Technically yes, but WSL2 is remarkably close to optimal in terms of throughput. Unlike WSL1 (a type 2 hypervisor), WSL2 requires Hyper-V (a type 1 hypervisor), meaning Windows also runs as a VM once it’s enabled. The Linux vGPU driver still needs to go through the Windows Nvidia driver as far as I know, but that is seldom the bottleneck for CUDA applications.