Hello,
I’m eager to get performance improvements for my rtx 3070. So I tried today the torch-nightly1.8-cudnn8.1 version via miniconda, it installed successfully, but I got the following warning when setting the cuda device:
import torch
torch.cuda.set_device(1)
/home/corsair/miniconda3/envs/torch1.8_cudnn8.1/lib/python3.8/site-packages/torch/cuda/init.py:104: UserWarning:
GeForce RTX 3070 with CUDA capability sm_86 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_37 sm_50 compute_37.
If you want to use the GeForce RTX 3070 GPU with PyTorch, please check the instructions at Start Locally | PyTorch
warnings.warn(incompatible_device_warn.format(device_name, capability, " ".join(arch_list), device_name))
/home/corsair/miniconda3/envs/torch1.8_cudnn8.1/lib/python3.8/site-packages/torch/cuda/init.py:104: UserWarning:
GeForce RTX 2070 SUPER with CUDA capability sm_75 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_37 sm_50 compute_37.
If you want to use the GeForce RTX 2070 SUPER GPU with PyTorch, please check the instructions at Start Locally | PyTorch
warnings.warn(incompatible_device_warn.format(device_name, capability, " ".join(arch_list), device_name))
I think something is off, I was working with nightly 1.8-cudnn8.0.5 just fine with both my gpus. 1.9-cudnn8.1 also works fine.