search

xinntao / Real-ESRGAN-ncnn-vulkan

NCNN implementation of Real-ESRGAN. Real-ESRGAN aims at developing Practical Algorithms for General Image Restoration.
https://github.com/xinntao/Real-ESRGAN
Other
1.42k stars 170 forks source link

realesrgan-ncnn-vulkan (v0.2.0) stucks/hangs on Google Colab (Ubuntu 22.04 LTS / Tesla T4 GPU /Vulkan 1.3.231) #62

Open scopeallen opened 4 months ago

scopeallen commented 4 months ago

Command:

! ~/realesrgan-ncnn-vulkan -i ~/test.png -o /root/output.png -s 4 -m ~/models -n xinntao-realesrgan-x4plus -f png

Output:

[0 llvmpipe (LLVM 15.0.7, 256 bits)] queueC=0[1] queueG=0[1] queueT=0[1] [0 llvmpipe (LLVM 15.0.7, 256 bits)] bugsbn1=0 bugbilz=0 bugcopc=0 bugihfa=0 [0 llvmpipe (LLVM 15.0.7, 256 bits)] fp16-p/s/a=1/1/1 int8-p/s/a=1/1/1 [0 llvmpipe (LLVM 15.0.7, 256 bits)] subgroup=8 basic=1 vote=1 ballot=1 shuffle=1

and stucks/hangs here forever.

NeedsMoar commented 2 months ago

Well, you're running it on the llvmpipe software renderer which is likely the only thing that provides a high enough vulkan version if the upscaler chose that (and it's not stuck forever, it'll eventually output something someday, I seem to recall those colab instances having pretty terrible CPU allocation since the GPU was the main purpose), or possibly you just need to run vulkaninfo and specify -g # to specify the device number corresponding to the T4 if its drivers have a compatible vulkan.