Is a Quadro or Tesla GPU necessary?

[sixsamuraisoldier]

Would it work with GTX GPUs?
If not, what is the reason why it won't work with GTX GPUs?

[sixsamuraisoldier]

The GTX Titan X can be run in TCC mode which should support Peer to Peer transfer. Could it be possible to utilize this feature rather than GPUDirect RDMA?

[oscarbg]

also interested to know if even Geforce can support that at least in Linux world..

[kaigai]

This work stands on NVIDIA's GPUDirect RDMA.
They said, it works on Tesla and Quadro GPUs.
http://docs.nvidia.com/cuda/gpudirect-rdma/

In my test, the feature is workable on Quadro K1200, Tesla K20c and Tesla M40.
On the other hands, GTX980 didn't work; although it is never low-end GTX.

I don't know the status of TITAN X in Pascal generation, because its brand is not GTX no longer but not either of Tesla or Quadro. I also want to know the status.

[dinedal]

I have a TITAN X. How can I test this?

[kaigai]

No. It is Tesla or Quadro only feature.

https://github.com/pg-strom/devel/wiki/GPU-Availability-Matrix