• 615781
  • 1
  • Public Content

3rd Generation Intel® Xeon® Scalable Processors

Software and workloads used in performance tests may have been optimized for performance only on Intel® microprocessors. Performance tests, such as SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations, and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products.

Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available updates. See configuration disclosure for details. No product or component can be absolutely secure.

Refer to https://software.intel.com/articles/optimization-notice for more information regarding performance and optimization choices in Intel software products.

Estimates of SPECrate®2017_​int_​base and SPECrate®2017_​fp_​base based on Intel internal measurements. SPEC® , SPECrate® and SPEC CPU® are registered trademarks of the Standard Performance Evaluation Corporation. See www.spec.org for more information.

Claim Processor Family System Configuration Measurement Measurement Period
[14] Up to 39% More Bandwidth with Intel® Optane™ PMem 200 series 512GB module vs. Intel® Optane™ PMem 100 series 512GB module 3rd Generation Intel® Xeon® Platinum processor & Intel® Optane™ persistent memory 200 series New: 1-node, 1x pre-production CPX6 Processor @ 2.9GHz on Intel - Cedar Island Customer Reference Board (CRB) with DRAM: (per socket) 6 slots / 32GB / 2666 MT/s, PMem: (per socket) 1x 512GB Intel® Optane™ PMem 200 series module at 15W (192GB DRAM, 512GB PMem) total memory, ucode WW12'20 (pre-production) , running Fedora 30 kernel 5.1.18-200.fc29.x86_​65,using MLC v3.8 with App-Direct. Source: 2020ww18_​CPX_​BPS_​BG, test by Intel on 31 Mar 2020. Baseline: 1-node, 1x Intel® Xeon® Platinum 8280L processor @ 2.7GHz on Intel - Purley Customer Reference Board (CRB) with DRAM: (per socket) 6 slots / 32GB / 2666 MT/s, PMem: (per socket) 1x 512GB Intel® Optane™ PMem 100 series module at 15W (192GB DRAM, 512GB PMem) total memory, ucode 0x04002F00, running Fedora 29 kernel 5.1.18-200.fc29.x86_​64,using MLC v3.8 with App-Direct workload. Source: 2020ww22_​CPX_​BPS_​BG, test by Intel on 27 Apr 2020. MLC ver3.8 with App Direct (from Intel Optane PMem 200 series demo) New: March 31, 2020

Baseline: April 27, 2020

[13] Multi-generation ResNet-50 Training Throughput Performance Improvement with Intel DL Boost supporting INT8 and BF16 3rd Generation Intel® Xeon® Platinum processor

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

Training: ResNet-50 v1.5 Throughput New: Jun 01, 2020

2nd Gen: Jun 01, 2020

1st Gen: Jun 02, 2020

Baseline: Jun 08, 2020

[12] Multi-generation ResNet-50 Inference Throughput Performance Improvement with Intel DL Boost supporting INT8 and BF16 3rd Generation Intel® Xeon® Platinum processor

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.

Inference: ResNet-50 v1.5 Throughput New: Jun 01, 2020

2nd Gen: Jun 01, 2020

1st Gen: Jun 02, 2020

Baseline: Jun 08, 2020

[11] 1.9x average performance gain on popular workloads with the new 3rd Gen Intel® Xeon® Platinum 8380H processor vs. 5-year old platform 3rd Generation Intel® Xeon® Platinum processor

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.

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 New: May 20, 2020

Baseline: May 20, 2020

[10] Process up to 1.98x more OLTP database transactions per minute with the new 3rd Gen Intel® Xeon® Scalable platform vs. 5-year old 4-socket platform 3rd Generation Intel® Xeon® Platinum processor 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.

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.

HammerDB OLTP Database New: April 20, 2020

Baseline: April 20, 2020

[9] Up to 1.93x higher AI training performance with 3rd Gen Intel® Xeon® Scalable processor supporting Intel® DL Boost with BF16 vs. prior generation on ResNet50 throughput for image classification 3rd Generation Intel® Xeon® Platinum processor New: 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 5.4.0-26,28,29-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, BF16, BS=512, test by Intel on 5/18/2020.

Baseline: 1-node, 4x Intel® Xeon® Platinum 8280 processor on Intel Reference Platform (Lightening 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.

