TAS made an impact in Einstein@Home in April:

[petrusbroder]

The preliminary stats for TAS' april efforts are in.

Please check for the stats in TAS' forum Stats and Milestones: TAS E@H

A summary:

1. TA overtook two teams (AMD Users and Overclockers.com) on the last day of the project.
2. TA was overtaken by one team (Off-Topic) which has more than twice as many crunchers compared to TA.
3. TA left the Knights Who Say Ni in the dust: April 1st they were within 17 000 credits och overtaking TA, May 1 st they are 238 260 credits behind TA. No overtaking within the next month unless they produce at least 7950 credits/day more than TA (which they never have done in the past ... the future I do not know ...

Some weeks ago I predicted ...
... that TA will reach a total of 1 090 000 credits in april. TA reached 1 126 398 credits (+3.33%).
... that TAS will have produced 267 500 credits in april. TAS reached 272 627 credits (+1.91%).
... that TAS will reach a total of 332 000 credits in April. TAS reached 333 329 credits (+0.4%).
... that TAS total would comprise some 30% of TA's total credits. TAS reached 29.6%. (-1,33%)

I have the very strong impression, that the TAS effort has inspired at least some other TA-members to crunch at Einstein@home, because we observed, that the number of TA's credits increased more than the number which could be attributed to TAS and the "usual" production of TA.

Why not check out the E@H teams page? Just for fun... or TA's stat page...

Thanks all TAS-members for a super great month of crunching, which really har helped TA and which gave us -TAS - a lot of fun. You can also the fun @ TAS and crunch our next project ...

 

[GLeeM]

Thanks Petrus!

Those were some amazing predictions you made, and how close you guessed.

Howdidyoudothat?

 

[Rattledagger]

272.627 Cobblestones. :beer:
9088 cred/day.
TAS' part: 48.1%

Nice production. :beer:



Now, the question for May would be, can TAS manage to beat these numbers in CPDN...

 

[mrwizer]

Originally posted by: Rattledagger
272.627 Cobblestones. :beer:
9088 cred/day.
TAS' part: 48.1%

Nice production. :beer:



Now, the question for May would be, can TAS manage to beat these numbers in CPDN...

I wish, but for me personally, I cannot add as much to CPDN.

 

[BlackMountainCow]

:beer::beer:

 

[petrusbroder]

Originally posted by: Rattledagger
272.627 Cobblestones. :beer:
9088 cred/day.
TAS' part: 48.1%

Nice production. :beer:

Now, the question for May would be, can TAS manage to beat these numbers in CPDN...

I certainly wish ... but somehow I doubt it for several reasons:

1. The WUs are much larger and that takes time. That means that if I see that production is low (say after 10 days) I certainly can add a cruncher, but those credits would not be "really" done until that WU is finished - 4 - 5 weeks later. And by that time TAS may have moved on ...
2. I can not put all my crunchers on CPDN - have quite a few oldies, which work OK for smaller WUs (I still get 10 crunched even if it takes 20 days compared to the fast ones which may take only 6 days...)
3. I still do not understand how the credits are calculated ... so how can I make a prediction which is better than a wild guess - and thus assign crunchers to the project
4. I do not want to quit serti@home and Einstein@home (gee, why not ....)

So my guess is: we will not be able to reproduce the impact that TAS had on E@H in april

on the other hand: we will certainly try! :thumbsup:

 

[BlackMountainCow]

Indeed, I think that CPDN is a really hard project for any PC. I've had several overcloked computers fail on CPDN that run just fine with any other DC and are Prime95 stable. I don't think that TAS will have so much of an impact this time but we surely will give TA a good boost in overall creds. It's just a shame that the old credits from CPDN classic didn't trasnfer to the BOINC version. For CPDN classic TAS could definitly have a good impact, but BOINC just makes more sense.

:beer:

 

[amdxborg]

Thanx for the stats Peter! :beer:

TAS did great in E@H, hopefully we can improve on this for CPN. With more members joining in, we could really get in there!

 

[Rattledagger]

In CPDN, you'll upload a trickle every 10.802 timesteps, and gets... 94,5175 Cobblestones/trickle. Every wu has 72 trickles.
How many seconds per timestep is displayed in the screensaver, or when uploaded a trickle. It will be some slight variation through a wu, but you'll get a fairly accurate estimate of cobblestone/day after running only a couple of minutes.

Anyway, it's highly unlikely most of the computers manages to finish exactly 1 wu
in 31 days, so let's see on total production:

272.627 CS is... 40 CPDN-wu.

Hmm, the gap to #19 and The Knights Who Say Ni! is only 207.220 CS or 31 wu, but they're crunching 11.128 CS/day while Anandtech only 9.423 CS/day meaning 1.705 CS/day more so catching them will be difficult...



As for computers crashing in CPDN due to unstable overclock, this is nothing new.

 

[GLeeM]

Originally posted by: Rattledagger
As for computers crashing in CPDN due to unstable overclock, this is nothing new.

I thought my duallies overclock (1.6@2.5 ) was stable when it could do four F@H BigWUs.

CPDN froze the computer, twice. The mouse pointer moved but nothing else The temps were good 45C

Sorry, I'm going to have to sit this one out. I will keep doing E@H and some Seti Classic instead.

 

[networkman]

It was a little later than planned, but I did get the 2nd of two DL360s online and running E@H for the TeAm this evening.

 

[BlackMountainCow]

Originally posted by: GLeeM

Originally posted by: Rattledagger
As for computers crashing in CPDN due to unstable overclock, this is nothing new.

I thought my duallies overclock (1.6@2.5 ) was stable when it could do four F@H BigWUs.

