Platinum 8380H processor (pre-production 28C, 250W) on Intel Reference Platform (Cooper City) with 384 GB (24 slots / 16GB / 3200) total memory, ucode 0x700001b, HT on, Turbo on, with Ubuntu 20.04 LTS, Linux 5.4.0-29-generic, Intel SSD 800GB OS Drive, Inference: ResNet-50 v1.5 Throughput, https://github.com/Intel-tensorflow/tensorflow -b bf16/base, commit#6ef2116e6a09, Modelzoo: https://github.com/IntelAI/models/ -b v1.6.1, Imagenet dataset, oneDNN 1.4, FP32, INT8-VNNI, BF16, BS=128, 4 instances, 28-cores/instance, test by Intel on 06/01//2020.

2nd Gen Intel Xeon Scalable Processor (Cascade Lake): 1-node, 4x Intel® Xeon® Platinum 8280 processor (28C, 205W) on Intel Reference Platform (Lightning Ridge) with 768 GB (24 slots / 32 GB / 2933 ) total memory, ucode 0x4002f00, HT on, Turbo on, with Ubuntu 20.04 LTS, Linux 5.4.0-29-generic, Intel SSD 800GB OS Drive, Inference: ResNet-50 v1.5 Throughput, https://github.com/Intel-tensorflow/tensorflow -b bf16/base, commit# 6ef2116e6a09, Modelzoo: https://github.com/IntelAI/models/ -b v1.6.1, Imagenet dataset, oneDNN 1.4, FP32, INT8-VNNI, BS=128, 4 instances, 28-cores/instance, test by Intel on 06/01/2020.

Intel Xeon Scalable Processor (Skylake): 1-node, 4x Intel® Xeon® Platinum 8180 processor (28C, 205W) on Intel Reference Platform (Lightning Ridge) with 768 GB (24 slots / 32 GB / 2666 ) total memory, ucode 0x2000069, HT on, Turbo on, with Ubuntu 20.04 LTS, 5.4.0-26-generic, Intel SSD 800GB OS Drive, Inference: ResNet-50-v1.5 Throughput, https://github.com/Intel-tensorflow/tensorflow -b bf16/base, commit#6ef2116e6a09, Modelzoo: https://github.com/IntelAI/models/ -b v1.6.1, Imagenet dataset, oneDNN 1.4, FP32, INT8, BS=128, 4 instances, 28-cores/instance, test by Intel on 6/02/2020.

Baseline: Intel Xeon processor E7 v4 (Broadwell): 1-node, 4x Intel® Xeon® processor E7-8890 v4 (24C, 165W) on Intel Reference Platform (Brickland) with 512 GB (32 slots /16GB/ 1600) total memory, ucode 0xb000038, HT on, Turbo on, with Ubuntu 20.04 LTS, Linux 5.4.0-29-generic, Intel SSD 800GB OS Drive, Inference: ResNet-50-v1.5 Throughput, https://github.com/Intel-tensorflow/tensorflow -b bf16/base, commit#6ef2116e6a09, Modelzoo: https://github.com/IntelAI/models/ -b v1.6.1, Imagenet dataset, oneDNN 1.4, FP32, BS=128, 4 instances, 24-cores/instance, test by Intel on 6/08/2020.

2nd Gen: June 01, 2020

1st Gen: June 02, 2020

Baseline: June 08, 2020

Average performance based on Geomean of est SPECrate®2017_​int_​base 1-copy, est SPECrate®2017_​fp_​base 1-copy, est SPECrate®2017_​int_​base, est SPECrate®2017_​fp_​base, STREAM Triad, Intel distribution of LINPACK, Virtualization and OLTP Database workloads. Results have been estimated or simulated.

New: SPECcpu_​2017, STREAM, LINPACK Performance: 1-node, 4x 3rd Gen Intel® Xeon® Platinum 8380H processor (pre-production 28C, 250W) on Intel Reference Platform (Cooper City) with 768 GB (24 slots / 32 GB / 3200) total memory, microcode 0x87000016, HT on for SPECcpu, off for STREAM, LINPACK), Turbo on, with Ubuntu 19.10, 5.3.0-48-generic, 1x Intel 240GB SSD OS Drive, est SPECcpu_​2017, STREAM Triad, Intel distribution of LINPACK, test by Intel on 5/15/2020. HammerDB OLTP Database Performance: New: 1-node, 4x 3rd Gen Intel® Xeon® Platinum 8380H processor (pre-production 28C, 250W) on Intel Reference Platform (Cooper City) with 768 GB (24 slots / 32 GB / 3200) total memory, microcode 0x700001b, HT on, Turbo on, with Redhat 8.1, 4.18.0-147.3.1.el8_​1.x86_​64, 1x Intel 240GB SSD OS Drive, 2x6.4T P4610 for DATA, 2x3.2T P4610 for REDO, 1Gbps NIC, HammerDB 3.2, Popular Commercial Database, test by Intel on 5/13/2020. Virtualization Performance: New: 1-node, 4x 3rd Gen Intel® Xeon® Platinum 8380H processor (pre-production 28C, 250W) on Intel Reference Platform (Cooper City) with 1536 GB (48 slots / 32 GB / 3200 (@2933)) total memory, microcode 0x700001b, HT on, Turbo on, with RHEL-8.1 GA, 4.18.0-147.3.1.el8_​1.x86_​64, 1x Intel 240GB SSD OS Drive, 4x P4610 3.2TB PCIe NVMe, 4 x 40 GbE x710 dual port, Virtualization workload, test by Intel on 5/20/2020.

