4 ms·
Are you running the scripts in the optimizedSD folder instead of the scripts folder? For me a 512x512 image uses 3.7GB of VRAM. You can reduce the resolution to
by squeaky-clean 4y ago
Are you running the scripts in the optimizedSD folder instead of the scripts folder? For me a 512x512 image uses 3.7GB of VRAM. You can reduce the resolution to use less (to a point, the model needs a fixed amount).
- stavros 4y agoI am, yes. I don't think it's the memory because it segfaults immediately, even if I try 64x64... It's very odd, I'm running Ubuntu 22.04, I'm not sure if the CUDA version is somehow wrong.
- squeaky-clean 4y agoOh damn that's weird. Any chance the segfault is on the CPU side? When I try to generate too large of an image I get an allocation error rather than a segfault. Also crazy idea but could it be related to open source / proprietary driver differences? I haven't done nvidia on Linux in a long time so I don't know if that's in a more reasonable place yet.
- stavros 4y agoHmm, thanks for the info, it might be a CPU error! There's zero information in the error, it just says "error SIGSEGV" or something similar. I don't think it's an open source driver issue, I'm using nVidia's proprietary one... It's so odd, it just dies outright. Maybe it is a CPU error, as you say, I think I'll investigate there.