RuntimeError: No GPU detected, Please check your CUDA version and installation

[Pengxin1028]

Hi,

I met this problem during refine step even if I downloaded the right tensorflow-gpu version.
**10-20 11:26:01, INFO
######Isonet starts refining######

10-20 11:26:05, ERROR No GPU detected, Please check your CUDA version and installation
10-20 11:26:05, ERROR Traceback (most recent call last):
File "/home/spuser/IsoNet/bin/refine.py", line 25, in run
run_whole(args)
File "/home/spuser/IsoNet/bin/refine.py", line 105, in run_whole
check_gpu(args)
File "/home/spuser/IsoNet/bin/refine.py", line 275, in check_gpu
raise RuntimeError('No GPU detected, Please check your CUDA version and installation')
RuntimeError: No GPU detected, Please check your CUDA version and installation**

Here is my system info:
System: Centos 7,
CUDA version: 10.2
Driver version: 440.31
GPU: 4X RTX2080Ti

Any thoughts?

Best,
Pengxin

[procyontao]

Hi Pengxin,

What is the version of cudnn?

[Pengxin1028]

My cudnn version is cudnn8-8.2.4

[procyontao]

Did you check whether your versions matches the ones in table? I can not guarantee that cuda10.2 will work because it is not listed. https://www.tensorflow.org/install/source#gpu

[Pengxin1028]

Hi, I used the wrong version of cudnn. I should use cudnn7.6.
The refinement seems to working now!10-20 13:55:05, INFO
######Isonet starts refining######

10-20 13:55:10, INFO Done preperation for the first iteration!
10-20 13:55:10, INFO Start Iteration1!
10-20 13:55:14, INFO Noise Level:0.0
10-20 13:55:34, INFO Done preparing subtomograms!
10-20 13:55:34, INFO Start training!
10-20 13:55:36, INFO Loaded model from disk
10-20 13:55:36, INFO begin fitting
Epoch 1/10
49/100 [=============>................] - ETA: 17s - loss: 0.1405 - mse: 0.0348 - mae: 0.1405

Much appreciate your help!

[aseedb]

I got the same error but in my case it was because I had the wrong version of tensorflow-gpu. Pip would not install version 2.3 because of dependency requirements not met. Finally I got it to install it by downgrading python from 3.9 to 3.8. This solved the issue and now isonet training is running smoothly.

[jychoi0616]

Dear team,

I got the similar error since I updated CUDA to 12.2. as below:

2024-04-08 13:46:35.112110: E external/local_xla/xla/stream_executor/cuda/cuda_driver.cc:274] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected

File "/data/IsoNet/bin/refine.py", line 227, in check_gpu
raise RuntimeError('No GPU detected, Please check your CUDA version and installation')
RuntimeError: No GPU detected, Please check your CUDA version and installation

I already installed tensorflow 2.15.0 and python3.9 for the isonet conda environment to match cuda 12.2.
Could you help me in this please?

[procyontao]

Hi, please check whether the information here: https://groups.google.com/g/isonet/c/fAygyajy6Zc can help.