Hi, does any one ever deal with the virtual addressing / memory with CUDA? I will need to deal with large files that couldn’t be fit in th global memory. Here is only post that I found vaguely talking about “virtual addressing” but I couldn’t find it anywhere in the programming guide. (see below) Thanks in Advance. http://forums.nvidia.com/index.php?showtop…tual+addressing
There’s no such thing as file mapping in CUDA. If you need to work with large files then you have to upload them partially.
Thanks AndreiB for the reply. When I read wumpus and vpodlozhnyuk ’ post (See link in the first post) , it gave me an impression that CUDA could do virtual Memory somehow - which could save me the trouble to do file mapping.
The Tesla architecture has a 64-bit MMU, and supports ‘virtual memory’ by mapping memory from the CPU into GPU memory space. But this does not include external hardware like harddrives.
I see. Totally make sense. :-) Thanks wumpus!