S5R3
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4312
Credit: 250233153
RAC: 35953
22 Sep 2007 16:06:15 UTC
Topic 193160
(moderation:
We generated the first Workunits of S5R3, some of you probably already got them.
The code of the Apps hasn't changed very much compared to the 4.4x Apps of S5R2. If you would run the same S5R2 workunit with both of them you'd find the S5R3 App being slightly faster.
As some paramaters of the Workunit split-up have changed, we needed to adjust our "credit normalization factor" (roughly credit per template). We may have to tune this again in the next few days. Please watch your credit per hour ratio and report here (and also take into account a normal fluctuation of +-5%).
BM
BM
S5R3
)
I have seen a few S5R3 complete recently. I am going by the app in the results indicating app 4.04 for S5R3. I am not really positively sure they are true S5R3 WUs.
Comparing WU CPU seconds to credit granted (on my computer# 997488) :
a) before S5R3 CPU sec / credit was around 112.4 (stock Mac Intel App)
b) 2 WUs with S5R3 CPU sec / credit is in the 135 range. (App 4.04)
For Windows, the picture
)
For Windows, the picture seems to be different:
I compared a few results for this host and credits/per hour changed from 18.7 to 18.5 , which is within natural fluctuation.
Anybody with Linux results?
CU
H-BE
RE: I compared a few
)
Thanks -- very helpful! Other users, please post your observations.
Bruce
Director, Einstein@Home
RE: RE: I compared a
)
Just so you know, people are posting their results on this thread as well, just so you don't miss this information. :)
RE: RE: I compared a
)
Please remember that the particular results that Bikeman is listing involve a comparison of the new S5R3 app with version 4.38 of the S5R2 app. Because of the overheads introduced during the extensive beta testing process, the 4.38 version was about 10 - 15% slower than the versions in place earlier in the S5R2 run.
By maintaining parity with 4.38 you are effectively building in a 15% pay cut over what was considered fair in the earlier stages of S5R2. This relates to Windows and I'm not sure how it affects other platforms.
With S5R3 looking to be quite a long run (probably around a year or more), I think it would be appropriate to adjust up to the previous "fair" rate of pay so as to give yourself the best possible chance of attracting as many as possible of those who place a high value on the "pay rate" :).
A rising "participation rate" and a steadily declining "time to completion" is always good for morale :).
Cheers,
Gary.
On FreeBSD@IntelPentiumD
)
On FreeBSD@IntelPentiumD 2.8GHz i've got a decrease of hour credit rate from ~12.12 to ~8.67: on S5R2 last large units spending 190,600-190,800 CPU secs gave 642.1 credits, now spending ~94,250 CPU secs is giving only 219.42 credits.
Computer - http://einsteinathome.org/host/907953/tasks
First AMD/Windows result will
)
First AMD/Windows result will be in in an hour. It doesn't look slow to me but I'll have to look at some more data to prove it ;-)
Bernd or Bruce: Would you like some more Linux results? It would be no problem to let my (Core Duo) laptop crunch under Linux for a while. That would mean Ubuntu Feisty with one of the latest kernels, so, the results should be quite comparable.
Just looked at this
)
Just looked at this AMD/Windows result and it's absolutely pathetic (sorry to put it in such a drastic way but the effect is really kinda dramatic)... 11,7 creds/hour on a 3500+. I'm crunching WU Nr. 2 now to see if it was maybe something that happens only once, and I'll check the debug output for trouble, but the first impression is definitely not so hot. This was an 18 creds/hour box under Linux in S5R2, and about 17 with the later Windows app. That would mean a rough 30% drop in credits (the results from S5R2 don't show up online any more since this box had a break over the summer but I kept some charts).
No idea what's wrong, my box not getting along with the science app, maybe AMD CPUs having trouble again, or simply different credit calculation. But it's not looking good so far. My Core Duo can do more in terms of credit/hour (Bikeman will confirm those are 15 cred/hour boxes) and one core of that benchmarks significantly lower. 11 or 12 is about what I could expect from my old Celeron on a good day.
So, to make this extra clear: I'll definitely stick with the project, I don't mean this as a complaint and I'm not blaming anyone. It's just that we should get back on our performance and mine is really not so hot, and I hope there's an explanation or a way to fix it.
Cheers,
Annika
Hi! Hmm...we'll see, your
)
Hi!
Hmm...we'll see, your wingman has an AMD Athlon 64 X2 Dual Core Processor 4000+ under Windows, I'm curious how long that machine will take. Once I get S5R3 units on a dual-boot PC, I'll try to lket one unit crunch under Windows and then under Linux. That would be a Pentium M, tho.
Maybe some WUs need a credits re-calibration. Your WU, Annika, was on the very low end of the frequency band, for example. I've seen others where the credits / h stats matched within only a few %, so this might be a problem with some frequencies only.
CU
H-B
Yeah, it might be the
)
Yeah, it might be the frequency band.
No idea yet, but my box is crunching away happily, so there'll soon be more results to compare.
It can hardly be a software issue, everything is rather vanilla-like... freshly installed Windows XP, latest BOINC core client from the official download area, no tweaking tools except AsusProbe (which only does monitoring) and Cool and Quiet is off... benchmarks are perfectly okay... you get the picture.
I do hope it's not AMD-related again, since all the boxes that have been reported to show encouraging results seem to have been Intels.