It's what we use in the D3D12 backend. Ideally, implement https://github.com/Traverse-Research/gpu-allocator/issues/235 before doing the switch to avoid changing the behavior too much.