CUDA 10.2 not detecting GPU with 470xx drivers

Hi,

I have a GT650M Kepler card with compute capability 3.0. Recent drivers dropped support for it, so i switched back to 470xx drivers on arch AUR, and installed cuda 10.2, also from AUR. However, none of the CUDA programs work:

> ./bandwidthTest 
[CUDA Bandwidth Test] - Starting...
Running on...

cudaGetDeviceProperties returned 999
-> unknown error
CUDA error at /dvs/p4/build/sw/rel/gpgpu/toolkit/r10.2/demo_suite/bandwidthTest/bandwidthTest.cu:255 code=999(cudaErrorUnknown) "cudaSetDevice(currentDevice)" 
> ./nbody
Run "nbody -benchmark [-numbodies=<numBodies>]" to measure performance.
...

NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.

Error: only 0 Devices available, 1 requested.  Exiting.
> ./vectorAdd 
[Vector addition of 50000 elements]
Failed to allocate device vector A (error code unknown error)!
> ./randomFog
Random Fog
==========

CUDA error at /dvs/p4/build/sw/rel/gpgpu/toolkit/r10.2/demo_suite/randomFog/../../samples/common/inc/helper_cuda.h:775 code=999(cudaErrorUnknown) "cudaGetDeviceCount(&device_count)" 
> ./deviceQuery 
./deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount returned 999
-> unknown error
Result = FAIL
> 
> nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Wed_Oct_23_19:24:38_PDT_2019
Cuda compilation tools, release 10.2, V10.2.89
# sudo pacman -Q | grep nvidia
nvidia-470xx-dkms 470.103.01-1
nvidia-470xx-utils 470.103.01-1
nvidia-prime 1.0-4
opencl-nvidia-470xx 470.103.01-1

Also, another strange thing is :

> clinfo 
Number of platforms                               0

Which could mean opencl isn’t working, however glxgears run just fine.

Graphics work just fine, but CUDA doesn’t :

> nvidia-smi
Sat Mar 12 22:33:49 2022       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.103.01   Driver Version: 470.103.01   CUDA Version: 11.4     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  Off  | 00000000:01:00.0 N/A |                  N/A |
| N/A   60C    P8    N/A /  N/A |    775MiB /  2000MiB |     N/A      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

How do I get CUDA 10.2 to work?

I had also installed driver version 470.103 and had the same problem. I reinstalled version 470.94, and it is working again.

Thanks @npape, you’re a godsend! All’s well now.

For anyone experiencing the issues on Arch : get the .tar.gz file from “download” link on this page, or directly download it via this link. The rest is the same as with any package from AUR.