Less credit with Seti@home 7 ?

Message boards : Number crunching : Less credit with Seti@home 7 ?

To post messages, you must log in.

AuthorMessage
Profile Seneca
Volunteer tester
Avatar

Send message
Joined: 17 Oct 02
Posts: 32
Credit: 5,632,707
RAC: 6,458
Germany
Message 1405201 - Posted: 19 Aug 2013, 19:38:00 UTC
Last modified: 19 Aug 2013, 19:38:35 UTC

With my actual machine, I got RAC scores of 4000 and (slightly) above if rumming about 8 hours a day. Since seti@home 7 I see no more than about 2800.

Intel i7-2600k@3.4GHz, NVIDIA GTX570
Win7Pro x64, NVIDIA driver 320.49, CUDA 5.50

Any hint ?
ID: 1405201 · Report as offensive
OzzFan
Volunteer tester
Avatar

Send message
Joined: 9 Apr 02
Posts: 15126
Credit: 45,594,079
RAC: 46,865
United States
Message 1405206 - Posted: 19 Aug 2013, 19:41:27 UTC - in response to Message 1405201.  

The forum has practically imploded with all the complaints about the credits (and RAC) being less than it was before. I'm surprised you didn't notice the other threads.
ID: 1405206 · Report as offensive
Profile cov_route
Avatar

Send message
Joined: 13 Sep 12
Posts: 342
Credit: 10,268,140
RAC: 686
Canada
Message 1405214 - Posted: 19 Aug 2013, 19:57:30 UTC

Here is the main thread about CreditNew & credit:

http://setiathome.berkeley.edu/forum_thread.php?id=72169
ID: 1405214 · Report as offensive
Lionel

Send message
Joined: 25 Mar 00
Posts: 665
Credit: 351,387,981
RAC: 140,670
Australia
Message 1405283 - Posted: 19 Aug 2013, 22:59:37 UTC - in response to Message 1405214.  

Seneca

With the introduction of v7, credit dropped to circa 50% of that for v6. Since then they have played with a knob or two and tried to tweak the system, albeit without any success. The issue is that there is a fundamental design flaw with the system (most probably due to the introduction of auto correlation). As I have said to others before, I do not think tweaking is the answer. They need to look at the design (conceptual and logical) and determine the root cause of the issue (which should include fundamental analysis across identical work units run against v6 and v7). Once they understand the issue and behavioural differences, then they can re-frame the solution and test, with volunteers if necessary. However at the moment they are not doing this which will only lead to more angst and frustration at both ends.

I almost get the impression that they don't know what they are doing, or whether they understand how they should go about solving the problem.

cheers
ID: 1405283 · Report as offensive
Profile Ageless
Avatar

Send message
Joined: 9 Jun 99
Posts: 13822
Credit: 3,269,733
RAC: 0
Netherlands
Message 1405294 - Posted: 20 Aug 2013, 0:21:26 UTC - in response to Message 1405283.  

I almost get the impression that they don't know what they are doing, or whether they understand how they should go about solving the problem.

Did you miss Eric's answer, or did you just dismiss the answer as it wasn't what you wanted it to be?
Jord

Ancient Astronaut Theorists suggest that in many ways, you can be considered an alien conspiracy!
ID: 1405294 · Report as offensive
Profile Seneca
Volunteer tester
Avatar

Send message
Joined: 17 Oct 02
Posts: 32
Credit: 5,632,707
RAC: 6,458
Germany
Message 1405579 - Posted: 20 Aug 2013, 15:24:49 UTC

OK - I've failed to read back that far ... sorry for asking about sth already discussed that much.

I've got the clue - for some design reason the RAC values drop by 50%.

I'll watch the behaviour if it comes up again, but I presume that will not be soon.

I'm somewhat puzzled about the RAC calculation ... I would presume that a workunit consumes an amount of calculations, which converts to credits somehow (credits = operations x factor). Looks like that's not the truth ?

Is there any doc how Credits (and RAC) is calculated ?
ID: 1405579 · Report as offensive
Tutankhamon "Communist"
Volunteer tester
Avatar

Send message
Joined: 1 Nov 08
Posts: 6091
Credit: 37,750,513
RAC: 18,042
Sweden
Message 1405584 - Posted: 20 Aug 2013, 18:31:46 UTC - in response to Message 1405579.  

