0

enter image description here

I want to install CUDA, so I need to check the driver and CUDA version, but when I input nvidia-smi I get the following error: Command "nvidia-smi" not found, but can be installed with: My graphics card should be a MTT S80.

I installed CPU-version pytorch, but when I ran my code this error was reported:

Traceback (most recent call last):
File "/home/geohpc/lqqq/ugs/main.py", line 217, in <module>
main(params0, model0)
File "/home/geohpc/lqqq/ugs/main.py", line 187, in main
exp_lr_scheduler
File "/home/geohpc/lqqq/ugs/train.py", line 130, in train
model.cuda()
File "/home/geohpc/下载/shqyes/envs/DL/lib/python3.7/site-
packages/torch/nn/modules/module.py", line 749, in cuda
return self._apply(lambda t: t.cuda(device))
File "/home/geohpc/下载/shqyes/envs/DL/lib/python3.7/site-
packages/torch/nn/modules/module.py", line 641, in _apply
module._apply(fn)
File "/home/geohpc/下载/shqyes/envs/DL/lib/python3.7/site-
packages/torch/nn/modules/module.py", line 664, in _apply
param_applied = fn(param)
File "/home/geohpc/下载/shqyes/envs/DL/lib/python3.7/site-
packages/torch/nn/modules/module.py", line 749, in <lambda>
return self._apply(lambda t: t.cuda(device))
File "/home/geohpc/下载/shqyes/envs/DL/lib/python3.7/site-packages/torch/cuda/__init__.py",
line 221, in _lazy_init
raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled
ERROR conda.cli.main_run:execute(49): `conda run python /home/geohpc/lqqq/ugs/main.py`
failed. (See above for error)
karel
  • 114,770
lqqq
  • 11
  • 3
    Does this answer your question? How do I know which NVIDIA driver I need? – Daniel T Jan 31 '24 at 05:30
  • Show the actual output of lspci in its entirety. The MTT S80 is a Chinese product that is not using the AMD Radeon nor NVIDIA systems, and therefore the nVidia drivers won't help you – Thomas Ward Jan 31 '24 at 06:24
  • Based on Thomas' answer and your previous question, you should stop trying to install CUDA. The simplest solution is to use CPU until you hit a bottleneck. Then, based on what libraries you are using, you could use TensorFlowLite with Vulkan, or PyTorch with an OpenCL extension – Daniel T Jan 31 '24 at 06:38
  • Thanks for the advice, it means a lot to me! – lqqq Jan 31 '24 at 07:40
  • Forgive me, my friend. I install CPU-version pytorch as you suggested, when I run my code, an error was reported (I added it to the original article), Do you know how to solve it? – lqqq Jan 31 '24 at 13:38
  • @lqqq Replace all occurrences (remember to search in your entire project folder /home/geohpc/lqqq/ugs/) of model.cuda() with model, which means remove all .cuda() calls, and that will fix your error. If there are more errors, remove all .to() calls (but as before keep the variable) – Daniel T Jan 31 '24 at 14:11
  • 1
    Note that the code specific questions to remove any CUDA references should be its own question. Your core question here was "which nvidia package do i need" which the answer is "None" – Thomas Ward Jan 31 '24 at 20:28
  • Also note that code specific questions do not belong on AskUbuntu. They may belong on another website on the StackExchange network, but porting things away from CUDA is hard. Even though I was able to (AMD), I would not recommended it, and I suggest just getting a Nvidia card with CUDA – Daniel T Jan 31 '24 at 20:37

1 Answers1

1

I would suspect based on the hardware you've stated that none of the NVIDIA drivers will help you, because the MTT S80 is not an NVIDIA graphics card.

A quick cursory google search shows that the MTT S80 is a Chinese brand graphics card that does not use an NVIDIA chipset, nor an AMD chipset. It's actually a PowerVR based card, and from various sources I could tell online, the MTT S80's actual software/drivers support just is not there for any use.

Also, given that it is not an NVIDIA card, you won't get nvidia-smi ever detecting the card. This is supported by the fact you ran lspci | grep -i nvidia and got nothing back. Therefore, I don't believe you will be able to use the NVIDIA CUDA toolkit with an MTT S80.

Thomas Ward
  • 74,764
  • Forgive me, my friend. I install CPU-version pytorch as you suggested, when I run my code, an error was reported (I added it to the original article), Do you know how to solve it? – lqqq Jan 31 '24 at 13:47
  • @lqqq Answers on that front are going to be entirely different than the original question asked. You should post your separate issue as a separate question. – Thomas Ward Jan 31 '24 at 15:16