SuperPi on GPU, were going CUDA

**Don_Dan** · 10-02-2008, 12:25 PM

Just wanted to share my 8800 GT screen:

**sjbuck** · 11-07-2008, 09:58 AM

Whatever happened to superpi - cuda?

**[XC] gomeler** · 11-07-2008, 12:45 PM

Stillborn

**crazyheinz** · 11-12-2008, 02:14 AM

what about this program?

**Vatos_locos** · 11-27-2008, 07:37 AM

what happened is this going out at all ?

**Warboy** · 12-23-2008, 08:49 AM

Yea what did happen to this?

**Calmatory** · 12-23-2008, 06:40 PM

Probably noticed the poor performance gains due to poorly threadable algorithm used. CUDA is just like finding a needle from haystack, a burden that no one really wants to stick to unless they really have to.

**BenchZowner** · 12-24-2008, 04:25 AM

@warboy:
As far as I know, they stopped because it wasn't an easy task and proved to be time consuming.
@calmatory
Not exactly.
But proper parallel coding isn't the easiest thing on earth...

**MarcusFoX** · 12-24-2008, 09:20 AM

that cuda-z program looks pritty sweet.. but why did that drop the whole superpi cuda thing...

**Serotoninn** · 04-26-2009, 04:10 PM

And who's idea was this? Because somebody didn't do their homework...

Pi was never an easy thing to multi-thread... And only now do we finally have a multi-threaded pi program for multiple cores - let alone CUDA...

http://www.xtremesystems.org/forums/...d.php?t=221773

p.s.

Originally Posted by dinos22

i look forward to doing a 20 second 32M SuperPi run

eek

more like 50m in 20 seconds

**FlawleZ** · 04-26-2009, 04:46 PM

Originally Posted by Serotoninn

And who's idea was this? Because somebody didn't do their homework...

Pi was never an easy thing to multi-thread... And only now do we finally have a multi-threaded pi program for multiple cores - let alone CUDA...

http://www.xtremesystems.org/forums/...d.php?t=221773

p.s.

more like 50m in 20 seconds

I believe he was implying 20s 32M on one core. Definitely a dead thread resurrected.

**poke349** · 04-26-2009, 05:09 PM

Interesting old thread... I can only laugh at the idea... because the math needed for such a thing isn't quite there yet...

Originally Posted by FlawleZ

I believe he was implying 20s 32M on one core. Definitely a dead thread resurrected.

I don't think that's what he meant. The power of the GPU is that they have MANY cores... and that you'll need to use them to get a 20s 32M time.

And yes... This new multithreaded pi program IS able to compute 50 million digits in 20 seconds.

50,000,000 digits:
19.3864 - Movieman - Dual 2.93GHz Xeon Gainestown
20.553 - spdycpu - Core i7 920 @ 4.3GHz (21 * 205)
20.7997 - poke349 - Dual 3.2GHz Xeon Harpertown
21.7047 - tet5uo - 3.20GHz Core i7 @ 4.00 GHz
25.774 - Serotoninn - 2.66GHz Core i7 @ 3.20GHz

I'll re-post the link to the thread:
http://www.xtremesystems.org/forums/...d.php?t=221773

**ShtopoRrr** · 04-27-2009, 04:40 AM

Sorry, I lazy to read whole thread.
Is there any cuda/openCL programmers? Can you port SuperPi to gpgpu, without adding multithreading? Using only 1 shader. Just imagine: very low-cost 8-sp Graphic card making WR!! I want to see OpenCL-PI among HWBOT benchmarks!

Sorry for my bad English

**Tuvok-LuR-** · 04-27-2009, 04:42 AM

Why should single threaded PI on gpu beat a cpu?

**trinibwoy** · 04-27-2009, 04:42 AM

What would be the point running a serial program on a GPU? The problem isn't cuda or opencl. The problem is that either the math isn't parallelizable or people aren't smart enough (yet) to make it so.

**kromosto** · 04-27-2009, 04:51 AM

Originally Posted by ShtopoRrr

Sorry, I lazy to read whole thread.
Is there any cuda/openCL programmers? Can you port SuperPi to gpgpu, without adding multithreading? Using only 1 shader. Just imagine: very low-cost 8-sp Graphic card making WR!! I want to see OpenCL-PI among HWBOT benchmarks!

Sorry for my bad English

i think it is nice idea when i found time i will implement this, dont know will somebody use it or not but can give it a chance.

**ShtopoRrr** · 04-27-2009, 04:55 AM

It isn't good to mix CPU & GPU results. I meant that low-cost GPU beat hi-end one. And of course, it is very difficult to parallelize PI calculation, so I proposed idea of single-threaded PI program.

kromosto, good luck, i want to use this program

**poke349** · 04-27-2009, 05:48 AM

Originally Posted by trinibwoy

What would be the point running a serial program on a GPU? The problem isn't cuda or opencl. The problem is that either the math isn't parallelizable or people aren't smart enough (yet) to make it so.

This is correct.

The math IS parallelizable. And the proof is here (already posted):

http://www.xtremesystems.org/forums/...d.php?t=221773

HOWEVER,

It is very difficult to do. It was hard enough to parallel it into several threads for several cores. So it will be MUCH harder to parallel it into hundreds of threads for a GPU...

Also GPUs right now don't have the right instruction set to be efficient with this kind of computation. (They are too specific for graphics.) So I won't be surprised if even the best of GPU implementations have trouble beating a CPU implementation.

**ShtopoRrr** · 04-30-2009, 08:37 PM

And what about wPrime on GPU? Multithreaded of course

**ShtopoRrr** · 06-18-2009, 12:45 PM

Any updates?

**haylui** · 06-18-2009, 10:03 PM

starting value for benching Pi is ?
512M Pi? 1024M Pi?

**ShtopoRrr** · 06-19-2009, 04:39 AM

It is good to be like CPU SuperPi, for single-threaded GPU SuperPi, I think

**K404** · 06-19-2009, 05:05 AM

Photoshop on GPU?

Video encoding?

Theres important things this can be used for than SPi.

**ShtopoRrr** · 06-19-2009, 06:30 AM

I dont think so. SuperPi is a main and simple calculation benchmark, a Standard. imho

**K404** · 06-19-2009, 07:50 AM

But only benchers care about it. For CUDA to develop it has to have real-world uses that developers can score funding to code for and turn heads to CUDA that way.

Thread: SuperPi on GPU, were going CUDA

Thread Tools

Search Thread

Rate This Thread

Display

Bookmarks

Bookmarks

Posting Permissions