OK - I've failed to read back that far ... sorry for asking about sth already discussed that much.

I've got the clue - for some design reason the RAC values drop by 50%.

I'll watch the behaviour if it comes up again, but I presume that will not be soon.

I'm somewhat puzzled about the RAC calculation ... I would presume that a workunit consumes an amount of calculations, which converts to credits somehow (credits = operations x factor). Looks like that's not the truth ?

Is there any doc how Credits (and RAC) is calculated ?


It sure is a "doc", sort of: http://boinc.berkeley.edu/trac/wiki/CreditNew

Good luck understanding it.
This is a test of the Emergency Moron System. Had there been a real moron in the room, there would've been a small mushroom cloud in the place where the idiot had been standing.
ID: 1405584 · Report as offensive
Profile Link
Avatar

Send message
Joined: 18 Sep 03
Posts: 805
Credit: 1,678,562
RAC: 22
Germany
Message 1405880 - Posted: 21 Aug 2013, 9:54:05 UTC - in response to Message 1405579.  

I'm somewhat puzzled about the RAC calculation ... I would presume that a workunit consumes an amount of calculations, which converts to credits somehow (credits = operations x factor). Looks like that's not the truth ?

Well, it is like that, but...

Since you are running stock applications: if you have data for the used CPU time and credit awarded for v6 CPU workunits and if you compare them with v7 CPU workunits you should see, that you get there about the same amount of credit per CPU-day. So there should not be any significant change on your CPU.

Where you get less credits is from your GPU. The new autocorrelation stuff is not using your GPU as efficiently as the old stuff did, so your GPU is performing less calculations in the same time, so here you get less credit per day.
.
ID: 1405880 · Report as offensive
Profile Cliff Harding
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 18 Aug 99
Posts: 1265
Credit: 73,394,999
RAC: 19,435
United States
Message 1405899 - Posted: 21 Aug 2013, 11:20:26 UTC - in response to Message 1405880.  

Since you are running stock applications: if you have data for the used CPU time and credit awarded for v6 CPU workunits and if you compare them with v7 CPU workunits you should see, that you get there about the same amount of credit per CPU-day. So there should not be any significant change on your CPU.

Where you get less credits is from your GPU. The new autocorrelation stuff is not using your GPU as efficiently as the old stuff did, so your GPU is performing less calculations in the same time, so here you get less credit per day.


With all of the v7 work being generated and returned that the credits would have balanced out by now, any idea when they will become normalized to v6 or is that an impossibility?


I don't buy computers, I build them!!
ID: 1405899 · Report as offensive
Profile Link
Avatar

Send message
Joined: 18 Sep 03
Posts: 805
Credit: 1,678,562
RAC: 22
Germany
Message 1406062 - Posted: 21 Aug 2013, 17:54:24 UTC - in response to Message 1405899.  

With all of the v7 work being generated and returned that the credits would have balanced out by now, any idea when they will become normalized to v6 or is that an impossibility?

According to Eric's post it's the case already for stock CPU apps, which are the reference.

Optimized and GPU is another story, from what has been posted in the other thread part of the issue why opt. CPU apps aren't that much faster than stock anymore is because they had to use some crappy compiler and not the one from Intel.
.
ID: 1406062 · Report as offensive
Sirius B
Volunteer tester
Avatar

Send message
Joined: 26 Dec 00
Posts: 14939
Credit: 2,129,460
RAC: 619
Ireland
Message 1406085 - Posted: 21 Aug 2013, 18:47:11 UTC - in response to Message 1406062.  

With all of the v7 work being generated and returned that the credits would have balanced out by now, any idea when they will become normalized to v6 or is that an impossibility?

According to Eric's post it's the case already for stock CPU apps, which are the reference.

Optimized and GPU is another story, from what has been posted in the other thread part of the issue why opt. CPU apps aren't that much faster than stock anymore is because they had to use some crappy compiler and not the one from Intel.


Well with all the cash & hardware donations for the project, why not have a donation run to get a licensed copy of Intel's compiler for Lunatics or will it cost too much?
ID: 1406085 · Report as offensive
Profile Cliff Harding
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 18 Aug 99
Posts: 1265
Credit: 73,394,999
RAC: 19,435
United States
Message 1406101 - Posted: 21 Aug 2013, 19:50:36 UTC - in response to Message 1406085.  

