[SOLVED] It appears that CUDA is broken in the latest NVIDIA beta and short lived drivers for Slackware users

RadicalDreamer · 07-23-2018, 07:06 PM

Anybody else with NVIDIA cards experiencing this?

https://devtalk.nvidia.com/default/t...vers-on-linux/

slacktroll · 07-24-2018, 09:13 AM

It fails for me with the beta vulkan driver and the new short term for me as well.

I went back to 390.67 and nvenc works like a charm. I guess there's no reason for the newer drivers at all, unless you want to use the latest dxvk.

dxvk 0.54 is "legacy" version and works great with 390.xx

RadicalDreamer · 07-24-2018, 09:40 AM

The 396.24 short lived driver is fine as well. Yes people can use older drivers just fine right now. I'm worried about the future where a xorg update forces these new drivers on us. I like cuda with blender and CUVID is vital to my existence.

I'm wondering if the issue has anything to do with the init system that shall not be named.

RadicalDreamer · 08-02-2018, 02:37 AM

The solution seems to be: "had the same issue after upgrading. Did some experimentation with my kernel config. Apparently .45 needs "Numa Memory Allocation and Scheduler Support"(CONFIG_NUMA) enabled, at least on my particular hardware/software combo. Had always assumed that option was for multisocket systems."
https://devtalk.nvidia.com/default/t...ers-on-linux/2

RadicalDreamer · 08-04-2018, 01:21 AM

The problem still exists in NVIDIA short lived driver 396.51

RadicalDreamer · 08-21-2018, 09:29 PM

The problem still exists in NVIDIA short lived driver 396.54.
https://devtalk.nvidia.com/default/t...kernel-module/

slacktroll · 08-23-2018, 12:10 PM

so what you are saying is that CONFIG_NUMA=y is enough to get it working?

RadicalDreamer · 08-23-2018, 03:36 PM

Quote:

Originally Posted by slacktroll

so what you are saying is that CONFIG_NUMA=y is enough to get it working?

That appears to be the case. I don't roll my own kernels so I don't know for sure but that is what I gather from the threads on NVIDIA forum. NVIDIA's response in my last post was "The CONFIG_NUMA requirement wasn't intentional and is being tracked in bug 2316155" so NVIDIA is aware and acknowledges it at least.

slacktroll · 08-24-2018, 11:46 AM

Quote:

Originally Posted by RadicalDreamer

That appears to be the case. I don't roll my own kernels so I don't know for sure but that is what I gather from the threads on NVIDIA forum. NVIDIA's response in my last post was "The CONFIG_NUMA requirement wasn't intentional and is being tracked in bug 2316155" so NVIDIA is aware and acknowledges it at least.

Recompiled kernel from -current, added CONFIG_NUMA=y (and yes to everything it asked for) , downloaded NVIDIA-Linux-x86_64-396.51.02.run

nvenc in ffmpeg now works again.

Thanks :-)

RadicalDreamer · 09-16-2018, 01:15 AM

The new 4.14.70 Linux kernel in Current has

Code:

NUMA n -> y

so CUDA is now operational again in Slackware with the Short-Lived NVIDIA driver thanks to our dear BDFL!