So I’m trying to using
cudaMallocManaged to allocated 12GB of memory on an 8GB card. This works just fine on Linux with an RTX2060 Super, but on Windows 10 with an RTX2080 Super it doesn’t work and throws an “Out of memory” error. When allocating less than 8GB it works fine, but it seems the whole ‘transcend GPU limits’ thing that Unified Memory is all about doesn’t work on Windows 10. And I assume I’ll be unable to use WSL/WSL2 or a VM to get past this. And thanks to switching to team Ryzen I won’t be able to run Linux on this machine because there is no integrated graphics.
Are there any flags I can set or options I can use to get the Linux behaviour under windows?