Performance numbers in WSL2

luptonix

New member
I recently ran AI-Benchmark on a Windows machine in Ubuntu-WSL2 with an RTX 2060 6GB.

The results for most benchmark tests are comparable to the 2060 numbers (some faster, some slower, all within 20%) at https://ai-benchmark.com/ranking_deeplearning_detailed.html However, the last 3 benchmark tests (Pixel-RNN, LSTM-Sentiment, GNMT-Translation) run MUCH slower.

Here's the preamble:
>> AI-Benchmark-v.0.1.2
>> Let the AI Games begin..

* TF Version: 2.9.2
* Platform: Linux-5.10.102.1-microsoft-standard-WSL2-x86_64-with-glibc2.29
* CPU: N/A
* CPU RAM: 6 GB
* GPU/0: NVIDIA GeForce RTX 2060
* GPU RAM: 3.8 GB
* CUDA Version: 11.8
* CUDA Build: V11.8.89

And here are the slow results:
17/19. Pixel-RNN

17.1 - inference | batch=50, size=64x64: 3871 ± 173 ms
17.2 - training | batch=10, size=64x64: 17698 ± 519 ms

18/19. LSTM-Sentiment

18.1 - inference | batch=100, size=1024x300: 1938 ± 163 ms
18.2 - training | batch=10, size=1024x300: 7475 ± 129 ms

19/19. GNMT-Translation

19.1 - inference | batch=1, size=1x20: 727 ± 32 ms

The numbers in 17.1, 17.2, and 18.2 are 10x slower than the numbers at https://ai-benchmark.com/ranking_deeplearning_detailed.html

Has anyone else encountered this?

Thanks!
 
Top