Some tips that might make you use it smoothly on Windows 10

[white0rchardpUnK]

First of all, thank the developers for their hard work to allow us to use such a very good tool. I also encountered a lot of problems when trying to make it work, and it took me a lot of time. So I hope my tips will save you time. About my configuration: 3080Ti + Intel CPU + Win10 with Anaconda installed

• Make sure your conda and pip commands are excuted in the same directory, which is this project.
• Check your Windows graphics driver version and CUDA version before installing anything in Anaconda. Sometimes the two versions are inconsistent and weird problems will occur.
• For now, the most suitable CUDA version is 11.3 for those who uses 30series graphics card. So when you execute the installation instructions, please specify cudatoolkit=11.3
• conda install pytorch==1.8.0 torchvision==0.9.0 torchaudio==0.8.0 cudatoolkit=10.2 -c pytorch This instructions in the tutorial may cause some strange problems such as not being able to find CUDA devices. My theory is that the source of priority installation is different according to the different conda configuration files. So please use conda install pytorch torchvision torchaudio cudatoolkit=11.3 to avoid these problems.（If you are not using a 30 series graphics card, specify the CUDA version as 10.2）
• If you are using PROXY like v2ray, please exit it or just close the system proxy before you use any pip commands.
• If something goes wrong and you need to reconfigure the environment, please run Anaconda Navigator with administrator privileges, and then delete the environment in it. This is much cleaner than you can delete using the command line.

[whenupdate]

If we use google colab with T4, how gonna this work ? upgrade cudatool ?

update : have upgrade conda with another jupyter notebook with installing new cuda 11.3 . but still same, stuck at 2 iterations/s . 2GB max usage.

[white0rchardpUnK]

If we use google colab with T4, how gonna this work ? upgrade cudatool ?

update : have upgrade conda with another jupyter notebook with installing new cuda 11.3 . but still same, stuck at 2 iterations/s . 2GB max usage.

I got a long time not using the Colab, maybe it is the driver issue? You can check the details about all the CUDA versions and driver versions from here.