Credit
log in

Advanced search

Message boards : Number crunching : Credit

1 · 2 · 3 · 4 . . . 10 · Next
Author Message
Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 2
Message 2 - Posted: 15 Jun 2009, 17:57:46 UTC
Last modified: 4 Aug 2009, 12:34:57 UTC

Credit is fixed. It has been set to 3.19 credits per 2^24 numbers processed, or about 400 credits per workunit. The alternate method which is being considered is to grant 1 credit per every 1501972363 steps. That will allow the application to be self adjusting as larger numbers are being calculated and the average steps per number increases.

Note: All credit calculations are based on a Pentium III 800 Mhz machine which is rated at 98.7 credits per day which is as close as I can get to the baseline 100 credit per day machine. Credit calculations are and will continue to be determined using the stock CPU application.

2009-08-04 Update: Credit is no longer fixed. It is calculated using the total steps calculated and is self-adjusting. As the numbers being calculated get larger, they will on average require more steps to calculate. Using total steps calculated also allows the size of the WU to change without affecting the credits. If the size doubles (e.g. twice as many numbers are calculated per WU) then the credit will also double automatically.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 2
Message 6 - Posted: 3 Jul 2009, 3:20:07 UTC

Aaaarrghh!! It looks like the CPU version is calculating one extra number per WU which throws off the total steps calculated (the last number reported). I thought I had them all reporting the same. On the other hand, the number of steps is constantly increasing so it looks like I have to get the validator to look at lines 3 and 4 to determine validity and then grant credit according to the average steps between the two. That, or a new CPU app which goes back to skipping the last number. As soon as the baseline machine (P3 800Mhz) finishes a WU (this weekend) I'll have a better idea on how many credits to grant per steps calculated. Maybe I can add a checksum or something and then not require a quorum either. That would be cool. And, because is dawned on me today, there's no need to report the cpu time in the stderr since the WU already reports it. That should shave off a millisecond or two. :wink:

In the mean time, if you see an "inconclusive" result, let me know. I'll manually check the status and grant credit if lines 2 and 3 of the result match.

For example:

CPU:
collatz_1245883152_43
2361183347662436149608
2361183347664583633256
2361183347662476382916
1405
1100773226328

CUDA:
collatz_1245883152_43
2361183347662436149608
2361183347664583633256
2361183347662476382916
1405
1100773225877

Profile Logan
Avatar
Send message
Joined: 2 Jul 09
Posts: 124
Credit: 37,455,338
RAC: 0
Message 7 - Posted: 3 Jul 2009, 8:48:41 UTC - in response to Message 6.
Last modified: 3 Jul 2009, 8:56:56 UTC

Aaaarrghh!! It looks like the CPU version is calculating one extra number per WU which throws off the total steps calculated (the last number reported). I thought I had them all reporting the same. On the other hand, the number of steps is constantly increasing so it looks like I have to get the validator to look at lines 3 and 4 to determine validity and then grant credit according to the average steps between the two. That, or a new CPU app which goes back to skipping the last number. As soon as the baseline machine (P3 800Mhz) finishes a WU (this weekend) I'll have a better idea on how many credits to grant per steps calculated. Maybe I can add a checksum or something and then not require a quorum either. That would be cool. And, because is dawned on me today, there's no need to report the cpu time in the stderr since the WU already reports it. That should shave off a millisecond or two. :wink:

In the mean time, if you see an "inconclusive" result, let me know. I'll manually check the status and grant credit if lines 2 and 3 of the result match.

For example:

CPU:
collatz_1245883152_43
2361183347662436149608
2361183347664583633256
2361183347662476382916
1405
1100773226328

CUDA:
collatz_1245883152_43
2361183347662436149608
2361183347664583633256
2361183347662476382916
1405
1100773225877


Hi Slicker!!! I am giving a short tour in your project...;)

Completed, validation inconclusive WUs:

976
964
955
954
931
909
904
900
838
819
810
809
657
758
724
711
699
690
660
656
656
631
627
626

And growing....;)
____________
Logan.