Baseline: SPECcpu_​2017, STREAM, LINPACK Performance: 1-node, 4x Intel® Xeon® processor E7-8890 v3 on Intel Reference Platform (Brickland) with 512 GB (32 slots / 16 GB / 2133 (@1600)) total memory, microcode 0x16, HT on for SPECcpu, off for STREAM, LINPACK), Turbo on, with Ubuntu 20.04 LTS, 5.4.0-29-generic, 1x Intel 480GB SSD OS Drive, est SPECcpu_​2017, STREAM Triad, Intel distribution of LINPACK, test by Intel on 5/15/2020. HammerDB OLTP Database Performance: 1-node, 4x Intel® Xeon® processor E7-8890 v3 on Intel Reference Platform (Brickland) with 1024 GB (64 slots / 16GB / 1600) total memory, microcode 0x16, HT on, Turbo on, with Redhat 8.1, 4.18.0-147.3.1.el8_​1.x86_​64, 1x Intel 800GB SSD OS Drive, 1x1.6T P3700 for DATA, 1x1.6T P3700 for REDO, 1Gbps NIC, HammerDB 3.2, Popular Commercial Database, test by Intel on 4/20/2020. Virtualization Performance: 1-node, 4x Intel® Xeon® processor E7-8890 v3 on Intel Reference Platform (Brickland) with 1024 GB (64 slots / 16GB / 1600) total memory, microcode 0x0000016, HT on, Turbo on, with RHEL-8.1 GA, 4.18.0-147.3.1.el8_​1.x86_​64, 1x Intel 240GB SSD OS Drive, 4x P3700 2TB PCIe NVMe, 4 x 40 GbE x710 dual port, Virtualization workload, test by Intel on 5/20/2020.

Baseline: May 20, 2020

Baseline: 1-node, 4x Intel® Xeon® processor E7-8890 v3 on Intel Reference Platform (Brickland) with 1024 GB (64 slots / 16GB / 1600) total memory, microcode 0x16, HT on, Turbo on, with Redhat 8.1, 4.18.0-147.3.1.el8_​1.x86_​64, 1x Intel 800GB SSD OS Drive, 1x1.6T P3700 for DATA, 1x1.6T P3700 for REDO, 1Gbps NIC, HammerDB 3.2, Popular Commercial Database, test by Intel on 4/20/2020.

Baseline: April 20, 2020

Baseline: 1-node, 4x Intel® Xeon® Platinum 8280 processor on Intel Reference Platform (Lightning Ridge) with 768 GB (24 slots / 32 GB / 2933 ) total memory, ucode 0x4002f00, HT on, Turbo on, with Ubuntu 20.04 LTS, Linux 5.4.0-26,28,29-generic, Intel 800GB SSD OS Drive, ResNet-50 v1.5 Throughput, https://github.com/Intel-tensorflow/tensorflow -b bf16/base, commit#828738642760358b388d8f615ded0c213f10c99a, Modelzoo: https://github.com/IntelAI/models/ -b v1.6.1, Imagenet dataset, oneDNN 1.4, FP32, BS=512, test by Intel on 5/18/2020.

Baseline: May 18, 2020

Baseline: May 18, 2020

Baseline: 1-node, 1x Intel® Xeon® 8280L 28C @ 2.7GHz processor on Neon City with Single PMem module config (6x32GB DRAM; 1x{128GB,256GB,512GB} Intel® Optane™ PMem 100 series module at 15W) ucode Rev: 04002F00 running Fedora 29 kernel 5.1.18-200.fc29.x86_​64, and MLC ver 3.8 with App-Direct. Source: 2020ww18_​CPX_​BPS_​DI. Tested by Intel, on 27 Apr 2020

Baseline: April 27, 2020

New: 1-node, 4x 3rd Gen Intel® Xeon® Platinum 8380H processor (pre-production 28C, 250W) on Intel Reference Platform (Cooper City) with 1536 GB (48 slots / 32 GB / 3200 (@2933)) total memory, microcode 0x700001b, HT on, Turbo on, with Ubuntu 18.04.4 LTS, 5.3.0-53-generic, 1x Intel 240GB SSD OS Drive, 4x P4610 3.2TB PCIe NVMe, 4 x 40 GbE x710 dual port, CloudXPRT vCP - Data Analytics, Kubernetes, Docker, Kafka, MinIO, Prometheus, XGBoost workload, Higgs dataset, test by Intel on 5/27/2020.

