Incompatibility with cuda, cudnn, torch and conda/anaconda

windows
1

i found an nvidia compatibility matrix, but that didnt work. testing with 2 PC’s with 2 different GPU’s and have updated to what is documented, at least i think so. not sure what to do now. i have been trying for a week.

NVIDIA-SMI 522.06 Driver Version: 522.06 CUDA Version: 11.8

import torch.cuda
torch.cuda.is_available()
False
print(torch.rand(2,4))
tensor([[0.9383, 0.1120, 0.1925, 0.9528],
[0.1820, 0.6980, 0.4946, 0.2403]])
torch.cuda.is_available()
False

using python 3.9.13 and 3.8.10

2

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

3

(base) C:\Users\rossroxas>python -m torch.utils.collect_env
Collecting environment information…
PyTorch version: 2.0.0+cpu
Is debug build: False
CUDA used to build PyTorch: Could not collect
ROCM used to build PyTorch: N/A

OS: Microsoft Windows 10 Enterprise
GCC version: (x86_64-posix-seh, Built by strawberryperl.com project) 8.3.0
Clang version: Could not collect
CMake version: Could not collect
Libc version: N/A

Python version: 3.9.16 (main, Mar 8 2023, 10:39:24) [MSC v.1916 64 bit (AMD64)] (64-bit runtime)
Python platform: Windows-10-10.0.19045-SP0
Is CUDA available: False
CUDA runtime version: 11.8.89
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration:
GPU 0: NVIDIA RTX A5500
GPU 1: NVIDIA RTX A5500
GPU 2: NVIDIA RTX A5500

Nvidia driver version: 522.06
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture=9
CurrentClockSpeed=2694
DeviceID=CPU0
Family=179
L2CacheSize=28672
L2CacheSpeed=
Manufacturer=GenuineIntel
MaxClockSpeed=2694
Name=Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz
ProcessorType=3
Revision=21767

Architecture=9
CurrentClockSpeed=2694
DeviceID=CPU1
Family=179
L2CacheSize=28672
L2CacheSpeed=
Manufacturer=GenuineIntel
MaxClockSpeed=2694
Name=Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz
ProcessorType=3
Revision=21767

Versions of relevant libraries:
[pip3] numpy==1.24.3
[pip3] pytorch-gpu==0.0.1
[pip3] torch==2.0.0
[pip3] torch-package==1.0.1
[pip3] torchlib==0.1
[pip3] torchutils==0.0.4
[pip3] torchvision==0.15.1
[conda] numpy 1.24.3 pypi_0 pypi
[conda] pytorch-gpu 0.0.1 pypi_0 pypi
[conda] torch 2.0.0 pypi_0 pypi
[conda] torch-package 1.0.1 pypi_0 pypi
[conda] torchlib 0.1 pypi_0 pypi
[conda] torchutils 0.0.4 pypi_0 pypi
[conda] torchvision 0.15.1 pypi_0 pypi

4

Based on the output you are installing the CPU-only binary. Which install command did you use?

5

i tried:

pip install torch
nvidia for the CUDA graphics driver and cudnn. I had the impression that everything was included and maybe distributed so that i can check the GPU after the graphics driver install.

6

i have a few different versions of python

Python version: 3.8.10 (tags/v3.8.10:3d8993a, May 3 2021, 11:48:03) [MSC v.1928 64 bit (AMD64)] (64-bit runtime)
Python platform: Windows-10-10.0.19045-SP0
Is CUDA available: True
CUDA runtime version: 11.8.89
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA RTX A5500
GPU 1: NVIDIA RTX A5500
GPU 2: NVIDIA RTX A5500

Nvidia driver version: 522.06
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture=9
CurrentClockSpeed=2693
DeviceID=CPU0
Family=179
L2CacheSize=28672
L2CacheSpeed=
Manufacturer=GenuineIntel
MaxClockSpeed=2693
Name=Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz
ProcessorType=3
Revision=21767

Architecture=9
CurrentClockSpeed=2693
DeviceID=CPU1
Family=179
L2CacheSize=28672
L2CacheSpeed=
Manufacturer=GenuineIntel
MaxClockSpeed=2693
Name=Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz
ProcessorType=3
Revision=21767

Versions of relevant libraries:
[pip3] numpy==1.16.6
[pip3] torch==2.0.0+cu118
[pip3] torchaudio==2.0.1+cu118
[pip3] torchvision==0.15.1+cu118
[conda] cudatoolkit 11.8.0 h09e9e62_11 conda-forge
[conda] mkl 2023.1.0 h8bd8f75_46356
[conda] mkl-include 2023.1.0 haa95532_46356
[conda] numpy 1.23.5 pypi_0 pypi
[conda] pytorch-gpu 0.0.1 pypi_0 pypi
[conda] torch 2.0.0 pypi_0 pypi
[conda] torch-package 1.0.1 pypi_0 pypi
[conda] torch-utils 0.1.2 pypi_0 pypi
[conda] torchlib 0.1 pypi_0 pypi
[conda] torchutils 0.0.4 pypi_0 pypi
[conda] torchvision 0.15.1 pypi_0 pypi

7

Use the install commands from our website.
E.g. for torch==2.0.0+cu117 on Windows you should use:

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117

Yes, all dependencies are included in the binaries. You would only need a properly installed NVIDIA driver.

8

i think one of the confusing things is finding the matrix on git i found doesnt really give straight forward line up of which versions are compatible with cuda and cudnn

9

i am actually running 11.8

10

also, do i need to use anaconda or miniconda?

11

I don’t understand which “matrix on git” you are referring to as you can just select the desired PyTorch release and CUDA version in my previously posted link.
As also mentioned your locally installed CUDA toolkit won’t be used unless you build PyTorch from source or a custom CUDA extension since the binaries ship with their own dependencies.

Again, your locally installed CUDA toolkit won’t be used, only the NVIDIA driver.

No, you don’t need conda.

12

thank you for the replies! ill test things out and update when i can!

13

the website says anaconda is a prerequisite. if that is not accurate, cant i just use python?