NVIDIA remote gpu access

Accelerated Computing CUDA CUDA Programming and Performance
1

Dear CUDA members,

I’m writing this message because I wonder if NVIDIA can provide to the whole CUDA community a service that we can access GPUs in order to test our applications (Like Nokia with their mobiles). Many of our work we can only test in our devices, and there is a need to know how our work would perform in different GPUs.

Another approach are people who would like to start working/learning CUDA and they have no CUDA-GPU, the only solution is emulation (far away for real results/performance).

This could be attached with nSight which is designed in a remote-programming fashion (as far as I know, I need a better card to install/try it).

What you readers of this message think about this? would you support this idea?

Cheers!

2

There already exist several cluster-on-demand offerings, that include access to GPU units attached to cluster nodes, I’ve somewhat discussed some of these in this thread. So wouldn’t expect NVIDIA directly providing this type of service.

3

It would be interesting to have a service that could give me performance estimates from a wide range of cards… Sure cuda apps scale very well but that doesn’t let you lay claim to what performance to be expected on a specific card.

For me, just some service where i could schedule runs on a specific plattform and recieve the output some time later would be great. No need for instant access in my case…

Free would be great :)