Cannot install CUDA on CI server due to timeout

Hello.

I’m trying to configure my AppVeyor CI server to install CUDA but it fails due to timeout. Here is my CI log:

Build started
git clone -q ...
git fetch -q origin +refs/pull/34/merge:
git checkout -qf FETCH_HEAD
Running Install scripts
appveyor DownloadFile https://developer.nvidia.com/compute/cuda/9.1/Prod/local_installers/cuda_9.1.85_win10 -FileName cuda_9.1.exe
Downloading cuda_9.1.exe (1,446,894,456 bytes)...1%
Downloading cuda_9.1.exe (1,446,894,456 bytes)...10%
Downloading cuda_9.1.exe (1,446,894,456 bytes)...20%
Downloading cuda_9.1.exe (1,446,894,456 bytes)...30%
Downloading cuda_9.1.exe (1,446,894,456 bytes)...40%
Downloading cuda_9.1.exe (1,446,894,456 bytes)...50%
Downloading cuda_9.1.exe (1,446,894,456 bytes)...60%
Downloading cuda_9.1.exe (1,446,894,456 bytes)...70%
Downloading cuda_9.1.exe (1,446,894,456 bytes)...80%
Downloading cuda_9.1.exe (1,446,894,456 bytes)...90%
Downloading cuda_9.1.exe (1,446,894,456 bytes)...100%
Start-Process -FilePath "cuda_9.1.exe" -ArgumentList "-s nvcc_9.1 cublas_9.1 cublas_dev_9.1 cufft_9.1 cufft_dev_9.1 npp_9.1 npp_dev_9.1 Display.Driver" -Wait

Then in hangs for 1 hour without writing any logs and then CI cancels the build due to timeout.

Why it takes so much time? Is there any workaround to reduce it? I don’t want to increase timeout, because it’s not fonny when build takes several hours.

It seems that the last parameter Display.Driver require some user interaction and it doesn’ work on UI-less CI server. However, documentation says that it’s mandatory for CUDA.

What should I do?

You’ll need to pass the necessary instructions for silent install. Run the actual command (interactively, on a server with a display) that is being issued in line 18, and see what happens.