stephenbrooks.orgForumMuon1Generalpeakstreaminc.com
Username: Password:
Search site:
Subscribe to thread via RSS
Stephen Brooks
2006-10-12 17:09:54
Don't worry, I'm keeping tabs on it.  I want double-precision (64 bit) floating point, though, so we might have to wait for 2008's GPUs.
BOHICASETI
2006-10-13 03:19:50
WOW!  Just watched the demo video they had.  Seems to be a serious jump in programming technologies.  I just got my dual Opteron in today, I'm ready when you are.  hehe
BOHICASETI
2007-01-13 06:35:16
I decided to check out their web site again and watched their demo link and now it says that it supports single AND Double-precision float point operations.  Just thought that I'd throw that out there.  Might be worth another visit to see if it's another possibility.
Stephen Brooks
2007-02-19 12:16:27
Their interface supports it, but does the graphics card?  If not, it will just emulate the double precision in software.
waffleironhead
2007-04-17 18:57:48
http://www.peakstreaminc.com/product/free_trial_download.php

they now offer a free evaluation.
waffleironhead
2007-04-17 19:02:38

"When do we get double precision support?  Is it in the pipeline?
PeakStream supports double precision today.  The PeakStream Platform executes all double precision calculations on the CPU, so mixed double/single precision code can be written with the PeakStream API.  Graphics vendors and the Cell consortium have gradually increased their precision support over time, and we expect double precision support in GPUs and Cell processors in the future.  "


Stephen Brooks
2007-04-18 13:06:05
Thanks.  Yes, "all double precision calculations on the CPU" with a C not a G, so no speed-up for double precision code right now.
Stephen Brooks
2007-04-20 01:32:52
Heh, even the Inquirer have picked up on this
[TA]z
2007-04-24 00:28:48
I read a short paper on accelerating double precision FEM sims with GPUs from and from what I remember it seemed to be pretty successful.  It basically implemented a precision defect correction algorithm, which maintained precision and more than doubled speed.
Stephen Brooks
2007-04-27 12:41:24
Hmm I can see how that would work if you're doing a matrix solve (you can work out the residuals quickly on the CPU in double precision but do the time-consuming iterative part, at two or more scales of single precision, on the GPU), but I'm not sure it's very easy to apply it to particle-pushing as in Muon1.
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel charactermstdn.io RSS feed

Site has had 25159693 accesses.