Not able to use torch with NVIDIA Geforce MX130

Megha_jain · January 14, 2023, 3:47pm

I have a laptop with Intel UHD card and NVIDIA Geforce MX130. My CUDA version is 12.0
Environment path variable is C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0.
torch version 1.13.1.
Please help me with following errors
But torch.cuda.is_available() is shown as false
but for tensor flow
tf.test.is_built_with_cuda() is shown as true
tf.config.experimental.list_physical_devices(‘gpu’) is []

Additional info:

nvcc --version is

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Mon_Oct_24_19:40:05_Pacific_Daylight_Time_2022
Cuda compilation tools, release 12.0, V12.0.76
Build cuda_12.0.r12.0/compiler.31968024_0

from numba import cuda
cuda.detect()

shows

Found 1 CUDA devices
id 0 b’NVIDIA GeForce MX130’ [SUPPORTED]
Compute Capability: 5.0
PCI Device ID: 0
PCI Bus ID: 1
Watchdog: Enabled
Compute Mode: WDDM
FP32/FP64 Performance Ratio: 32
Summary:
1/1 devices are supported

nvidia-smi

shows

±----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| No running processes found |
±----------------------------------------------------------------------------+

ptrblck · January 14, 2023, 8:22pm

Could you post the output of python -m torch.utils.collect_env please?

Megha_jain · January 15, 2023, 1:15am

Thanks for the reply. Following is the output

Traceback (most recent call last):
File “C:\Users\Megha\anaconda3\lib\runpy.py”, line 188, in _run_module_as_main
mod_name, mod_spec, code = _get_module_details(mod_name, _Error)
File “C:\Users\Megha\anaconda3\lib\runpy.py”, line 111, in get_module_details
import(pkg_name)
File "C:\Users\Megha\anaconda3\lib\site-packages\torch_init.py", line 139, in
raise err
OSError: [WinError 126] The specified module could not be found. Error loading “C:\Users\Megha\anaconda3\lib\site-packages\torch\lib\caffe2_nvrtc.dll” or one of its dependencies.

Megha_jain · January 15, 2023, 1:20am

By doing above exercise, I also found that my path is C:\Users\Megha\AppData\Local\Microsoft\WindowsApps whereas output of os.path.dirname(sys.executable) is C:\Users\Megha\anaconda3.

Can this be creating some issue? Sorry in advance as I am new to all this and trying to self learn

ptrblck · January 15, 2023, 10:25pm

Based on this topic it seems as if the installation might be corrupt somehow and reinstalling it fixed the issue. Do you see the caffe2_nvrtc.dll in the posted folder?

AlanJohnstone · January 15, 2023, 11:08pm

What is the “posted folder”
Alan

ptrblck · January 15, 2023, 11:13pm

The install folder where the missing .dll should be located.

Megha_jain · January 16, 2023, 1:47am

I have reinstalled torch but again I am facing the same problem. What might be causing faulty installation

Megha_jain · January 16, 2023, 3:00am

Sorry for the confusion but I think I am not able to uninstall torch in the first place.
conda uninstall torch
Collecting package metadata (repodata.json): …working… done
Solving environment: …working… failed

Note: you may need to restart the kernel to use updated packages.

PackagesNotFoundError: The following packages are missing from the target environment:

torch

However, output of pip show torch is

ame: torch
Version: 1.13.1
Summary: Tensors and Dynamic neural networks in Python with strong GPU acceleration
Home-page: https://pytorch.org/
Author: PyTorch Team
Author-email: packages@pytorch.org
License: BSD-3
Location: c:\users\megha\anaconda3\lib\site-packages
Requires: typing_extensions
Required-by: torchvision, torchaudio
Note: you may need to restart the kernel to use updated packages.

ptrblck · January 16, 2023, 7:43am

Try to run conda uninstall pytorch instead as this should be the conda package name.