BOINC FAQ Service (Ahora, también disponible en Español/Now available in Spanish)

Profile Crunch3r
Volunteer moderator
Project developer
Project tester
Avatar
Send message
Joined: 30 Jun 09
Posts: 219
Credit: 7,515,635,101
RAC: 15,467
Message 8 - Posted: 3 Jul 2009, 13:25:52 UTC - in response to Message 7.

OMG ! I got robbed too ! Horrible to say the least !
664
626
925
923
and probably more... was to shocked to count any further!

So... am i right to assume that we'll get a zillion credits for those WU as a compensation for all the trouble/pain not getting credit in the first place ? :p





____________

Team BOINC United.Join Science that matters.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 2
Message 16 - Posted: 3 Jul 2009, 16:01:00 UTC

Credit was awarded this morning.

I also put in another new validator build that I hope fixes the problem once and for all. So far, it looks to be working OK.

Profile Crunch3r
Volunteer moderator
Project developer
Project tester
Avatar
Send message
Joined: 30 Jun 09
Posts: 219
Credit: 7,515,635,101
RAC: 15,467
Message 18 - Posted: 3 Jul 2009, 16:49:25 UTC - in response to Message 16.

Credit was awarded this morning.

I also put in another new validator build that I hope fixes the problem once and for all. So far, it looks to be working OK.


Shall I take that as a "NO" to my zillion credit request ? :p

____________

Team BOINC United.Join Science that matters.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 2
Message 22 - Posted: 4 Jul 2009, 14:32:10 UTC - in response to Message 18.

Credit was awarded this morning.

I also put in another new validator build that I hope fixes the problem once and for all. So far, it looks to be working OK.


Shall I take that as a "NO" to my zillion credit request ? :p


A blond is watching the news with her husband when news anchor states that two Brazilian men were killed while sky diving. The blonds starts crying immediately. Her husband says to her, "Honey, accidents happen. It isn't like you knew those people. So, why are you crying?" She responds, "It is just so overwhelming. So many people! How many is a Brazilian, anyway?"

Actually, the baseline machine finally returned a WU. It claimed 376 credits, but it was also returned about 8% fewer total steps than average. Credits have now been set to around 400 per WU.

Profile Logan
Avatar
Send message
Joined: 2 Jul 09
Posts: 124
Credit: 37,455,338
RAC: 0
Message 23 - Posted: 4 Jul 2009, 16:00:22 UTC - in response to Message 22.
Last modified: 4 Jul 2009, 16:03:00 UTC

Credit was awarded this morning.

I also put in another new validator build that I hope fixes the problem once and for all. So far, it looks to be working OK.


Shall I take that as a "NO" to my zillion credit request ? :p


A blond is watching the news with her husband when news anchor states that two Brazilian men were killed while sky diving. The blonds starts crying immediately. Her husband says to her, "Honey, accidents happen. It isn't like you knew those people. So, why are you crying?" She responds, "It is just so overwhelming. So many people! How many is a Brazilian, anyway?"

Actually, the baseline machine finally returned a WU. It claimed 376 credits, but it was also returned about 8% fewer total steps than average. Credits have now been set to around 400 per WU.


As the same... the half of credits for the same work...
____________
Logan.

BOINC FAQ Service (Ahora, también disponible en Español/Now available in Spanish)

Profile Mumps [MM]
Volunteer tester
Avatar
Send message
Joined: 24 Jun 09
Posts: 16
Credit: 197,247,161
RAC: 107
Message 24 - Posted: 4 Jul 2009, 23:27:08 UTC
Last modified: 4 Jul 2009, 23:28:58 UTC

So, any thoughts on what's happening with my ATI Cruncher? Everything it returns seems to me to finish way to quickly, but there's nothing in the output to help me figure out if it's just a broken Brook install or something else.

The ATI results have yet to validate at all... Only those you manually granted credit for it seems...

Host

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 2
Message 25 - Posted: 5 Jul 2009, 3:50:46 UTC - in response to Message 24.
Last modified: 5 Jul 2009, 3:52:55 UTC

So, any thoughts on what's happening with my ATI Cruncher? Everything it returns seems to me to finish way to quickly, but there's nothing in the output to help me figure out if it's just a broken Brook install or something else.

