Folding Forum

My folding machines are running Fedora 33 and have been operating fine until today when Fedora rolled out an update to Linux kernel 5.9.8. I rebooted one machine to test that everything still worked and got these errors in the log:

CUDA: Not detected: cuInit() returned 999
OpenCL: Not detected: clGetPlatformIDs() returned -1001

WARNING:FS00:No CUDA or OpenCL 1.2+ support detected for GPU slot 00: gpu:1:0 TU116 [GeForce GTX 1650 SUPER]. Disabling.

After looking at the dmesg output, I found this error:

kernel: nvidia_uvm: Unknown symbol radix_tree_preloads (err -2)

After researching this error, I found this thread:

https://www.phoronix.com/forums/forum/l ... ther-month
NVIDIA Doesn't Expect To Have Linux 5.9 Driver Support For Another Month
While NVIDIA is usually quite timely in supporting new versions of the Linux kernel and aim to have out a driver by the end of the release candidates for new series, in the case of the recently minted Linux 5.9 kernel it's taking a lot longer...

So anyone who wants to fold on GPU with kernel version 5.9 will have to wait until NVIDIA rolls out a newer driver than 455.38.

Thanks for sharing + your investigation

The latest akmod-nvidia package from RPMFusion installs version 455.45 of the driver which fixes folding on the 5.9.8 Linux kernel.

That's usually the case.
When you update linux kernels, you'll have to reinstall the Nvidia drivers, occasionally needing to install a newer driver.

For future reference: I just repeated steps 2.7-2.9 of the process laid out on the page that comes up by clicking the link below (it's known to work for Fedora distribution users), to upgrade to 455.45.01 (you'll need to use "Short Lived" as one of the key words when looking on the NVidia Drivers Search/Download page for now). But it will ask you if you want to update to the newer version you have downloaded, as part of the prompts. Select "YES".

https://www.if-not-true-then-false.com/ ... dia-guide/

Paul

For some reason the gpu gets disabled in folding@home and bonic as well if the "cuda" libaries from @system are installed (never had this problem until 5.9). they are not required to run cuda apps because the nvidia driver has the required libraries, but if you develop apps using cuda you are f**ked, luckly I run machine learning apps in java on nvidia gpus using a precompiled cuda libary from maven but if you use c++ you are in trouble.