[SOLVED] Wanted - Headless CUDA

[SOLVED] Wanted - Headless CUDA

Postby jaywalker » May 7th, '16, 20:59

This is for a Mageia 5 system running on an AMD A10-5600 APU. I tried to use the GPU acceleration for Cycles rendering in Blender but my 6-GPU core APU appears to be incompatible. My solution was to buy a GeForce GTX 750 Ti and use CUDA instead of OpenCL. So far, so good. I have the displays up and running as before on the AMD graphics processor (still using the fglrx driver) but I cannot for the life of me guess how to get the nvidia kernel module to load so that I can use its CUDA support. It makes no logical sense to use the Mageia recommended route of XFdrake or similar as that would wreck the xorg.conf for my 3-head setup using fglrx.

If I could determine where the various kernel modules and files should be located I could probably make this work for CUDA. Is there a recommended method for setting up a headless graphics card for use as a parallel processor?

Richard
Last edited by jaywalker on May 8th, '16, 14:14, edited 1 time in total.
jaywalker
 
Posts: 341
Joined: Nov 17th, '11, 02:38
Location: Belfast, Northern Ireland

Re: Wanted - Headless CUDA

Postby doktor5000 » May 8th, '16, 13:38

Did you do a quick search for "headless cuda" ? There are several relevant results, did you already check those out?
E.g. https://sites.google.com/site/akohlmey/ ... u-coolness or https://devtalk.nvidia.com/default/topi ... lf-at-80c/ or https://reference.wolfram.com/language/ ... dless.html
Cauldron is not for the faint of heart!
Caution: Hot, bubbling magic inside. May explode or cook your kittens!
----
Disclaimer: Beware of allergic reactions in answer to unconstructive complaint-type posts
User avatar
doktor5000
 
Posts: 17630
Joined: Jun 4th, '11, 10:10
Location: Leipzig, Germany

Re: Wanted - Headless CUDA

Postby jaywalker » May 8th, '16, 14:13

Yes Doc, I spent Friday evening and most of Saturday following red-herrings. It was all very interesting and kept me teetering on the edge of impending success but most of the cases I came across were either trying to run headless bitcoin-mining servers or were using their Nvidia cards as their main display device. There were a few who were using laptops (I think) with Intel graphics cards and Nvidia GPUs for rendering but their issues usually had more to do with some intermediate software they all use; I think primus was mentioned, but I could be mis-remembering.

I can report some success, however. Before posting yesterday I had been trying for a few hours to figure out where to put the nvidia stuff so that I could load the kernel module and see what happens. I was having trouble reconciling the rpm contents list with what I could see had actually been installed. In the end I stumbled across the gem of information that the nouveau module was loaded and in use! I think it showed up in the output from lspci -v. I blacklisted that and tried again, just a plain ol' modprobe nvidia. It did the trick and I was able to get Blender to see it and use it - around 5 or 6 times faster to render a particular scene than OpenCL on the CPU. Went to bed content, but today I am still trying to get back to that happy place.

However, I am pleased to consider this issue is now solved.
Thank you for your kind attention,

Richard
jaywalker
 
Posts: 341
Joined: Nov 17th, '11, 02:38
Location: Belfast, Northern Ireland


Return to Video

Who is online

Users browsing this forum: No registered users and 1 guest

cron