advice needed by a PhD student

Lev · November 17, 2011, 9:41am

I agree, but it is hard to make such a program without learning cuda,there could be a lot difficult moments.

pasoleatis · November 17, 2011, 9:58am

Well, for some applications, you want professional programmers, but “amateurs” can get useful work done with CUDA. I learned CUDA programming as a Ph.D. physics student, and I’ve taught CUDA programming to undergraduate and graduate physics students. We’re not implementing really novel parallel algorithms, but it still helps us get work done faster than a pure CPU implementation.

Actually, I find that CUDA makes it easier for inexperienced programmers to make efficient code. I seldom get anywhere near peak efficiency on a CPU, but I can often get within a factor of 2 or 3 of the maximum throughput (either memory or compute) on a CUDA device.

To be successful with CUDA as a non-professional, you do need to be selective about how you apply the technology. Not all problems have an easy solution with CUDA, and you have to do a little cost-benefit analysis to decide where your development effort is best allocated. In the 4 years I’ve been using CUDA, I think I have only written 5 non-trivial programs with it. These five programs are very useful to me, and I use at least one of them every day, but compared to all the code I’ve written over that time, CUDA is not the dominant component.

I am one of the lucky ones. My problems relies heavy on FFT I ported very easy my codes from FORTRAN to CUDA and I got 50-80 times spped-up comapred to single core cpu codes.

Lev · November 17, 2011, 10:10am

Do you use nvidia fft?

pasoleatis · November 17, 2011, 11:30am

Yes. I use cufft, real to complex and back inplace transforms. I only spend 2 weeks to port my code from fortran. for the cpu version I used fftw.

esem · November 30, 2011, 11:07am

Hi, pasoleatis,

how did you get such a speed-up? I compared CUFFT (CUDA 4.0, Tesla C2050) and MKL libraries for 2D Real to Complex FFT, and got max 15 times speed up. (I measeured only the time needed to compute the FFT without data transfer)

For which sizes did you get 50-80 times speed-up?

thanks, esem

pasoleatis · November 30, 2011, 3:49pm

Hello,

I do not have access to MKL libraries. I compared with a single core runs on clusters composed of AMD processors. It depends on what you compare with. I compare it to my usual run.

esem · December 4, 2011, 9:20pm

Can you please give the timing you get on GPU for some FFTs? For example, how long does it take to compute FFT for 256 x 256 matrix Real to Complex on both CPU and GPU? 512 x 512?

It is very important for me to understand whether I achieved the peak performance.

Regards, esem