RicciH8
(Ricci)
January 26, 2021, 10:49am
1
Hello,
I’ve seen many posts online from last year about this but I’m still to find a solution that works for me.
I get this message: “GeForce RTX 3090 with CUDA capability sm_86 is not compatible with the current PyTorch installation”.
Is there a solution now or a planned release date for a version with compatibility? I’ve tried most answers from other posts with no avail.
Thanks in advance,
Ricci
1 Like
ptrblck
January 26, 2021, 10:57am
2
You would have to install the PyTorch binaries with the CUDA11 runtime from here or alternatively you could also build form source.
3 Likes
ykpgrr
(Yakup Görür)
April 22, 2021, 10:28am
5
You can use ‘lambda stack’ instead of installation pytorch from source.
Fast and easy:
Instructions for getting TensorFlow and PyTorch running on NVIDIA's GeForce RTX 30 Series GPUs (Ampere), including RTX 3090, RTX 3080, and RTX 3070.
1 Like
I did so via conda (cudatoolkit=11.3) and yet torch.version.cuda is 10.2.
In that case your install command might either have been wrong or you might have different PyTorch versions installed in the same environment. Create a new virtual environment and reinstall PyTorch with the CUDA 11 runtime.
1 Like