ResNet-50 v1.5 Image Classification Training Throughput New: May 18, 2020

Baseline: May 18, 2020

[8] Up to 1.9x higher AI inference performance with 3rd Gen Intel® Xeon® Scalable processor supporting Intel® DL Boost with BF16 vs. prior generation with FP32 on BERT throughput for natural language processing 3rd Generation Intel® Xeon® Platinum processor New: 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 5.4.0-26,28,29-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, BF16, BS=32, 4 instances, 28-cores/instance, test by Intel on 5/18/2020. Baseline: 1-node, 4x Intel® Xeon® Platinum 8280 processor on Intel Reference Platform (Lightening 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=32, 4 instances, 28-cores/instance, test by Intel on 5/18/2020. BERT-Large (QA) Squad Inference Throughput New: May 18, 2020

Baseline: May 18, 2020

[7] 225X faster access to data with Intel® Optane™ persistent memory 200 series vs. NVMe SSD Intel® Optane™ persistent memory 200 series New: Intel® Optane™ persistent memory idle read latency compared to Baseline: Intel® SSD DC P4610 Series TLC NAND solid state drive idle read latency. Memory idle read latency NA
[6] Average of 25% higher memory bandwidth vs. prior gen 3rd Generation Intel® Xeon® Platinum processor & Intel® Optane™ persistent memory 200 series New: 1-node, 1x Intel® Xeon® pre-production CPX6 28C @ 2.9GHz processor on Cooper City with Single PMem module config (6x32GB DRAM; 1x{128GB,256GB,512GB} Intel® Optane™ PMem 200 series module at 15W), ucode pre-production running Fedora 29 kernel 5.1.18-200.fc29.x86_​64, and MLC ver 3.8 with App-Direct. Source: 2020ww18_​CPX_​BPS_​BG. Tested by Intel, on 31 Mar 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

MLC ver 3.8 with App-Direct New: March 31, 2020

Baseline: April 27, 2020

[5] Up to 1.92x higher performance on cloud data analytics usage models with the new 3rd Gen Intel® Xeon® Scalable processor vs. 5-year old 4-socket platform 3rd Generation Intel® Xeon® Platinum processor

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.

CloudXPRT vCP- Data Analytics, Kubernetes, Docker, Kafka, MinIO, Prometheus, XGBoost workload, Higgs dataset New: May 27, 2020

Baseline: May 27, 2020

[4] Up to 2.2x more Virtual Machines with the new 3rd Gen Intel® Xeon® Scalable platform and Intel® SSD Data Center Family vs. 5-year old 4-socket platform 3rd Generation Intel® Xeon® Platinum processor 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: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.

Virtualization workload New: May 20, 2020

Baseline: May 20, 2020

[3] The new Intel® 3D NAND SSDs deliver an improved balance of performance and capacity for your storage requirements-including up to 33% better performance and 40% lower latency. Intel® SSD D7-P5500 series

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.

Read/Write IOPS, Latency N/A
[2] Up to 1.87x higher AI Inference performance with 3rd Gen Intel® Xeon® Scalable processor supporting Intel® DL Boost with BF16 vs. prior generation using FP32 on ResNet50 throughput for image classification 3rd Generation Intel® Xeon® Platinum processor New: 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 5.4.0-26,28,29-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, BF16, BS=56, 4 instances, 28-cores/instance, test by Intel on 5/18/2020.

Baseline: 1-node, 4x Intel® Xeon® Platinum 8280 processor on Intel Reference Platform (Lightening 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.

ResNet-50 Inference Throughput New: May 18, 2020

Baseline: May 18, 2020

[1] Up to 1.7x more AI training performance with 3rd Gen Intel® Xeon® Scalable processor supporting Intel® DL Boost with BF16 vs. prior generation on BERT throughput for natural language processing 3rd Generation Intel® Xeon® Platinum processor New: 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 5.4.0-26,28,29-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, BF16, BS=12, test by Intel on 5/18/2020.

Baseline: 1-node, 4x Intel® Xeon® Platinum 8280 processor on Intel Reference Platform (Lightening 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.

BERT-Large (QA) Squad Training Throughput New: May 18, 2020

Baseline: May 18, 2020