The Performance Impact From Different Arch Linux Kernel Flavors

Written by Michael Larabel in Operating Systems on 25 January 2023 at 12:00 PM EST. Page 5 of 8. 51 Comments.
TensorFlow benchmark with settings of Device: CPU, Batch Size: 16, Model: ResNet-50. 6.1.7-hardened1-1-hardened was the fastest.
TensorFlow benchmark with settings of Device: CPU, Batch Size: 32, Model: ResNet-50. 6.1.7-hardened1-1-hardened was the fastest.

Like with the video encode benchmarks, making use of the Linux 5.15 LTS kernel on Arch Linux led to slower speeds when it came to AI performance with TensorFlow.

Neural Magic DeepSparse benchmark with settings of Model: NLP Sentiment Analysis, 80% Pruned Quantized BERT Base Uncased, Scenario: Asynchronous Multi-Stream. 6.1.7-arch1-1 was the fastest.
Neural Magic DeepSparse benchmark with settings of Model: NLP Sentiment Analysis, 80% Pruned Quantized BERT Base Uncased, Scenario: Asynchronous Multi-Stream. 6.1.7-arch1-1 was the fastest.
Neural Magic DeepSparse benchmark with settings of Model: NLP Question Answering, BERT base uncased SQuaD 12layer Pruned90, Scenario: Asynchronous Multi-Stream. 6.1.7-arch1-1 was the fastest.
Neural Magic DeepSparse benchmark with settings of Model: NLP Question Answering, BERT base uncased SQuaD 12layer Pruned90, Scenario: Asynchronous Multi-Stream. 6.1.7-arch1-1 was the fastest.
Neural Magic DeepSparse benchmark with settings of Model: CV Detection, YOLOv5s COCO, Scenario: Asynchronous Multi-Stream. 6.1.7-hardened1-1-hardened was the fastest.
Neural Magic DeepSparse benchmark with settings of Model: CV Detection, YOLOv5s COCO, Scenario: Asynchronous Multi-Stream. 6.0.5.14.realtime1-3-rt was the fastest.

Neural Magic's DeepSparse was performing the slowest as well when using the Linux 5.15 LTS kernel on this AMD Zen 4 desktop.

GNU Octave Benchmark benchmark with settings of . 6.1.7-arch1-1 was the fastest.

Those using GNU Octave as an alternative to MATLAB should fine the stock kernel or Zen options to be providing the best performance.


Related Articles