SMT Proves Worthwhile Option For 128-Core AMD EPYC "Bergamo" CPUs

Written by Michael Larabel in Processors on 20 July 2023 at 10:57 AM EDT. Page 7 of 8. 22 Comments.
ASTC Encoder benchmark with settings of Preset: Fast. EPYC 9754 1P: SMT Off was the fastest.
ASTC Encoder benchmark with settings of Preset: Thorough. EPYC 9754 2P: SMT On was the fastest.
ASTC Encoder benchmark with settings of Preset: Exhaustive. EPYC 9754 2P: SMT On was the fastest.

ASTC texture encoding performance was mixed based upon settings whether SMT paid off.

Graph500 benchmark with settings of Scale: 26. EPYC 9754 2P: SMT Off was the fastest.
Graph500 benchmark with settings of Scale: 26. EPYC 9754 2P: SMT Off was the fastest.
Graph500 benchmark with settings of Scale: 26. EPYC 9754 2P: SMT Off was the fastest.

Graph500 was faster with SMT disabled.

MariaDB benchmark with settings of Clients: 4096. EPYC 9754 1P: SMT Off was the fastest.

The MariaDB MySQL database server was slightly faster with SMT disabled.

TensorFlow benchmark with settings of Device: CPU, Batch Size: 512, Model: AlexNet. EPYC 9754 2P: SMT Off was the fastest.
TensorFlow benchmark with settings of Device: CPU, Batch Size: 256, Model: GoogLeNet. EPYC 9754 1P: SMT Off was the fastest.
TensorFlow benchmark with settings of Device: CPU, Batch Size: 512, Model: GoogLeNet. EPYC 9754 2P: SMT Off was the fastest.
TensorFlow benchmark with settings of Device: CPU, Batch Size: 512, Model: ResNet-50. EPYC 9754 2P: SMT Off was the fastest.

TensorFlow performance tended to be similar at 1P but when running in a dual socket configuration is where having SMT off tended to produce better results.

Neural Magic DeepSparse benchmark with settings of Model: NLP Document Classification, oBERT base uncased on IMDB, Scenario: Asynchronous Multi-Stream. EPYC 9754 2P: SMT Off was the fastest.
Neural Magic DeepSparse benchmark with settings of Model: NLP Sentiment Analysis, 80% Pruned Quantized BERT Base Uncased, Scenario: Asynchronous Multi-Stream. EPYC 9754 2P: SMT Off was the fastest.
Neural Magic DeepSparse benchmark with settings of Model: CV Detection, YOLOv5s COCO, Scenario: Asynchronous Multi-Stream. EPYC 9754 2P: SMT Off was the fastest.
Neural Magic DeepSparse benchmark with settings of Model: NLP Text Classification, DistilBERT mnli, Scenario: Asynchronous Multi-Stream. EPYC 9754 2P: SMT Off was the fastest.
Neural Magic DeepSparse benchmark with settings of Model: NLP Token Classification, BERT base uncased conll2003, Scenario: Asynchronous Multi-Stream. EPYC 9754 2P: SMT Off was the fastest.

Neural Magic's DeepSparse software benefited from disabling SMT on the Bergamo server configurations tested.


Related Articles
Ampere Computing 2024 Roadmap Update: 256 Core 3nm CPU In 2025
Benchmarking The First RISC-V Cloud Server: Scaleway EM-RV1 Performance
AMD Announces Ryzen PRO 8840 & PRO 8000G Series CPUs
Intel Announces Gaudi 3 AI Accelerator, Intel Xeon 6 Brand
AMD Zen 4 vs. Zen 4C Performance, Zen 4C Core Scaling With Ryzen 5 8500G
Intel Core i3 14100 / i5 14500 vs. AMD Ryzen 5 8500G / 8600G In 500+ Benchmarks