The Compelling AVX-512 Performance Advantage On AMD EPYC 9005 "Turin"

Written by Michael Larabel in Processors on 31 January 2025 at 10:56 AM EST. Page 5 of 5. 5 Comments.

👁 OpenVINO benchmark with settings of Model: Noise Suppression Poconet-Like FP16, Device: CPU. EPYC Turin: AVX-512 Enabled was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Prompt Processing 512. EPYC Turin: AVX-512 Enabled was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Prompt Processing 2048. EPYC Turin: AVX-512 Enabled was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Prompt Processing 512. EPYC Turin: AVX-512 Enabled was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Prompt Processing 512. EPYC Turin: AVX-512 Enabled was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Prompt Processing 1024. EPYC Turin: AVX-512 Enabled was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Prompt Processing 2048. EPYC Turin: AVX-512 Enabled was the fastest.

👁 ONNX Runtime benchmark with settings of Model: fcn-resnet101-11, Device: CPU, Executor: Standard. EPYC Turin: AVX-512 Enabled was the fastest.

👁 ONNX Runtime benchmark with settings of Model: bertsquad-12, Device: CPU, Executor: Standard. EPYC Turin: AVX-512 Enabled was the fastest.

👁 ONNX Runtime benchmark with settings of Model: Faster R-CNN R-50-FPN-int8, Device: CPU, Executor: Standard. EPYC Turin: AVX-512 Enabled was the fastest.

👁 ONNX Runtime benchmark with settings of Model: ResNet101_DUC_HDC-12, Device: CPU, Executor: Standard. EPYC Turin: AVX-512 Enabled was the fastest.

Overall these results aren't too different from the original AVX-512 testing of EPYC Turin back in October. Check that out though if wanting to see the 256-bit vs. 512-bit data path difference with 5th Gen EPYC. Today's testing though jives with the former results that were done using a different CPU (EPYC 9755) on AMD's reference server platform. Here with this Supermicro 1P EPYC server we see the same great uplift and with impact to power/thermals. Plus this round of testing being done with a newer Linux software stack, GCC 14.2, and various new/updated benchmarks.

For those wanting to see more of this fresh AVX-512 benchmark data, see this OB result page for all of the data I collected in full.

👁 Geometric Mean Of All Test Results benchmark with settings of Result Composite, AMD EPYC Turin 2025 New AVX-512 Benchmarks. EPYC Turin: AVX-512 Enabled was the fastest.

When taking the geometric mean of all the AVX-512 workloads tested on this EPYC 9655(P) Supermicro server, AVX-512 yielded 1.57x the performance of the same hardware/software but with AVX-512 forced off.

👁 CPU Power Consumption Monitoring Overview benchmark with settings of Accumulated CPU Power Consumption Monitoring.

On average with AVX-512 enabled the EPYC 9655 96-core processor had a 259 Watt power draw with 384 peak compared to a 244 Watt average with a 389 Watt peak when AVX-512 was disabled within the BIOS. There overall tended to be fairly similar power results and nothing like the extremes observed during the early Intel AVX-512 generations.

👁 CPU Peak Freq Monitoring Overview benchmark with settings of Accumulated CPU Peak Freq Monitoring.

When AVX-512 was disabled the CPU peak frequency tended to be slightly higher with a 3.3GHz vs. 3.5GHz average but again for overall performance having AVX-512 enabled was of great benefit and providing the best performance by far.

👁 CPU Temp Monitoring Overview benchmark with settings of Accumulated CPU Temp Monitoring.

With little difference to the EPYC Turin CPU power consumption, the CPU core temperature didn't see any dramatic change: 52.2 degrees on average with AVX-512 enabled (default) or 50.7 degrees with AVX-512 disabled. This EPYC server build was enjoying SilverStone XE360-SP5 liquid cooling.

Those are the fresh AVX-512 AMD EPYC benchmarks to share from GCC 14.2 on Ubuntu 24.10 for this Supermicro server build.

5 Comments

If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal or Stripe tips are also graciously accepted. Thanks for your support.

Page: 1 2 3 4 5

URL: https://www.phoronix.com/review/amd-epyc-turin-avx512/5

⇱ The Compelling AVX-512 Performance Advantage On AMD EPYC 9005 "Turin" - Phoronix

The Compelling AVX-512 Performance Advantage On AMD EPYC 9005 "Turin"