Running eth_mnist.py on GPU(cuda), but it took approximately 14 hours, am I doing something incorrect? #456
-
|
Good evening, |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
I am assuming that you are using Windows 10 under the Dev channel, therefore, the pre-release beta version WSL2 has GPU access. WSL2 added much-needed features compared to WSL1 like GPU access. But where WSL2 can access the GPU with no issue, performance-wise, it leaves much room for improvements (therefore, it beta status). The performance didn't improve much since June 2020 link Tip, when you run PyTorch task that use GPU under WSL2, check the Windows 10, Task Manager - performance tab, you can see that your GPU is underutilized. If you want to utilize your GPU, you have two options. 1) use native Linux - the preferred approach. 2) use native python on windows 10 with GPU access. (I never used this approach, so I don't know if it is working well.) PS, if you run the program without plots, it will but much faster |
Beta Was this translation helpful? Give feedback.
-
|
Good evening, Yes, I'm using Windows 10 under the Dev channel. Thank you for your answer. |
Beta Was this translation helpful? Give feedback.
I am assuming that you are using Windows 10 under the Dev channel, therefore, the pre-release beta version WSL2 has GPU access. WSL2 added much-needed features compared to WSL1 like GPU access. But where WSL2 can access the GPU with no issue, performance-wise, it leaves much room for improvements (therefore, it beta status). The performance didn't improve much since June 2020 link
Tip, when you run PyTorch task that use GPU under WSL2, check the Windows 10, Task Manager - performance tab, you can see that your GPU is underutilized.
If you want to utilize your GPU, you have two options. 1) use native Linux - the preferred approach. 2) use native python on windows 10 with GPU access. (I ne…