Sometimes I think that CPDN is over-picky when it comes to OCs. I mean, how can a PC be 48 hours Prime95 & memtest86+ stable and then fail in CPDN? Doesn't really make sense to me.

 

[ryokum]

Thanks for all your help TAS!!! You guys really rock!!!

 

[Rattledagger]

Originally posted by: BlackMountainCow
Sometimes I think that CPDN is over-picky when it comes to OCs. I mean, how can a PC be 48 hours Prime95 & memtest86+ stable and then fail in CPDN? Doesn't really make sense to me.

Well, different mathematical functions all uses the fpu, but not neccessarily the same part of the fpu.

So as an example let's say you've got 3 different functions, all using their own part of the fpu:
sin/cos, ln/exp, multiply/divide.

For easy's sake let's also say each of these functions takes exactly the same amount of cpu-time.


Let's say project A and project B both uses each of these functions 33% of the time, but with a small difference.
Project A runs one sin/cos, followed by one multiply/divide, followed by one ln/exp and starts again with sin/cos in an endless loop.
Project B runs 30 minutes sin/cos, 30 minutes multiply/divide, 30 minutes ln/exp before loops back to sin/cos.

Project A will spread-out the load and heat over the fpu all the time, so if there's a weak spot in one of the parts chances are you'll not detect it.
Project B on the other hand will use each part of the fpu exclusively for some time, so if one of them has a weak spot it's a much bigger chance it will error-out when the heat in this part of the fpu increases.



So while prime95 and memtest is stable, this is no guarantee another DC-project isn't using the "wrong" part of the system too much.

 

[Smoke]

You know, I think I understood that.

 

[amdxborg]

Originally posted by: ryokum
Thanks for all your help TAS!!! You guys really rock!!!

It's great helping out! :beer:

 

[BlackMountainCow]

I see what you mean Rattle, but it still doesn't really explain why CPDN is so picky. I mean, yes, in comparison to only ONE other DC program you'd have a point. But on my Athlon XP3400+ (overclocked to 2,4 GHz from a XP2500) I never had ANY problem with any other DC client, any game or Windows. And if you look into my sigline, you'll see that I've gone through quite some DCs by now. None of them had a problem. Only CPDN made me clock back to normal. I'd like to think that all the other DCs and games would have stressed my FPU/CPU in all its areas enough to ensure that it's really a stable OC.

 

[Rattledagger]

Well, windows doesn't normally use fpu, and most games is more dependant on the gpu than fpu. Besides, an error in a game often just means a blue pickel instead of a yellow for a split second, or a bullet doesn't go there it should or clicking a button doesnt work so must click again.

As for seti classic, you can spit out invalid results, but still get normal crunch-times. Only when you starts to spit out only 1-minute-wu do you know you've overclocked too far.


All overclocking is to get as close to the edge but not tip over, see http://setiweb.ssl.berkeley.edu/forum_thread.php?id=5976#93848

CPDN is AFAIK more demanding on hd, memory and cpu than any other DC-projects. It's therefore not unreasonable CPDN manages to push the weakest part just a little bit further than other projects, and therefore tip it over the edge.

 

[petrusbroder]

Hi mates,
Here are the final stats for TAS' effort in April when we crunched Einstein@home.
I have based the numbers on the data which were registered in May, 4, 2005 (for those who continued to crunch E@H after May 1) and on any data until May 14 for those, who stopped. I regarded any cruncher, who had 0 credits added to their account for 5 days or more as "stopped" and for those I took the total as of May 14.

TAS had
a total (March, 31): __ 60 702 cobblestones
a total (May, 4): ____ 355 263 cobblestones
a production of: ____ 294 561 cobblestones

TA had
a total (March 31): __ 560 099 cobblestones
a total (May, 4): __ 1 170 540 cobblestones
a production of: ____ 610 441 cobblestones

Thus TAS has produced 48.25% of TA's production in April.

TA increased its position among the teams from 13th place to 12th.
TA passed 2 teams (AMD Users and Overclockers.com) and was passed by 1 team (Off_Topic, which BTW now is in position 9).
TA passed a big milestone: 1 000 000 cobblestones.

March 31, TA's position in the charts was threatened by the "Knights Who Say Ni", who were overtaking TA at a good clip. April 30, TA had put a very healthy distance to the Knights (230 829 cobblestones)

TA had a maximum RAC of 18 903 (in March 31: 11453, in May 4: 17 801) and produced on the best day 41 918 cobblestones (April 27)

Among the crunchers petrusbroder, mrwizer, were in 1st or 2nd position for RAC for more than 3 weeks, TAS had 7 crunchers among the top 10, and 11 among the top 20. The team has 69 members, among them were 53 active in april...

Well that was quite a month, wasn't it?

In may TAS is crunching ClimatePreDictionNet and in June TAS will crunch Predictor.

Please join the fun @ TAS where there is a lot of fun, stats, some neffing and good mates!

 

[GLeeM]

Thanks petrusbroder (I see that the E@H TeAm has managed to keep their position!)

TAS rocks!

You can be a part of this, if you join now come help me get a jump start on Predictor.

 

[amdxborg]

Thanx Peter for the awesome work on the stats. We appriciate all your hard work, you deserve some ice cold :beer:

Great work TAS, we make a HUGE difference wherever we GO! Keep up the super crunching my fellow TeAmMates!

The TA Sentinel Rally TeAm could always do with more help. Even if you add just one pc, a difference WILL be made in a project. Come over to the forums and look around, the guys are great and we're always eager to welcome a new TeAmMate to the FUN!!