Cuda 6

[3LP]

http://www.cgrecord.net/2013/11/nvidia-cuda-6.html

I guess when they say "Unified Memory" it doesn't mean that the GPU is now capable of using the system ram as part of their ram, or does it?
Dreaming dreaming

Vlado, any insane news in this release that will change everything, again?

Stan

[ivaylo.ivanov]

"Simplifies programming by enabling applications to access CPU and GPU memory without the need to manually copy data from one to the other"
I.e. nothing new, just simplification for new CUDA developers . At least this is how I understand it...

[vlado]

I guess when they say "Unified Memory" it doesn't mean that the GPU is now capable of using the system ram as part of their ram, or does it? Dreaming dreaming

No, it doesn't. As Ivaylo noted, the additions in CUDA 6 are purely for the convenience of programmers.

Vlado, any insane news in this release that will change everything, again?

Nope. You will have to wait for the next-gen nVidia architecture for that.

Best regards,
Vlado

[3LP]

Nice!

Next gen from NVidia will be a major release then, but I wonder if they will not kill their tesla/quadro department if they allow real unified/stacking memory...

ie : just put 6 titan and have 36Go of ram for GPUs...

Thanks for the info guys

Stan

[vlado]

I'm sure there will be some performance penalty for paging to the system RAM...

Best regards,
Vlado

[3LP]

I'm sure too, but even if it was already possible to stack gpu ram only, it would be great.
And performance penalty can always be considered if you know how much and why and make a concient decision based on that.

Stan

[Dariusz Makowski (Dadal)]

and Im sure it will be at the price of powerfull PC so it wont be worth it at the end