Turning on (most) virtualization instructions have limited impact. NVIDIA GDS (GPU Direct Storage) is a feature which (I believe) enables the DMA engine on your processor to transfer memory from your NVME to your GPU, and possibly write to in-box NVME, or Nvme-over-fabric
In any event, the need for GDS should be record so it isn't lost, but it does not block (I believe).
If you run deviceTransferTest from the NVIDIA CUDA samples, one of the tests will fail if the kernel boot parameters are incorrect. GDS requires the ability to do RDMA, and GDS enables internal RDMA capability
This is mostly a duplicate of https://github.com/artificialwisdomai/origin/issues/111, however one is metal specific and the other is virtual machine specific.
Turning on (most) virtualization instructions have limited impact. NVIDIA GDS (GPU Direct Storage) is a feature which (I believe) enables the DMA engine on your processor to transfer memory from your NVME to your GPU, and possibly write to in-box NVME, or Nvme-over-fabric
In any event, the need for GDS should be record so it isn't lost, but it does not block (I believe).
If you run
deviceTransferTest
from the NVIDIA CUDA samples, one of the tests will fail if the kernel boot parameters are incorrect. GDS requires the ability to do RDMA, and GDS enables internal RDMA capability