Can some1 explain what this GPUDirect technology? Is it same as the one from Mellanox (shared pinned mem?)?
Thanks for any info,
We’re working on a white-paper about this. Yes, it’s basically pinned memory shared between CUDA and a device driver.
Oh… It mimimizes the trip back to a CPU register…! Thanks!
Good alternative! But it would be better to find a way to directly DMA via PCIe to the GPU (without RAM coming into picture)