Message boards :
Number crunching :
GPU crunching
Message board moderation
Author | Message |
---|---|
Hans Dorn Send message Joined: 3 Apr 99 Posts: 2262 Credit: 26,448,570 RAC: 0 |
Hi all, Like a few of you already know, I'm currently working on a GPU port of the seti app, targeted at nVidias 8800 series of graphics cards. I've already ported the 2 most trivial crunching steps - chirping and calculating the power spectrum - and created a little test app. The app just generates some text output, it is _no_ replacement for the seti app. If you have an 8800 card, it would be nice if you'd do a test run. The app, and sources, are available at Anton's site The whole port is pretty much a work in progress. Comments and suggestions are very welcome. Regards Hans |
KWSN - Chicken of Angnor Send message Joined: 9 Jul 99 Posts: 1199 Credit: 6,615,780 RAC: 0 |
Great to see you making progress, Hans! Any news on the Linux front? Regards, Simon. P.S.: I see Anton's site is using Enigma, a portal I co-develop :o) I may be able to help with the site theme if necessary. Donate to SETI@Home via PayPal! Optimized SETI@Home apps + Information |
Hans Dorn Send message Joined: 3 Apr 99 Posts: 2262 Credit: 26,448,570 RAC: 0 |
Great to see you making progress, Hans! I'm trying to wrap my mind around the FFT ATM, but making good progress - at least I hope so :o) I should have something presentable by the end of next week.... Regards Hans |
ML1 Send message Joined: 25 Nov 01 Posts: 20289 Credit: 7,508,002 RAC: 20 |
This is sounding rather good. I'll hold out on the next upgrade a short while to go with an 8800 series GPU or better... whichever your code fits. Good luck with the code! Regards, Martin (Stuck elsewhere in EDF :-( ) See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
ML1 Send message Joined: 25 Nov 01 Posts: 20289 Credit: 7,508,002 RAC: 20 |
Random thought: What happened to Eric K's work on doing this for the previous vector-based GPUs? And would that be of any help? Regards, Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
Hans Dorn Send message Joined: 3 Apr 99 Posts: 2262 Credit: 26,448,570 RAC: 0 |
Random thought: I've tried reading the sources of GPUFFTW, but it gave me a headache. Crunching via OpenGL apparently is not for the faint of heart :o) There's a tiny chance that we can backport some of the code to CG, but I'm not positive about that yet. Regards Hans |
ML1 Send message Joined: 25 Nov 01 Posts: 20289 Credit: 7,508,002 RAC: 20 |
I've tried reading the sources of GPUFFTW, but it gave me a headache. You have my sympathies. The first "F" in "FFT" gave me butterflies! Crunching via OpenGL apparently is not for the faint of heart :o) Interesting. Would that code still work as intended despite the very different GPU architecture? I guess you at least get the advantage of the faster PCI-E transfers. Good luck for the coding! Regards, Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
michael37 Send message Joined: 23 Jul 99 Posts: 311 Credit: 6,955,447 RAC: 0 |
I've tried reading the sources of GPUFFTW, but it gave me a headache. User Benher claimed here that there exists open source for FFT in the GPU. Is that the GPUFFTW that you mentioned? |
Hans Dorn Send message Joined: 3 Apr 99 Posts: 2262 Credit: 26,448,570 RAC: 0 |
Yep. It's one of them, reportedly the fastest. You can download a copy from here after filling out a really short form :o) Both windows and linux are supported. I tried using it for seti, but could get no speed improvement, mainly because of the overhead introduced by transferring and rearranging data. Regards Hans |
Andy Lee Robinson Send message Joined: 8 Dec 05 Posts: 630 Credit: 59,973,836 RAC: 0 |
You have my sympathies. The first "F" in "FFT" gave me butterflies! Hahahaha! Very good! :-) |
N/A Send message Joined: 18 May 01 Posts: 3718 Credit: 93,649 RAC: 0 |
Well done! I know that this'll sound like nit-picking or down-putting, but please don't forget the Mac OS (especially PowerPC). Keep up the good work! |
