Device to Device copy How to copy from Device to Device

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi

I have a 9800 GX2 and I need access to the whole 1GB of global mem (across both cards). If I run my kernel on one device, can I copy the output to the memory of the second device within the kernel? I have a huge image stack that I need to process and would like to not overwrite the input image. If this is possible, what is the function call to use within the kernel?

Thanks

2

The two cards have separate memory pools that cannot be shared in this fashion.

3

to my understanding he knows they have 2 seperate memory pools.

the way i read his question is can card #2 be uses as +512 to card#1`s 512 (like a cache or a storage)

isnt this how it works right now but usually the cache/storage is on the HD or system ram?

4

Basically I need access to the 1GB of global memory that is divided across 2 cards. If this isn’t possible is there a card that has 1GB of global memory for only 1 device?

5

Tesla C870, Tesla C1060, or GTX 280 (although you won’t be able to access the entire 1GB on that card, probably 960-980 megs)