5.2.13 (Windows and Linux) just posted

ktaylor
ktaylor
Joined: 9 Dec 05
Posts: 1
Credit: 0
RAC: 0

Has anyone compiled CPU

Has anyone compiled CPU specific science packages for Einstein yet? If so, where can I find them? I'm running a CPU specific package from http://www.marisan.nl/seti/ for SETI that shaved about 30 minutes off of my crunch times for their WU's. Would be nice to see if there's something for Einstein going in since I'm just starting to pull work for this project.

Michael Roycraft
Michael Roycraft
Joined: 10 Mar 05
Posts: 846
Credit: 157,718
RAC: 0

RE: Has anyone compiled CPU

Message 20188 in response to message 20187

Quote:
Has anyone compiled CPU specific science packages for Einstein yet? If so, where can I find them? I'm running a CPU specific package from http://www.marisan.nl/seti/ for SETI that shaved about 30 minutes off of my crunch times for their WU's. Would be nice to see if there's something for Einstein going in since I'm just starting to pull work for this project.

ktaylor,

No, there are not and will be no CPU-specific (or custom-optimized) apps for Einstein. First, the code is not open-source. Secondly, and more important, Einstein calculations are several degrees more precise than Seti, so the sort of apps you speak of would result in many more calculation errors, invalidating and breaking results.

microcraft
"The arc of history is long, but it bends toward justice" - MLK

Tern
Tern
Joined: 27 Jul 05
Posts: 309
Credit: 68,331,610
RAC: 36,537

RE: No, there are not and

Message 20189 in response to message 20188

Quote:
No, there are not and will be no CPU-specific (or custom-optimized) apps for Einstein.

While Michael is 100% correct, I'll add to this so you don't get the wrong idea... speed _IS_ important to the Einstein project. They're just going about it a different way than SETI is. Where SETI throws it out to 3rd parties to have umpteen different versions, Einstein tries to do the same kind of optimizations on the _standard_ version. Being a Mac guy, I'll have to use the Mac side as an example, but you'll get the idea; for SETI, there is the "standard" app, that will run on G3/G4/G5 processors, but is pretty slow. There is an optimized app that will run on any of those processors, and is faster, but still not up to the level of the Windows and Linux optimized versions. Then there are separate Altivec-optimized versions for G4 and G5, with separate ones for OS X 10.3 or OS X 10.4, that are right up there with YAOSCW. That means there are six different Mac SETI apps to choose from, to get the one that runs best on your Mac. (How many are there for Windows now? 20? 30?)

Meanwhile, the Einstein app is a little bit bigger... but when it starts, it actually looks to see what kind of Mac CPU it's on, and runs the right code for that CPU. If you're on a G4, it's running the G4-optimized version. Same for OS versions. It may not be able to get that last few seconds of performance that "specialized" versions would get, but it does a darn good job, and it's under the control of the project, not open to the "this one only passes validation 10% of the time but it runs in 12 seconds on a 386!" issues that SETI sometimes has.

My understanding is that the Einstein Windows app is not yet optimized to the level of the Mac app, because they've had to concentrate on graphics issues instead... but I'm quite sure it'll come.

Paul D. Buck
Paul D. Buck
Joined: 17 Jan 05
Posts: 754
Credit: 5,385,205
RAC: 0

My G5 went from about 10

My G5 went from about 10 hours running an Einsein@Home model to about 5 hours.

I had been using a SETI@Home app that had optimizations in it for some time, but my run times were still way up there 9,000 to 11,000 seconds (3-4 hours). Now, I am down to 1 hour to 1 hour 20 minutes.

THis is the reason I think Steve Jobs has lost his mind going to Intel for processors. Many of the video processing applications the Mac is famous for use Altivec. They will be giving that up ... to my mind, regardless of the troubles with IBM not a smart move to go to same-ol same-ol ...

Now, going to AMD, there they would have had potential. Imagine Altivec enabled Athlons ...

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,353
Credit: 2,223,336,802
RAC: 2,749,301

RE: Now, going to AMD,

Message 20191 in response to message 20190

Quote:
Now, going to AMD, there they would have had potential. Imagine Altivec enabled Athlons ...

I'll confess to not being up on the details of the latest x86 architectures, has AMD managed to surpass Intel in vector processing (Altivec is the PPC implementation of it)? The last time I looked (pre A64) floating point vector related benchmarks were the one area where Intel was consistantly smoking AMD, conversely nonvector FP was where AMD was blowing Intel out of the water.

WHen I put my current system together I noticed there were still benchmarks where each architecture smoked the other, but didn't look closely at which were which. A top of the line intel mobo+ gig of ram cost $100 more than the AMD equivilant, and on my budget $300 AMDs outperformed $200 P4s almost across the board.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.