Hi, I got this weird issue where I keep running out of Vram, even tho there are 65GB of Swap left to use. On Windows that’s not an issue, it just renders, on Linux it says “System out of GPU Memory”. I had 2 other people test this. Here’s how it goes
I have a 3080 10 GB, I can’t (Manjaro)
Friend 1 (Manjaro) has a 1070 8GB, he has no issues
Friend 2 (Arch) has a 3060 Ti 8GB, had to disable simplify, save and reload but then it worked.
All on the same settings. I tried everything and I made sure that I am running the latest drivers, proprietary and what not. Konsole says I am running 510.60.02.
We all tested this back on 3.1.0. The issue went away for a while after installing the “Video-Nvidia” proprietary driver, but possibly due to less performance intensive scenes. Now it’s back. I got a ton of geometry nodes stuff in this current scene but I literally rendered this scene last night in 4K with a different light setup. It used 5GB of Ram but I didn’t see how much Vram. It did give an error the first time, restarted and it worked. Today it doesn’t work. I didn’t bother my friends to test this one, I just fired up windows and rendered it on first try. Currently using 3.1.2
I can’t share with you the whole blend file cause it’s using paid assets and it just so happens those are the most Vram intensive. What can I do to fix the issue or figure out the cause?
Hi.
To monitor vRAM usage on Linux, you open the terminal: watch -n 0.5 nvidia-smi
To detect peaks in total vRAM usage while rendering you keep another terminal open by pasting all of the following: a=0; while true; do b=$(nvidia-smi --query-gpu=memory.used --format=csv|grep -v memory|awk '{print $1}'); [ $b -gt $a ] && a=$b && echo $a ; sleep .5; done
To make a comparison with other machines, it is also necessary to know the total system RAM of each of them. Cycles can use half of the available system RAM if vRAM is not enough.
It would also be interesting to know if vRAM usage is similar with CUDA or OptiX
It’s the same system, I am dual booting. Other than blender there is only firefox open in the background on both systems most of the time. Ram is 32GB. While the render is starting it goes up to about 15GB Ram usage max, but on the last successful render of the same scene in the top left it said ram usage 5.5GB and peak 8 something GB. If you still think I should write those commands in the terminal let me know and I will test later on PC, but I am almost convinced it’s not the <1gb vram the rest of the PC is using, based on the Ram and the 65 GB of free Swap memory, also the Windows performance.
Blender can be very imprecise on vRAM usage information, so to make correct measurements you should use the commands I shared with you earlier.
The thing is, users don’t have a scene to test with. The only thing I can recommend is to measure vRAM usage on Linux without having Blender open and do the same from Windows. If vRAM usage of both systems is similar when Blender is closed and only Linux gives problems using the same version of Blender and the same scene, then maybe it’s something that should be reported to the bug tracker. But developers will also need the scene for testing, and if you can’t share it with them then that will be a problem.
So I don’t know what to recommend either. Maybe try different versions of the nvidia driver
Ok I have no clue what’s going on. I couldn’t render it on Windows either, my screen would turn black for a second so I assume my gpu drivers were crashing. Anyways I reduced the unnecessary geometry and it’s fixed now so I guess that was the issue. On the bright side the scene now renders in record speeds.
On the other hand Friend 1 sent me this so who knows. Either way doesn’t really look like an issue with Blender
I think the problem still might be the drivers because yesterday I tried one of the commands from your forum post and the scene the whole time used ~5gb vram, blender reported ~4.9gb vram but the highest spike was 7gb so if your scene is 9 gb and one day it renders fine the other day blender might just store data differently and it spikes to 12gb or something and that obviously crashes because the drivers are messed up
So yeah in the event someone figures anything out here I’d love to know.