GPUDirect RDMA with Geforce GTX 10xx?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi, I want to implement a direct data transfer from our PCI-Express card to a GPU. I am a bit confused about the supported devices for GPUDirect RDMA. I want to use a Geforce GTX 1060. In this document (http://docs.nvidia.com/cuda/gpudirect-rdma/index.html#overview) I read that I need “Kepler-class GPUs and CUDA 5.0”. Later on in the same document I find “GPUDirect RDMA is available on both Tesla and Quadro GPUs.”. According to this page (CUDA GPUs - Compute Capability | NVIDIA Developer) the GeForce GTX 10xx support CUDA 6.1.

Does that mean I can use the GeForce GTX 10xx cards for GPUDirect RDMA? Or do I need a Quadro or Tesla Card?

Thanks,

Ginsengelf

2

You need a tesla or quadro card.

[url]https://devtalk.nvidia.com/default/topic/716091/geforce-gtx-780-and-gpudirect-rdma/[/url]

[url]https://www.microway.com/knowledge-center-articles/comparison-of-nvidia-geforce-gpus-and-nvidia-tesla-gpus/[/url]

3

As far as I know, the statements quoted from NVIDIA’s documentation are both correct, and are orthogonal to each other. Each states a necessary but not sufficient condition.

4

The microway comparison quotes that Titan X features Dual DMA engines. Does this mean it is suitable in a cluster environment to share GPU VRAM across nodes? Is this supported in Titan V?