The ATI results have yet to validate at all... Only those you manually granted credit for it seems...

Host


I see it is running Windows 7 x64. Anyone else gotten any WUs to validate using Windows 7?

What kind of ATI card is it?
What version of the ATI driver are you using?
Have you copy and renamed the 3 ATI DLLs to AMD?
Has it run MW ATI workunits OK?
What version of the ATI app are you using?

Profile Mumps [MM]
Volunteer tester
Avatar
Send message
Joined: 24 Jun 09
Posts: 16
Credit: 197,247,161
RAC: 107
Message 26 - Posted: 5 Jul 2009, 13:51:37 UTC - in response to Message 25.

So, any thoughts on what's happening with my ATI Cruncher? Everything it returns seems to me to finish way to quickly, but there's nothing in the output to help me figure out if it's just a broken Brook install or something else.

The ATI results have yet to validate at all... Only those you manually granted credit for it seems...

Host


I see it is running Windows 7 x64. Anyone else gotten any WUs to validate using Windows 7?

What kind of ATI card is it?

It's an integrated Mobility Radeon HD 3430 on my laptop

What version of the ATI driver are you using?

The Beta drivers for Win 7. Catalyst 9.6 (The only ones available for D/L)

Have you copy and renamed the 3 ATI DLLs to AMD?

Yup. Couldn't get even this far without that. :-)

Has it run MW ATI workunits OK?

Being a 3430, it won't run MilkyWay...

What version of the ATI app are you using?

The latest, 1.07.

Senilix
Send message
Joined: 30 Jul 09
Posts: 16
Credit: 281,188,980
RAC: 397,560
Message 1211 - Posted: 17 Sep 2009, 22:16:35 UTC

5K credits per wu with collatz 2.00 ?!?!? With an ellapse time of 30 min?!?!?

I hope this will last for the next couple of hours ;)

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 2
Message 1212 - Posted: 17 Sep 2009, 22:26:24 UTC

The new v2.0 apps are much faster than the v1.10 apps. The GPU apps are much more efficient and utilize the CPU much less which means the new apps take more advantage of their parallel processing capabilities. Also, because the new app uses 64-bit integer math, 64-bit platforms will finish twice as fast as 32-bit platforms. What does it all mean?

1. The WU size has been increased to keep the server load at a reasonable level. Credits are NOT fixed per WU but are a ratio of the steps calculated. So, a larger WU results in more steps and thus more credit.

2. Using the current credit for the v2.0 app is not consistent with the credit policy used for this project. A PIII 800 Mhz machine which should receive approx 100 credits per day according to its benchmarks. Using the v2.0 application, it would receive 1,000. In other words, 10 times too much.

3. Over the next several days, credit will be reduced until it reaches a point where the baseline machine is averaging 100 credits per day using the stock app. That means approx. a 10x reduction in credit if you are running a PIII 800. On the other hand, it means that you will receive the same credit next week that you did last week. If you have a 64-bit machine or a GPU, your credit will likely increase even though the credit per steps has been reduced.

4. a) Will there be SSE, SSE2, etc. optimized apps and b) will they run faster? Maybe and not much. Because the apps already use x86 intrinsics, there isn't a lot of gain for SSE or SSE2 since the math being done is almost all done by bit shifts and not be multiplication or division. As time permits, opt apps will be created and tested but I don't expect them to be much faster, if any. If you want optimized, get a 64 bit operating system. That will do much more than any SSE3 optimization will do.

5. Some will argue that you are doing more work and the credit shouldn't change. That isn't true for two reasons. First, if you want to get technical about it, Gipsel did 99.9% most of the work on the ATI and GPU clients. Crunch3r worked to make the BOINC client acknowledge ATI cards. I spent many nights swearing at the CUDA code and documentation and getting the BOINC server to acknowledge the BOINC ATI client. Secondly, the new app "cheats". It doesn't actually calculate all the steps so claiming credit for calculating all the steps isn't fair. The new app uses a lookup table so that it can process 20 steps in a single calculation. (See the Wiki page link on the home page for more information on how the optimizations work.) That's why it is 20 times faster. It isn't doing 20 times more integer operations per second. It does the same as it always did, but by "cheating" and looking ahead 20 steps at a time, if can check more numbers in the same amount of time.

