cudaHostAlloc limitations in WSL2

When running existing code, that works fine on non-WSL based linux installs. I am running into problems with cudaHostAlloc. It seems the limits placed on the amount of memory that can be allocated this way are significantly tighter in WSL2 (in WSL cudaHostAlloc starts failing at roughly 300MB, whereas on a non-WSL based linux multiple GB are no issue). Are there any settings related to this or is this a matter of the overall current state of cuda support in WSL2?


Hello and thanks for reaching out,

Are you speaking 300MB at once (single call to cudaHostAlloc) or multiple allocation that amount to 300MB ?