AMD EPYC Turin 8c vs. 12c Memory Channel DDR5 Comparison

AMD EPYC 9655P memory benchmarks by Michael Larabel for a future article.

HTML result view exported from: https://openbenchmarking.org/result/2411164-NE-AMDEPYCTU56&grs&sro.

Stream

Type: Copy

OpenVINO

Model: Person Detection FP16 - Device: CPU

Stream

Type: Scale

OpenVINO

Model: Person Detection FP16 - Device: CPU

Stream

Type: Add

Stream

Type: Triad

High Performance Conjugate Gradient

X Y Z: 104 104 104 - RT: 60

High Performance Conjugate Gradient

X Y Z: 144 144 144 - RT: 60

Algebraic Multi-Grid Benchmark

Xcompact3d Incompact3d

Input: X3D-benchmarking input.i3d

NAS Parallel Benchmarks

Test / Class: SP.C

OpenRadioss

Model: Chrysler Neon 1M

NAS Parallel Benchmarks

Test / Class: MG.C

NAS Parallel Benchmarks

Test / Class: FT.C

OpenFOAM

Input: drivaerFastback, Medium Mesh Size - Execution Time

Xcompact3d Incompact3d

Input: input.i3d 193 Cells Per Direction

NAS Parallel Benchmarks

Test / Class: BT.C

MBW

Test: Memory Copy, Fixed Block Size - Array Size: 8192 MiB

NAS Parallel Benchmarks

Test / Class: IS.D

TensorFlow

Device: CPU - Batch Size: 512 - Model: ResNet-50

MBW

Test: Memory Copy, Fixed Block Size - Array Size: 4096 MiB

RAMspeed SMP

Type: Add - Benchmark: Integer

RAMspeed SMP

Type: Copy - Benchmark: Integer

NAMD

Input: STMV with 1,066,628 Atoms

RAMspeed SMP

Type: Scale - Benchmark: Integer

ClickHouse

100M Rows Hits Dataset, Second Run

RAMspeed SMP

Type: Average - Benchmark: Integer

NAS Parallel Benchmarks

Test / Class: SP.B

ClickHouse

100M Rows Hits Dataset, Third Run

RAMspeed SMP

Type: Triad - Benchmark: Integer

ClickHouse

100M Rows Hits Dataset, First Run / Cold Cache

OpenVINO

Model: Machine Translation EN To DE FP16 - Device: CPU

OpenVINO

Model: Machine Translation EN To DE FP16 - Device: CPU

NAS Parallel Benchmarks

Test / Class: LU.C

Graph500

Scale: 26

GPAW

Input: Carbon Nanotube

Graph500

Scale: 26

OpenVINO

Model: Noise Suppression Poconet-Like FP16 - Device: CPU

Graph500

Scale: 26

Graph500

Scale: 26

NAS Parallel Benchmarks

Test / Class: CG.C

MBW

Test: Memory Copy - Array Size: 8192 MiB

OpenVINO

Model: Noise Suppression Poconet-Like FP16 - Device: CPU

MBW

Test: Memory Copy - Array Size: 4096 MiB

7-Zip Compression

Test: Compression Rating

GROMACS

Implementation: MPI CPU - Input: water_GMX50_bare

ONNX Runtime

Model: ZFNet-512 - Device: CPU - Executor: Standard

Apache HTTP Server

Concurrent Requests: 500

Whisperfile

Model Size: Medium

OpenVINO

Model: Road Segmentation ADAS FP16-INT8 - Device: CPU

OpenVINO

Model: Road Segmentation ADAS FP16-INT8 - Device: CPU

Whisperfile

Model Size: Small

Apache Cassandra

Test: Writes

NAMD

Input: ATPase with 327,506 Atoms

Epoch

Epoch3D Deck: Cone

ONNX Runtime

Model: GPT-2 - Device: CPU - Executor: Standard

LuxCoreRender

Scene: LuxCore Benchmark - Acceleration: CPU

XNNPACK

Model: FP32MobileNetV3Large

nginx

Connections: 500

ASTC Encoder

Preset: Fast

LiteRT

Model: Inception ResNet V2

ONNX Runtime

Model: T5 Encoder - Device: CPU - Executor: Standard

LiteRT

Model: DeepLab V3

LuxCoreRender

Scene: DLSC - Acceleration: CPU

OpenFOAM

Input: drivaerFastback, Small Mesh Size - Mesh Time

ONNX Runtime

Model: fcn-resnet101-11 - Device: CPU - Executor: Standard

OpenFOAM

Input: drivaerFastback, Medium Mesh Size - Mesh Time

uvg266

Video Input: Bosphorus 4K - Video Preset: Super Fast

XNNPACK

Model: FP16MobileNetV3Small

SPECFEM3D

Model: Tomographic Model

OpenVKL

Benchmark: vklBenchmarkCPU ISPC

SVT-AV1

Encoder Mode: Preset 8 - Input: Bosphorus 4K

Kvazaar

Video Input: Bosphorus 4K - Video Preset: Super Fast

Liquid-DSP

Threads: 128 - Buffer Length: 256 - Filter Length: 57

ONNX Runtime

Model: bertsquad-12 - Device: CPU - Executor: Standard

uvg266