michael37 Send message Joined: 23 Jul 99 Posts: 311 Credit: 6,955,447 RAC: 0 |
This is BRUTAL -- the computations are based on OpenGL. Is there an easier approach using the mathematical routines in CUDA? I realize you can't disclose much 'cause of NDA required to use CUDA, but some general direction would be great. |
citroja Send message Joined: 12 Dec 03 Posts: 192 Credit: 3,245,701 RAC: 0 |
quite interesting to say the least....I have two questions: 1) With the possibilities of dual cards in a system can/will that cause issues? 2) When/if GPU crunching is complete will there be a version for "older" cards (i.e. 7600, 7800, 7900)???? -citroja |
Eric Korpela Send message Joined: 3 Apr 99 Posts: 1382 Credit: 54,506,847 RAC: 60 |
GPU crunching might only be practical on PCI Express based cards because AGP has low GPU to main memory bandwidth. Eric quite interesting to say the least....I have two questions: @SETIEric@qoto.org (Mastodon) |
KWSN - Chicken of Angnor Send message Joined: 9 Jul 99 Posts: 1199 Credit: 6,615,780 RAC: 0 |
Hi, I've stumbled upon a site that may have an interesting product to help in developing for massively parallel architectures like new GPUs and the Cell BE (read: PS3). The site is http://www.rapidmind.net. I've applied for and received an evaluation license from them. The basic idea behind this product is that you can write your code in "normal" C++ and have the same code work on any supported architecture! That is to say, you write one codebase, do not have to use specialized architectural tools or languages, do not have to learn all sorts of platform-specific concepts you don't really need, ... Supported platforms: all ATI Radeon 1x00s, nVidia GeForce 6xxx/7xxx (and I assume, 8xxx as well) GPUs as well as Cell BE-based systems. Now, I haven't received a price quote yet, so I don't exactly know how expensive their product is (though I will find out). If it's less than 1000 bucks, it may be worth it (though I'd definitely like to test/benchmark before I spend any money). In any case, even just using the Eval version may give people working on these implementations some pointers (checking out what RapidMind's code does vs. their own for the same things) to get started. Of course, hand-tuned code may still be quicker than what their product comes up with - but contrast that with the development time delta, and the picture may change. Regards, Simon. Donate to SETI@Home via PayPal! Optimized SETI@Home apps + Information |
citroja Send message Joined: 12 Dec 03 Posts: 192 Credit: 3,245,701 RAC: 0 |
GPU crunching might only be practical on PCI Express based cards because AGP has low GPU to main memory bandwidth. The 7600s, 7800s, and 7900s all have PCIe versions (to my knowledge) I have a 7800 GTX and that was why I was curious. This may be important as the price of the 7xxx series starts to drop as the 8xxxx units begin to become popular. -citroja |
Hans Dorn Send message Joined: 3 Apr 99 Posts: 2262 Credit: 26,448,570 RAC: 0 |
Hi Simon, interesting pointer, thanks! BTW I'm having a hard time currently with getting my "hand-tuned" stuff to a speed level that's worth mentioning :o) Regards Hans |
Misfit Send message Joined: 21 Jun 01 Posts: 21804 Credit: 2,815,091 RAC: 0 |
More good work from our new devs. ;) me@rescam.org |
citroja Send message Joined: 12 Dec 03 Posts: 192 Credit: 3,245,701 RAC: 0 |
If there anyone wants/needs to test on an nvidia 7800 series card let me know. (still working on the card compatibility issues) -citroja |
MAX3400 Send message Joined: 4 May 00 Posts: 2 Credit: 1,502,870 RAC: 0 |
If there anyone wants/needs to test on an nvidia 7800 series card let me know. He, probably my first post on this forum but since my 7800GTX isn't doing anyting at all 99% of the time, I would like to help out with testing. Is there any info on performance after OC'ing the GPU? |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.