search

n00mkrad / flowframes

Flowframes Windows GUI for video interpolation using DAIN (NCNN) or RIFE (CUDA/NCNN)
GNU General Public License v3.0
1.47k stars 114 forks source link

CUDA-capable GPU device is not available, running on CPU instead #171

Open shookehh opened 2 years ago

shookehh commented 2 years ago

Screenshot (377)

I recently tried updating to 1.35 and ever since then it wouldn't use my GPU for interpolation even though my GPU supports CUDA. Is there a fix to this? perhaps I missed something

image

8BEATER-Il-S commented 1 year ago

Same problem. Can't figure out the solution myself, so I have to use the integrated GPU :(

belongtothenight commented 1 year ago

Same here. Hope this can be improved!

Info: image image image

Solving process:

  1. Tried all default settings in the installation process. (fresh install) (as I remembered it's working pretty fast, so I assume CUDA is working.)
  2. Remove and reinstall with "Existing Python+Pytorch Installation", all required libraries are preinstalled. (Above images) CUDA doesn't work, and it takes forever to run.
  3. Use pip to remove the "torch" library, remove and reinstall with all default settings except selecting GPU support with Nvidia, support for RTX16/20 series and older. (mine is GTX1650 Ti) CUDA works! It computes 9.5 times faster than before.

Despite the title saying "Flowframes 1.36.0 [GPU: Intel(R) UHD Graphics]", it still runs with Nvidia GPU.

ash-hashtag commented 1 year ago

from the task manager it is using my dgpu(rtx3050 laptop) but the title says radeon graphics, but the gpu usage is 0%, it is being used as per the task manager, but 0% and my cpu hits 99%, so is it really using the gpu?

dmenne commented 1 year ago

Definitively have a working CUDA, doing development with Toolkit regularly. Nevertheless, same message.

dmenne commented 1 year ago

Solved: I had used "Autho" Python, and there seems to be a problem with my Python310 even if all requested models are available. Reinstalled with embedded Python, everything works fine.

hutien commented 1 year ago

What do you use for interpolation ? RFIE, VULKAN , image i had this error, any helps ?

pik-a-c-h-u commented 1 year ago

I had the same issue showing no CUDA GPU. settings>graphics settings> browse> select flowframes.exe from install location and then set it to highperformance GPU , this fixed the issue for me

hutien commented 1 year ago

How do you do that, i dont have graphics settings on windows ?

Lograthmic commented 1 year ago

Hi there! Maybe you should try to reinstall CUDA and CUDNN. I had the same issue. I use 1.36.0 and found that pytorch need 11.0.x version of CUDA. image After reinstall CUDA, the issue has been solved.

pik-a-c-h-u commented 1 year ago

How do you do that, i dont have graphics settings on windows ?

search for graphics settings in windows search (start) (im using win11) its there

pik-a-c-h-u commented 1 year ago

Hi there! Maybe you should try to reinstall CUDA and CUDNN. I had the same issue. I use 1.36.0 and found that pytorch need 11.0.x version of CUDA. image After reinstall CUDA, the issue has been solved.

my issue got resolved after force starting it in dedicated GPU

hutien commented 1 year ago

I did it but i still have this here : image