stephenbrooks.orgForumMuon1GeneralState of GPU acceleration?
Username: Password:
Search site:
Subscribe to thread via RSS
2012-07-26 05:46:01
Hello all,

It's been 10 years, but I'm back.  I just looked at my profile here in the forum and it turns out I posted 8 times in 2002. I also seem to remember that I had the highest result for version 1 back then, so I'm special.  I don't suppose those stats are around somewhere, because it'd be interesting to see?

I'm curious to know if there's any work being done in the direction of graphics card acceleration via OpenCL for the project.  I imagine that in the interest of accuracy, double-precision would be required like it is for Milkyway@home.  But even though my current processor runs around 25 times faster than the processor I was using back in the day (Ivy Bridge @ 4.2 compared to an Athlon 700), it can't hold a candle to the ~teraflop my graphics card can do in double-precision.

Anyway, lovely to be back. 
2012-07-31 19:26:36
Muon1's always been double-precision, AFAIK.

here's the link to the stats for v1 to v3.11

As for gpu processing, I've no idea.

And welcome back.

Stephen Brooks
2012-08-01 14:46:47
Are you sure your graphics card can do a teraflop in *double* precision?  A lot of current cards can do a lot in single but the double precision is only a fraction (1/4 or 1/8 or so).  Unless you have a very expensive card of course...

I've got some simple things working in the GPU, but Muon1 is rather harder because it has to solve a problem with a generic configuration of accelerator components (I'd have to get a fairly complicated memory structure describing the accelerator from RAM onto the GPU).  It's possible, although another problem is my GPU on my workstation here doesn't support double precision (I've tried!)

Anyway, welcome back!  I'm glad I kept those version 1 stats pages up now
2012-08-02 09:12:44
Awesome!  Those stats are a definite trip down memory lane.  And at 7.066%, it looks like I did indeed hold the lead then. 

My card is a Radeon 7970, which is rated at around 940 gflops dual precision.  It's also at around a 20% over clock, and is a very fast card.  I think single-precision calculations are in the range of 4 tflops, which is far beyond any games I'm currently running.  So I know at least that the Radeons are in that range, even if Nvidia really capped their non-workstation cards.

If I had bought a second for a multi-GPU setup I'd think about sending it your way; as it stands, I'm happy to be an alpha Guinea Pig if it comes up, and if there were to be a fundraiser to get you a decent card for testing I'd chip in a bit.  As far as I can tell it's a cool project doing some awesome science.  When I'm back from vacation I'll throw a few more Ivy Bridge machines into the mix. 
Stephen Brooks
2012-08-02 20:04:29
For now, you can run Muon1 on the CPU and I believe there are some other projects around that make good use of the GPU.
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel RSS feed

Site has had 21696461 accesses.