search

gaomingqi / Track-Anything

Track-Anything is a flexible and interactive tool for video object tracking and segmentation, based on Segment Anything, XMem, and E2FGVI.
MIT License
6.3k stars 470 forks source link

Compatibility Issue with NVIDIA GeForce RTX 4090 and PyTorch #101

Open EricaCherry opened 11 months ago

EricaCherry commented 11 months ago

I've been trying to run the Track-Anything code on my machine which has an NVIDIA GeForce RTX 4090 GPU, but I've been encountering compatibility issues.

The problem arises due to the CUDA capability of my GPU (sm_89) not being compatible with the PyTorch version that the code supports. The current PyTorch installation supports CUDA capabilities sm_37, sm_50, sm_60, sm_61, sm_70, sm_75, and compute_37, but not sm_89 which is required for the RTX 4090.

I tried to resolve this by installing different versions of PyTorch that are compatible with CUDA 10.2 (which is installed on my machine), but none of them seem to support sm_89. I also attempted to run the code on the CPU as a workaround, and it worked, but Its super slow.

Could you please provide some guidance on how to resolve this issue? Is there a way to make the code compatible with newer GPUs like the RTX 4090?

EricaCherry commented 11 months ago

changed the device from cuda:3 to cuda:0 in the app.py file and it seems to work now...

what worked for me Python: 3.8.12 PyTorch: 1.10.0 CUDA: 11.7 NVIDIA Driver: 536.67 GPU: NVIDIA GeForce RTX 4090