Strange update dynamics

Product Help Anaconda & Miniconda
1

Please have a look at the following screenshot:

Screenshot_20221206_184357
Screenshot_20221206_1843572090×1752 134 KB

I had a conda env named ai with a working cuda-enabled pytorch installation:

>>> torch.cuda.is_available()
True

Just before the last command I ran conda update --all (for env ai, but in another shell). From the messages I see that the cpu pytorch versions got installed:

...
    pytorch-1.12.1             |cpu_py39h5e1f01c_0        80.6 MB
...
    torchvision-0.13.1         |cpu_py39h378ed51_0         6.1 MB
...

As you can see, after the update, CUDA is no longer available:

(ai) F:\2022-12>python
Python 3.9.15 (main, Nov 24 2022, 14:39:17) [MSC v.1916 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> torch.cuda.is_available()
False
>>>

I had to re-install pytorch using the conda instruction on the official website. Now CUDA is detected again.

Is this behaviour normal? How am I supposed to update the packages in an environment?

PS. This is a new installation on a new machine. I noticed that import torch is incredibly slow wrt to the (less powerful) old PC.

2

dear lordthistle.

Thank you for your contribution to the anaconda community.

The Pytorch Cuba module is provided by NVIDIA and should be installed in Anaconda by the method shown in the URL below.

https://anaconda.org/pytorch-test/pytorch-cuda

As you know, the official URL for the CUBA version of Pytorch is
The following URL is provided by NVIDIA.

There may be a distribution license issue, currently only the CPU version.
When using it with Anaconda, I think it is necessary to download it by the method at the above URL.

conda install -c pytorch-test pytorch-cuda

We apologize for the inconvenience, but thank you for your understanding.

best regards.