S39L Observation Thread

Elphidieus
Elphidieus
Joined: 20 Feb 05
Posts: 245
Credit: 20603702
RAC: 0

RE: RE: I'm now seeing

Message 26363 in response to message 26362

Quote:
Quote:
I'm now seeing completed quora for my S39L WUs. Claimed credits are peanuts. Actuals are typically 3.5 to 4 times the claims! I'm starting to feel guilty...
I run trux's calibrating client for this reason. Otherwise folks running the distributed ap will start to get caught in quora with folks running the optimized ap and be discouraged by the sudden drop in their credit. If you are interested, see truXoft Calibrating Client

I'm curious and confused actually..., I still fail to see what the calibration client really does even out the claims, when in fact the optimised client crunches WU in a far shorter amount of time, the overall RAC of the optimised clients is still theoretically, increases, relative to the quora of distributed clients...

unless this is made beneficial to those running multiple projects.....

archae86
archae86
Joined: 6 Dec 05
Posts: 3157
Credit: 7209794931
RAC: 936085

RE: I still fail to see

Message 26364 in response to message 26363

Quote:

I still fail to see what the calibration client really does even out the claims, when in fact the optimised client crunches WU in a far shorter amount of time, the overall RAC of the optimised clients is still theoretically, increases, relative to the quora of distributed clients...

unless this is made beneficial to those running multiple projects.....

When an infinitesimal number of folks are running low-claiming setups, their result is always the low claim in the quorum and gets discarded, so the effect is just selection bias--though still not zero. But when the fraction rises, others start noticing that low claims are dragging down their results.

This happened on SETI when P4 chips and others with better crunch relative to benchmark showed up. It got much more so with the "optimized" science applications. Based on posted comments, this effect directly influenced many decisions to leave the project.

The calibrating client is an attempt to claim in line with the useful science performed. While it is not entirely successful (for one thing, it takes about 30 results to train itself when started up fresh on a stable setup, and longer than that to react to a drastic shift such as the recent 4x improvement going for dist to S39L) but better than other solutions I know of.

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4527270
RAC: 0

RE: Incredible! Did you get

Message 26365 in response to message 26361

Quote:
Incredible! Did you get any feedback about this huge improvement?

Yes. Rytis wanted to forbid me because my Duron was faster than others.
His goal was to do a working BOINC application, not a faster one. :)

rbpeake
rbpeake
Joined: 18 Jan 05
Posts: 266
Credit: 1122127797
RAC: 747786

RE: Yes. Rytis wanted to

Message 26366 in response to message 26365

Quote:
Yes. Rytis wanted to forbid me because my Duron was faster than others.
His goal was to do a working BOINC application, not a faster one. :)


Your comment about this project being a planetheater is indeed an ironic one! The fossil fuel consumption necessary to run this project by powering all the computers crunching at less than maximum effiency will contribute to the greenhouse gases that are warming the planet..... :(

Crunchers For More Power
Crunchers For M...
Joined: 3 Aug 05
Posts: 69
Credit: 1071273
RAC: 0

RE: SETI has some very

Message 26367 in response to message 26351

Quote:

SETI has some very good optimised code.
I spent an afternoon to optimise it, and I could get just 10% improvement. (Crunch3r's Athlon XP SSE 2.09)

You mean, you had squeezed out another 10% out of a binary from Crunch3r? *wow*

Can I download that modified binary somewhere ;) ? Ten percent better is a good reason for me to try it :))

Maybe it's imaginable for you to work closely with Crunch3r to build a ultimate fast client for the upcoming Seti Enhanced ? :)

CU HiNuN

RandyC
RandyC
Joined: 18 Jan 05
Posts: 6572
Credit: 111139797
RAC: 0

RE: My AMD XP 2600+ has

Message 26368 in response to message 26360

Quote:

My AMD XP 2600+ has gone from ~4600 secs (S38) to ~4000 secs (S39L). I skipped S39 'cause the upgrades are coming too fast. :-P

Upgrading my XP 1600+ next.

Follow-up:

First 'pure' S39L WU for my XP 1600+ went from ~7550 sec. (S38) to 6627 sec. A 'mixed' s38/s39L WU ran 7070 sec.

Seti Classic Final Total: 11446 WU.

LiborA
LiborA
Joined: 8 Dec 05
Posts: 74
Credit: 337135
RAC: 0

Statisticks for CPU Pentium

Statisticks for CPU Pentium Xeon 2,4GHz (core Prestonia)

C37: Avarage time = 10125 sec (from 8 WU's)
S39: Average time = 5587 sec (from 12 WU's)
S39L: Average time = 5530 sec (from 12 WU's)

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4527270
RAC: 0

RE: RE: SETI has some

Message 26370 in response to message 26367

Quote:
Quote:
SETI has some very good optimised code.
I spent an afternoon to optimise it, and I could get just 10% improvement. (Crunch3r's Athlon XP SSE 2.09)

You mean, you had squeezed out another 10% out of a binary from Crunch3r? *wow*

Can I download that modified binary somewhere ;) ? Ten percent better is a good reason for me to try it :))

I upload to here. It was optimised to Duron (64+64kB cache), so don't rely on better time on other processors.
But, I think it will catch up the SSE3 code on A64 and P4.

AndyK
AndyK
Joined: 5 Jan 06
Posts: 21
Credit: 44767
RAC: 0

RE: RE: RE: SETI has

Message 26371 in response to message 26370

Quote:
Quote:
Quote:
SETI has some very good optimised code.
I spent an afternoon to optimise it, and I could get just 10% improvement. (Crunch3r's Athlon XP SSE 2.09)

You mean, you had squeezed out another 10% out of a binary from Crunch3r? *wow*

Can I download that modified binary somewhere ;) ? Ten percent better is a good reason for me to try it :))

I upload to here. It was optimised to Duron (64+64kB cache), so don't rely on better time on other processors.
But, I think it will catch up the SSE3 code on A64 and P4.

I tested it on my AMD XP 2100+, but it errored out within 2 seconds:
24.03.2006 20:32:09 Unrecoverable error for result 20dc98aa.10638.6706.36084.1.235_2 ( - exit code -1073741819 (0xc0000005))

Staying with crunch3rs opt. app. :-)

AndyK

Want to know your pending credit?

[img]http://tinyurl.com/438v3"[/img]
The biggest bug is sitting 10 inch in front of the screen.

Ed Parker
Ed Parker
Joined: 19 Feb 05
Posts: 11
Credit: 481732
RAC: 0

I guessing c39L only works on

Message 26372 in response to message 26370

I guessing c39L only works on certain computers? I have been running the c37 version on all four computers, today I saw c39L listed and thought I'd give it a try on a couple of units. Both are AMD K7's, 1.2g and 1.3g. After inserting the new albert, both started having "computation errors". So, I went back to c37 and all is well again.

For us bozo's that just install whatever is posted as the latest whizbang, maybe you should post some info as to what (in bozo terms) computers the program with operate with.

Comment viewing options

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