With all of the v7 work being generated and returned that the credits would have balanced out by now, any idea when they will become normalized to v6 or is that an impossibility?

According to Eric's post it's the case already for stock CPU apps, which are the reference.

Optimized and GPU is another story, from what has been posted in the other thread part of the issue why opt. CPU apps aren't that much faster than stock anymore is because they had to use some crappy compiler and not the one from Intel.


Well with all the cash & hardware donations for the project, why not have a donation run to get a licensed copy of Intel's compiler for Lunatics or will it cost too much?


+1


I don't buy computers, I build them!!
ID: 1406101 · Report as offensive
Profile j mercer
Avatar

Send message
Joined: 3 Jun 99
Posts: 1721
Credit: 11,819,254
RAC: 233
United States
Message 1406109 - Posted: 21 Aug 2013, 20:22:45 UTC - in response to Message 1406101.  

+2
...
ID: 1406109 · Report as offensive
Josef W. Segur
Volunteer developer
Volunteer tester

Send message
Joined: 30 Oct 99
Posts: 4504
Credit: 1,414,761
RAC: 0
United States
Message 1406113 - Posted: 21 Aug 2013, 20:28:14 UTC - in response to Message 1406085.  

With all of the v7 work being generated and returned that the credits would have balanced out by now, any idea when they will become normalized to v6 or is that an impossibility?

According to Eric's post it's the case already for stock CPU apps, which are the reference.

Optimized and GPU is another story, from what has been posted in the other thread part of the issue why opt. CPU apps aren't that much faster than stock anymore is because they had to use some crappy compiler and not the one from Intel.

Well with all the cash & hardware donations for the project, why not have a donation run to get a licensed copy of Intel's compiler for Lunatics or will it cost too much?

It's not a matter of having a licensed copy of the Intel tools, the issue is distributing open source software where the binaries have been produced by those proprietary tools. The GPL is necessary for work derived from the project sources but seems somewhat in conflict with the Intel EULAs for their compiler and libraries. Nobody at Lunatics is in a position to supply the money or time to deal with a possible court case which would be needed to clarify whether the legal system would judge that conflict real or not.

The GCC compiler tool chain used for the CPU builds is not crappy, but getting the best performance from it will take awhile. The r1846 CPU builds in the v0.41 installer were produced under some time pressure for the v7 rollout, better versions are being worked on.
                                                                  Joe
ID: 1406113 · Report as offensive
Profile cov_route
Avatar

Send message
Joined: 13 Sep 12
Posts: 342
Credit: 10,268,140
RAC: 686
Canada
Message 1406193 - Posted: 21 Aug 2013, 23:51:22 UTC - in response to Message 1406113.  

The GCC compiler tool chain used for the CPU builds is not crappy, but getting the best performance from it will take awhile. The r1846 CPU builds in the v0.41 installer were produced under some time pressure for the v7 rollout, better versions are being worked on.
                                                                  Joe

That probably explains why, in my tests, the SSE3 version of V8c_Bb ran at the same speed as the SSE2 version.

The switch to GCC might be good for AMD CPUs, putting them on a more equal footing for once. At least it will remove any nagging doubts about code-path shenanigans, real or imagined. I've heard good things about the gcc sse code generation.
ID: 1406193 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 5432
Credit: 66,425,273
RAC: 12,908
Russia
Message 1406271 - Posted: 22 Aug 2013, 4:08:45 UTC - in response to Message 1406193.  

GCC builds slower than ICC ones on my Athlon XP (SSE-only) too.
So, not the case, unfortunately.

SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1406271 · Report as offensive
Profile William
Volunteer tester
Avatar

Send message
Joined: 14 Feb 13
Posts: 2037
Credit: 14,583,619
RAC: 10,370
Message 1406434 - Posted: 22 Aug 2013, 13:02:42 UTC

Actually, if Joe hadn't started working on definitely GPL compliant opt V7 CPU apps, when no other dev would and rushed to make something that was at least somewhat better than stock, we wouldn't have had anything at all to put in the installer!

I'm in no rush to do the next one...
A person who won't read has no advantage over one who can't read. (Mark Twain)
ID: 1406434 · Report as offensive

Message boards : Number crunching : Less credit with Seti@home 7 ?


 
©2016 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.