Device to device copy = SLI copy? SLI copy feature? when?

Hi everybody,

I was porting an application from CUDA 0.8 to 1.0 and I was wondering:

The “Asynchronous device to device memory copy” feature that appears in the Release Notes is not yet a “Card to Card memory copy” (aka. memory copying through SLI), is it? I cannot find any reference in the documentation, so I guess it is talking about copying inside the same device :(

Any update about this feature? do I remember reading somewhere that Nvidia was working on it? It could be a very nice thing to have…

Cheerio ^_^

Yes, the device to device memory copy in the documentation refers to copies within a single GPU.

There is currently no support for fast copies between different cards, but it’s something we’re working on.

Are you still working on it ?