Baseline: 1-node, 4x Intel® Xeon® processor E7-8890 v3 on Intel Reference Platform (Brickland) with 1024 GB (64 slots / 16GB / 1600) total memory, microcode 0x0000016, HT on, Turbo on, with Ubuntu 18.04.4 LTS, 5.3.0-53-generic, 1x Intel 400GB SSD OS Drive, 4x P3700 2TB PCIe NVMe, 4 x 40 GbE x710 dual port, CloudXPRT vCP - Data Analytics, Kubernetes, Docker, Kafka, MinIO, Prometheus, XGBoost workload, Higgs dataset, test by Intel on 5/27/2020.

Intel contributes to the development of benchmarks by participating in, sponsoring, and/or contributing technical support to various benchmarking groups, including the BenchmarkXPRT Development Community administered by Principled Technologies.

Baseline: May 27, 2020

Baseline:1-node, 4x Intel® Xeon® processor E7-8890 v3 on Intel Reference Platform (Brickland) with 1024 GB (64 slots / 16GB / 1600) total memory, microcode 0x0000016, HT on, Turbo on, with RHEL-8.1 GA, 4.18.0-147.3.1.el8_​1.x86_​64, 1x Intel 240GB SSD OS Drive, 4x P3700 2TB PCIe NVMe, 4 x 40 GbE x710 dual port, Virtualization workload, test by Intel on 5/20/2020.

Baseline: May 20, 2020

33% better performance when using Intel ® SSD D7-P5500 series: Source - Intel. Comparing datasheet figures for 4KB Random Read QD256 performance between the Intel ® SSD D7-P5500 Series 7.68TB and Intel ® SSD DC P4510 Series 8TB with both drives running on PCIe 3.1. Measured performance was 854K IOPS and 641.8K IOPS for the D7-P5500 and DC P4510, respectively. Performance for both drives measured using FIO Linux CentOS 7.2 kernel 4.8.6 with 4KB (4,096 bytes) of transfer size with Queue Depth 64 (4 workers). Measurements are performed on a full Logical Block Address (LBA) span of the drive once the workload has reached steady state but including all background activities required for normal operation and data reliability. Power mode set at PM0. Any differences in your system hardware, software or configuration may affect your actual performance. Intel expects to see certain level of variation in data measurement across multiple drives.

40% lower latency when using Intel ® SSD D7-P5500 series: Source - Intel. Comparing datasheet figures for 4KB Random Write QD1 latency between the Intel ® SSD D7-P5500 Series 7.68TB and Intel® SSD DC P4510 Series 8TB with both drives running on PCIe 3.1. Measured latency was 15μs and 25μs for the D7-P5500 and DC P4510, respectively. Performance for both drives measured using FIO Linux CentOS 7.2 kernel 4.8.6 with 4KB (4096 bytes) of transfer size with Queue Depth 1 (1 worker). Measurements are performed on a full Logical Block Address (LBA) span of the drive once the workload has reached steady state but including all background activities required for normal operation and data reliability. Power mode set at PM0. Any differences in your system hardware, software or configuration may affect your actual performance. Intel expects to see certain level of variation in data measurement across multiple drives.

Baseline: 1-node, 4x Intel® Xeon® Platinum 8280 processor on Intel Reference Platform (Lightning Ridge) with 768 GB (24 slots / 32 GB / 2933 ) total memory, ucode 0x4002f00, HT on, Turbo on, with Ubuntu 20.04 LTS, Linux 5.4.0-26,28,29-generic, Intel 800GB SSD OS Drive, ResNet-50 v1.5 Throughput, https://github.com/Intel-tensorflow/tensorflow -b bf16/base, commit#828738642760358b388d8f615ded0c213f10c99a, Modelzoo: https://github.com/IntelAI/models/ -b v1.6.1, Imagenet dataset, oneDNN 1.4, FP32, BS=56, 4 instances, 28-cores/instance, test by Intel on 5/18/2020.

Baseline: May 18, 2020

Baseline: 1-node, 4x Intel® Xeon® Platinum 8280 processor on Intel Reference Platform (Lightning Ridge) with 768 GB (24 slots / 32 GB / 2933 ) total memory, ucode 0x4002f00, HT on, Turbo on, with Ubuntu 20.04 LTS, Linux 5.4.0-26,28,29-generic, Intel 800GB SSD OS Drive, BERT-Large (QA) Throughput, https://github.com/Intel-tensorflow/tensorflow -b bf16/base, commit#828738642760358b388d8f615ded0c213f10c99a, Modelzoo: https://github.com/IntelAI/models/ -b v1.6.1, Squad 1.1 dataset, oneDNN 1.4, FP32, BS=12, test by Intel on 5/18/2020.

Baseline: May 18, 2020