Thank you! Means a lot to know it’s been helpful! I’m adding to the full PyTorch + CUDA playlist every week: ruclips.net/p/PLXsFtK46HZxWKgwYZzh_c2MmGxIDnRGUT
Does the CUDA available if-else check work? If True, then it did detect and you should be able to move your tensors onto cuda and do your tensor ops there.
Hey Art, there should be no difference. The video was posted just 4 weeks ago, and at some point I'm bound to do an update on the CUDA installed on my machine, but from past experiences this is quite straightforward :)
i have rtx 3070 when i exucute (nvidia smi) i get cuda version 12 how can i get pytorch 12 version and wich cuda version should i install and cudann and toolkit plz
What does PyTorch 12 version mean? You can install the latest version of PyTorch through pip: pip install torch Latest “version” is PyTorch 2, which is something I would cover in future videos: pytorch.org/get-started/pytorch-2.0/
Ahh ok you mean you have CUDA version 12 and PyTorch 2? if that’s the case you can downgrade CUDA to 11.x, something that PyTorch officially supports or you might just wait a while until CUDA 12 is supported. Does the exercises not work for you? If you’re lucky you might be able to follow along the whole series without any problem with CUDA too
@@SamuelChan my training some times go fast and some times slow in the same code without change i print device when it train gives me cuda but it is like exucte in cpu
What graphics card are you using? Try a more expensive, parallelizable computation and you should have a noticeable difference between gpu and cpu training.
Bro be teaching at 1.25x speed by default. SUBSCRIBED!!
Too respectful of your time to waste it with empty fillers and pauses. ❤️
Excellent, Thorough and well put-together tutorial explaining the details and important stuff. Keep up the good work!
Thank you! Means a lot to know it’s been helpful!
I’m adding to the full PyTorch + CUDA playlist every week:
ruclips.net/p/PLXsFtK46HZxWKgwYZzh_c2MmGxIDnRGUT
Very informative presentation Samuel !
Thank you so much
Thank you! Glad to know they were helpful! :)
What if nvidia-smi shows me CUDA Version: 12.2 and only available on the pytorch site is 11.7 and 11.8?
Does the CUDA available if-else check work? If True, then it did detect and you should be able to move your tensors onto cuda and do your tensor ops there.
Can Kaggle or Google Colab be used? Don’t have a GPU that’s why?
It would be interesting to know how to work with CUDA 11.8 and higher
Hey Art, there should be no difference. The video was posted just 4 weeks ago, and at some point I'm bound to do an update on the CUDA installed on my machine, but from past experiences this is quite straightforward :)
i have rtx 3070 when i exucute (nvidia smi) i get cuda version 12 how can i get pytorch 12 version and wich cuda version should i install and cudann and toolkit plz
What does PyTorch 12 version mean?
You can install the latest version of PyTorch through pip: pip install torch
Latest “version” is PyTorch 2, which is something I would cover in future videos: pytorch.org/get-started/pytorch-2.0/
@@SamuelChan yes i mean that in site of pytorch we have just 11.8 but my cuda version in nvidia smi is 12 my be i have icompatible driver
Ahh ok you mean you have CUDA version 12 and PyTorch 2? if that’s the case you can downgrade CUDA to 11.x, something that PyTorch officially supports or you might just wait a while until CUDA 12 is supported.
Does the exercises not work for you? If you’re lucky you might be able to follow along the whole series without any problem with CUDA too
@@SamuelChan my training some times go fast and some times slow in the same code without change i print device when it train gives me cuda but it is like exucte in cpu
What graphics card are you using?
Try a more expensive, parallelizable computation and you should have a noticeable difference between gpu and cpu training.