_hiVe*
Send message
Joined: 9 Aug 09
Posts: 106
Credit: 162,673,032
RAC: 0
Message 1253 - Posted: 18 Sep 2009, 6:33:46 UTC
Last modified: 18 Sep 2009, 6:35:34 UTC

The credit is apparently normalized now, right Slicker?
I was not lucky enough to catch too many of those 5k WU's :[ shame /if only my wingmen would be faster back then, arrr :P/

For a same length WU the credit is 6.5x (fold) lower than it was yesterday.
I just need to know, whether thats final (or at least semi-final) or it'll still sink like a rock in the lake :)

The announcement on the main page you promised has been made, so I'd guess that the changes have been implemented...a reassurance would be nice though.

Thanks.

Profile STE\/E
Avatar
Send message
Joined: 12 Jul 09
Posts: 581
Credit: 761,710,729
RAC: 0
Message 1276 - Posted: 18 Sep 2009, 10:59:52 UTC - in response to Message 1275.



P.S. We need ability to edit post's....


We have...look near the message ID, the header above the text.
THe "EDIT" button is right there....


You have to do it within 1 Hour though or else you can't Edit the post anymore after that ...

David @ TPS
Send message
Joined: 31 Aug 09
Posts: 4
Credit: 228,339,243
RAC: 97,996
Message 1282 - Posted: 18 Sep 2009, 12:41:53 UTC - in response to Message 1276.

No trouble with verification here, but the credit slump is astounding!

Rosetta used to be my bottom-end project as far as credits.

From this morning:

Collatz -- 23K seconds = 25.08 credits
Rosetta -- 6.7K " = 26.11 credits
S@H -- 6.5K " = 27.23 credits

Profile Gipsel
Volunteer moderator
Project developer
Project tester
Send message
Joined: 2 Jul 09
Posts: 279
Credit: 77,026,643
RAC: 77,919
Message 1290 - Posted: 18 Sep 2009, 13:21:41 UTC - in response to Message 1282.

No trouble with verification here, but the credit slump is astounding!

Rosetta used to be my bottom-end project as far as credits.

From this morning:

Collatz -- 23K seconds = 25.08 credits
Rosetta -- 6.7K " = 26.11 credits
S@H -- 6.5K " = 27.23 credits

I guess you should remove the app_info.xml from both your computers so you get the new and much faster applications ;)

Profile mscharmack
Avatar
Send message
Joined: 2 Aug 09
Posts: 12
Credit: 31,952,998
RAC: 0
Message 1310 - Posted: 18 Sep 2009, 14:55:50 UTC

What about all those pending WU's that were completed before the reduced credits jumped in. Will they receive the original credit amount or will they receive the reduced amount?

Profile Logan
Avatar
Send message
Joined: 2 Jul 09
Posts: 124
Credit: 37,455,338
RAC: 0
Message 1311 - Posted: 18 Sep 2009, 15:00:07 UTC - in response to Message 1310.

What about all those pending WU's that were completed before the reduced credits jumped in. Will they receive the original credit amount or will they receive the reduced amount?


The reduced...

And I noticed now about a new reduction...:(
____________
Logan.

BOINC FAQ Service (Ahora, también disponible en Español/Now available in Spanish)

Profile Bruce
Avatar
Send message
Joined: 15 Jul 09
Posts: 294
Credit: 12,203,827
RAC: 0
Message 1316 - Posted: 18 Sep 2009, 15:23:12 UTC
Last modified: 18 Sep 2009, 15:25:26 UTC

What a pile of Stink!!! Those pendings that were run B4 the 18th SHOULD get the Credits they would of got B4 the reduction!!! I'm being penalized because my wingman didnt turn his/her WU in time. TOTALLY UNFAIR!!!!!! Considering we were given very little notice of the way the credit reduction would work.:-(
____________

1 · 2 · 3 · 4 . . . 10 · Next
Post to thread

Message boards : Number crunching : Credit


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.