Video Input: Bosphorus 4K - Video Preset: Very Fast

Timed Linux Kernel Compilation

Build: allmodconfig

Timed Linux Kernel Compilation

Build: defconfig

VVenC

Video Input: Bosphorus 4K - Video Preset: Faster

LiteRT

Model: Quantized COCO SSD MobileNet v1

PostgreSQL

Scaling Factor: 1000 - Clients: 800 - Mode: Read Write - Average Latency

XNNPACK

Model: FP16MobileNetV3Large

PostgreSQL

Scaling Factor: 1000 - Clients: 800 - Mode: Read Write

LuxCoreRender

Scene: Rainbow Colors and Prism - Acceleration: CPU

SPECFEM3D

Model: Mount St. Helens

NAS Parallel Benchmarks

Test / Class: EP.C

ONNX Runtime

Model: yolov4 - Device: CPU - Executor: Standard

XNNPACK

Model: QS8MobileNetV2

John The Ripper

Test: MD5

Kvazaar

Video Input: Bosphorus 4K - Video Preset: Very Fast

XNNPACK

Model: FP16MobileNetV2

Memcached

Set To Get Ratio: 1:10

uvg266

Video Input: Bosphorus 4K - Video Preset: Ultra Fast

SPECFEM3D

Model: Layered Halfspace

oneDNN

Harness: Deconvolution Batch shapes_1d - Engine: CPU

LiteRT

Model: NASNet Mobile

XNNPACK

Model: FP32MobileNetV3Small

WarpX

Input: Uniform Plasma

OpenFOAM

Input: drivaerFastback, Small Mesh Size - Execution Time

SVT-AV1

Encoder Mode: Preset 13 - Input: Bosphorus 4K

RawTherapee

Total Benchmark Time

srsRAN Project

Test: PDSCH Processor Benchmark, Throughput Total

LiteRT

Model: Mobilenet Float

Stargate Digital Audio Workstation

Sample Rate: 192000 - Buffer Size: 1024

Stargate Digital Audio Workstation

Sample Rate: 192000 - Buffer Size: 512

oneDNN

Harness: Recurrent Neural Network Training - Engine: CPU

Memcached

Set To Get Ratio: 1:100

XNNPACK

Model: FP16MobileNetV1

Chaos Group V-RAY

Mode: CPU

OpenVINO

Model: Person Vehicle Bike Detection FP16 - Device: CPU

XNNPACK

Model: FP32MobileNetV1

OpenRadioss

Model: INIVOL and Fluid Structure Interaction Drop Container

Kvazaar

Video Input: Bosphorus 4K - Video Preset: Ultra Fast

OpenVINO

Model: Person Vehicle Bike Detection FP16 - Device: CPU

oneDNN

Harness: Convolution Batch Shapes Auto - Engine: CPU

SVT-AV1

Encoder Mode: Preset 5 - Input: Bosphorus 4K

Liquid-DSP

Threads: 192 - Buffer Length: 256 - Filter Length: 512

Liquid-DSP

Threads: 64 - Buffer Length: 256 - Filter Length: 57

SPECFEM3D

Model: Water-layered Halfspace

oneDNN

Harness: IP Shapes 1D - Engine: CPU

miniBUDE

Implementation: OpenMP - Input Deck: BM2

miniBUDE

Implementation: OpenMP - Input Deck: BM2

SVT-AV1

Encoder Mode: Preset 3 - Input: Bosphorus 4K

Blender

Blend File: Fishy Cat - Compute: CPU-Only

VVenC

Video Input: Bosphorus 4K - Video Preset: Fast

oneDNN

Harness: IP Shapes 3D - Engine: CPU

Stargate Digital Audio Workstation

Sample Rate: 96000 - Buffer Size: 512

Timed Godot Game Engine Compilation

Time To Compile

John The Ripper

Test: WPA PSK

Blender

Blend File: BMW27 - Compute: CPU-Only

OpenVINO

Model: Age Gender Recognition Retail 0013 FP16-INT8 - Device: CPU

uvg266

Video Input: Bosphorus 4K - Video Preset: Medium

WarpX

Input: Plasma Acceleration

Stargate Digital Audio Workstation

Sample Rate: 96000 - Buffer Size: 1024

Liquid-DSP

Threads: 64 - Buffer Length: 256 - Filter Length: 512

uvg266

Video Input: Bosphorus 4K - Video Preset: Slow

SVT-AV1

Encoder Mode: Preset 3 - Input: Beauty 4K 10-bit

SVT-AV1

Encoder Mode: Preset 8 - Input: Beauty 4K 10-bit

Blender

Blend File: Junkshop - Compute: CPU-Only

LiteRT

Model: Inception V4

Memcached

Set To Get Ratio: 1:5

Blender

Blend File: Barbershop - Compute: CPU-Only

OpenVINO

Model: Face Detection Retail FP16-INT8 - Device: CPU

LuxCoreRender

Scene: Danish Mood - Acceleration: CPU

Blender

Blend File: Classroom - Compute: CPU-Only

Liquid-DSP

Threads: 128 - Buffer Length: 256 - Filter Length: 512

OpenVINO

Model: Handwritten English Recognition FP16-INT8 - Device: CPU

Build2

Time To Compile

m-queens

Time To Solve