Comment by PunchyHamster

We need low power but high PCIE lane count CPUs for that. Just purely for shoving models from NVMe to GPU