Which VM type (GPU) is compatible

Question

Which VM type (GPU) is compatible

Manoni 10

Update: solved, solution in answers

I'm using a VM type Standard_NC6s_v3 (GPU - 1 x NVIDIA Tesla V100) which is compatible with this image:

image="mcr.microsoft.com/azureml/curated/acpt-pytorch-1.13-cuda11.7

nvidia-smi gives me:

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.256.02   Driver Version: 470.256.02   CUDA Version: 11.7     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  Tesla V100-PCIE...  On   | 00000001:00:00.0 Off |                    0 |
| N/A   29C    P0    36W / 250W |    751MiB / 16160MiB |      5%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

and

torch.cuda.is_available()

is TRUE. So far so good.

However, I was wondering, if I'd like to use this image for python 3.10 and more recent versions of pytorch and cuda:

image="mcr.microsoft.com/azureml/curated/acpt-pytorch-2.2-cuda12.1

which VM types would be compatible? If I use this image on my Standard_NC6s_v3 then

torch.cuda.is_available()

is FALSE.

romungi-MSFT 48,911 Reputation points Microsoft Employee Moderator

2024-08-21T05:37:22.3333333+00:00

@Manoni It would be great if you could add your resolution on this thread as an answer. It would help others with similar issue in the future. Thanks!!

1 answer

Your answer

romungi-MSFT 48,911 Reputation points Microsoft Employee Moderator

2024-08-21T05:37:22.3333333+00:00

@Manoni It would be great if you could add your resolution on this thread as an answer. It would help others with similar issue in the future. Thanks!!

Answer 1

Manoni 10

To be able to use this image:

mcr.microsoft.com/azureml/curated/acpt-pytorch-2.2-cuda12.1

I specified exactly these dependencies in my yaml file:

dependencies:
  - python=3.10
  - pytorch=2.2

And then torch.cuda.is_available() is TRUE.

Share via

Which VM type (GPU) is compatible

1 answer

Your answer