faster copying to gpu

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello all,

Transferring data from the CPU to the GPU has been the most time consuming process for some of my image processing work with CUDA. I was wondering if there were any good ways to get data straight from disk (or firewire) to GPU. I have heard the term “DMA” tossed around, but am not really sure what it is or if it applies. Any thoughts?

Thanks,

Dave

2

You could use pinned memory on the host ( see the bandwidth example) to improve the transfer rate.
There is no way to get data straight from disk or firewire.