Skip to content

Overview

Note: The Workload Services Framework is a benchmarking framework and is not intended to be used for the deployment of workloads in production environments. It is recommended that users consider any adjustments which may be necessary for the deployment of these workloads in a production environment including those necessary for implementing software best practices for workload scalability and security.

Introduction

The STREAM benchmark is a simple synthetic benchmark program that measures sustainable memory bandwidth (in MB/s) and the corresponding computation rate for simple vector kernels.

Test Case

This workload provides several tests cases with the following configuration parameters to measure the sustained memory bandwidth.

  • instruction_set: Specify the instruction set: sse, avx2 or avx3, default is sse.
  • ntimes: Specify the number of iterations of each kernel (Default:100, Min=2).
  • NO_OF_STREAM_ITERATIONS: Specify the NO_OF_STREAM_ITERATIONS to run the workload internally in a loop and it will pick the maximum value among them, default is 1
  • THP_ENABLE: Specify the THP_ENABLE: always or never, default is never,This will enable the transparanet huge pages if set to always else will disable it

Use the following commands to show the list of test cases:

cd build
cmake ..
cd workload/Stream
./ctest.sh -N

 **privileged mode ***
- the privileged mode is to run the workload with numactl with root access .if you wish to use it you can set the flag to true ,when executing workload through docker. In kubernetes it is set to true.

```bash
   ./ctest.sh -R <you_test_case> --set ENABLE_PRIVILEGED_MODE=true -V

Test cases:

Intel specific platform testcases with ubuntu22.04 as base image and ubuntu 24.04 as base image with two oneapi kit,one is 2022 with icc compiler and other is 2024 with icx as default compiler

icc compiler with ubuntu22.04 testcase 

Test #1: test_stream_icc_sse
Test #2: test_stream_icc_avx2
Test #3: test_stream_icc_avx3
Test #4: test_stream_icc_sse_gated
icc compiler with ubuntu24.04 testcase 
Test #1: test_static_stream_icc_ubuntu24_sse
Test #2: test_static_stream_icc_ubuntu24_avx2
Test #3: test_static_stream_icc_ubuntu24_avx3
Test #4: test_static_stream_icc_ubuntu24_pkm
Test #5: test_static_stream_icc_ubuntu24_avx512_icpc
icx compiler with ubuntu22.04 testcase 
Test #1: test_static_stream_icx_ubuntu22_sse
Test #2: test_static_stream_icx_ubuntu22_avx2
Test #3: test_static_stream_icx_ubuntu22_avx3
Test #4: test_static_stream_icx_ubuntu22_pkm
Test #5: test_static_stream_icx_ubuntu22_avx512_icpc
Test #6: test_static_stream_icx_ubuntu22_sse_gated
icx compiler with ubuntu24.04 testcase 
Test #1: test_static_stream_icx_ubuntu24_sse
Test #2: test_static_stream_icx_ubuntu24_avx2
Test #3: test_static_stream_icx_ubuntu24_avx3
Test #4: test_static_stream_icx_ubuntu24_pkm
Test #5: test_static_stream_icx_ubuntu24_avx512_icpc
Test #6: test_static_stream_icx_ubuntu24_sse_gated

AMD specific platform(ROME & MILAN) testcases

Test #1: test_stream_amd_sse
Test #2: test_stream_amd_avx2
Test #3: test_stream_amd_sse_gated
Test #4: test_stream_amd_pkm

AMD specific platform(GENOA) testcases

Test #1: test_stream_amd_sse
Test #2: test_stream_amd_avx2
Test #3: test_stream_amd_avx3
Test #4: test_stream_amd_sse_gated
Test #5: test_stream_amd_pkm

Docker Image

The workload provides 4 docker images: stream, stream-amd, stream-amd-zen4 and stream-arm.

Run the workload as follows:

stream_icc is built on top of ubuntu22.04 and stream_icc_ubuntu24 is built on top of ubuntu24.04 - stream_icc stream_icc_ubuntu24 is for Intel specific platform built with icc compiler

stream_icx_ubuntu22 is built on top of ubuntu22.04 and stream_icx_ubuntu24 is built on top of ubuntu24.04 - stream_icx_ubuntu22 stream_icx_ubuntu24 is for Intel specific platform built with icx compiler - stream-amd is for AMD and built with AOCC4 compiler - stream-amd-zen4 is for AMD Zen4 platform using AMD's prebuilt binaries - stream-arm is for ARM platforms, compiled with gcc12

mkdir -p logs
id=$(docker run --rm --detach -e INSTRUCTION_SET=sse -e NTIMES=100 stream)

# or
id=$(docker run --rm --detach -e INSTRUCTION_SET=sse -e NTIMES=100 stream-amd)
docker exec $id cat /export-logs | tar xf - -C logs
docker rm -f $id

# or
id=$(docker run --rm --detach -e INSTRUCTION_SET=sse -e NTIMES=100 stream-arm)
docker exec $id cat /export-logs | tar xf - -C logs
docker rm -f $id

KPI

Run the kpi.sh script to parse the KPIs from the validation logs.

KPI output example:

Copy Best Rate (MB/s): 77748.9
Copy Avg time (s): 0.098781
Copy Min time (s): 0.098399
Copy Max time (s): 0.101331
Scale Best Rate (MB/s): 78363.0
Scale Avg time (s): 0.098296
Scale Min time (s): 0.097628
Scale Max time (s): 0.100837
Add Best Rate (MB/s): 78671.7
Add Avg time (s): 0.146458
Add Min time (s): 0.145867
Add Max time (s): 0.148147
*Triad Best Rate (MB/s): 78888.7
Triad Avg time (s): 0.146033
Triad Min time (s): 0.145466
Triad Max time (s): 0.148240

Index Info

  • Name: Stream
  • Category: Synthetic
  • Platform: SPR, ICX, EMR,MILAN,GENOA,ROME,SKL,BERGAMO,CLX,ARMv8,ARMv9,ARMv10
  • Keywords:
  • Permission: