Advanced search

Message boards : Graphics cards (GPUs) : Have work units gotten longer????

Author Message
mscharmack
Avatar
Send message
Joined: 20 Aug 07
Posts: 18
Credit: 1,319,274
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwat
Message 8787 - Posted: 23 Apr 2009 | 13:32:30 UTC

My SLI system was completing two work units every 19-20 hours. Now it appears they are taking 24 to 38 hours. Has anyone else seen a jump in their completion times?

Profile Michael Goetz
Avatar
Send message
Joined: 2 Mar 09
Posts: 124
Credit: 7,573,744
RAC: 32
Level
Ser
Scientific publications
watwatwatwatwatwatwatwat
Message 8788 - Posted: 23 Apr 2009 | 13:37:05 UTC - in response to Message 8787.

Forgive the dumb question, but just checking the obvious, easy stuff first:

You didn't accidently leave SLI turned on, right? That would reduce the number of GPUs available to CUDA down to 1.

Mike

Joe
Send message
Joined: 1 Sep 08
Posts: 37
Credit: 5,864,088
RAC: 0
Level
Ser
Scientific publications
watwatwatwatwat
Message 8791 - Posted: 23 Apr 2009 | 14:41:34 UTC - in response to Message 8788.

My time to complete a Wu jumps from about 8h to a little more than 10h - Vista 64Bit. With XP 32Bit from about 6h to 7h... Both with a GTX295

mscharmack
Avatar
Send message
Joined: 20 Aug 07
Posts: 18
Credit: 1,319,274
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwat
Message 8794 - Posted: 23 Apr 2009 | 17:55:00 UTC - in response to Message 8788.
Last modified: 23 Apr 2009 | 17:55:57 UTC

No, both XFX 9600 GSO's are operating independently. I just thought it was kind of odd to see a jump in times when they were pretty steady before and both are doing seperate work units.

ignasi
Send message
Joined: 10 Apr 08
Posts: 254
Credit: 16,836,000
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 8854 - Posted: 24 Apr 2009 | 18:02:34 UTC - in response to Message 8794.

If WUs do have the same name, you can assume they are of the same length.
However, between different names (which refer to different experiments) there may be differences. We are usually running different systems, that have different computational costs... so completion times may vary, though we intend to homogenize them as much as possible.

ignasi

Scott Brown
Send message
Joined: 21 Oct 08
Posts: 144
Credit: 2,973,555
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwat
Message 8868 - Posted: 24 Apr 2009 | 20:04:29 UTC - in response to Message 8854.

If WUs do have the same name, you can assume they are of the same length.


Any possibility of knowing what the naming schemes mean?


ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 8904 - Posted: 25 Apr 2009 | 13:18:09 UTC

mscharmack,

something odd is happening with your machine. If you take a look at one of your recent results you'll see that there are many "No heartbeat from core client for 30 sec - exiting" messages. This means that the GPU-Grid client looses connection to the BOINC client and subsequentially shuts down. So the actual time per step didn't grow, the WUs didn't grow (at least not that much), but the wall clock time until WU completion increases.

This may either mean that the BOINC client stopped working (you'd have to restart it manually) or that the communication is disrupted, which could be due to a overzealous firewall or any other strange windows / network issue. Sorry, can't be more specific than this.

MrS
____________
Scanning for our furry friends since Jan 2002

Profile Megacruncher TSBT
Send message
Joined: 7 Aug 08
Posts: 8
Credit: 971,097,823
RAC: 7,770
Level
Glu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 8969 - Posted: 27 Apr 2009 | 0:05:13 UTC - in response to Message 8904.

I've certainly found that WU are taking longer and that as a result my daily credit is plummeting. What gives?
____________
The Scottish Boinc Team

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 8971 - Posted: 27 Apr 2009 | 2:56:28 UTC - in response to Message 8969.

I've certainly found that WU are taking longer and that as a result my daily credit is plummeting. What gives?

Are you using BOINC 6.6.20?

Best alternative, roll back to 6.5.0 ... a reported problem ... 6.6.23 is a potential alternative, but it looks to ME like the screwed up work fetch so you have to watch and reset debts (for me every 24-48 hours) which means that these later versions are not yet ready for prime time ... 6.6.24 will not use the second GPU if you have more than one ...

Waiting for 6.6.25 ... :)

Profile JockMacMad TSBT
Send message
Joined: 26 Jan 09
Posts: 31
Credit: 3,877,912
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwat
Message 8983 - Posted: 27 Apr 2009 | 11:18:29 UTC - in response to Message 8971.

Paul

I am using 6.6.24 and all 4 GPU's (2x295) are working (still!).

Now I have tweaked the registry to force all the GPU's so maybe the same will work for you.

Or is it just in the situation where they are 2 different GPU's ?


Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 9005 - Posted: 27 Apr 2009 | 19:05:58 UTC

No, I have the same... two GTX 295s ... though I think that they are from different MFGRs (pretty sure they are) ... which may be the cause. The trouble is the new test does not tell you WHAT failed.

I suggested a change that would do that ... though I don't know if they are going to add that code to 6.6.25 or not ...

There maybe a different change with a cc_config flag to force use of all GPUs ...

Profile Megacruncher TSBT
Send message
Joined: 7 Aug 08
Posts: 8
Credit: 971,097,823
RAC: 7,770
Level
Glu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 9009 - Posted: 27 Apr 2009 | 19:44:03 UTC - in response to Message 8971.

I've certainly found that WU are taking longer and that as a result my daily credit is plummeting. What gives?

Are you using BOINC 6.6.20?

Best alternative, roll back to 6.5.0 ... a reported problem ... 6.6.23 is a potential alternative, but it looks to ME like the screwed up work fetch so you have to watch and reset debts (for me every 24-48 hours) which means that these later versions are not yet ready for prime time ... 6.6.24 will not use the second GPU if you have more than one ...

Waiting for 6.6.25 ... :)


I'm using 6.6.15. I've got no problem getting work and on my 2 X 260GTX machine both GPUs are crunching away. Nothing is crashing and there are no errors. It's just that WUs are taking two or three times longer to run - for the same credit.

____________
The Scottish Boinc Team

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 9021 - Posted: 27 Apr 2009 | 21:26:44 UTC - in response to Message 9009.

I've certainly found that WU are taking longer and that as a result my daily credit is plummeting. What gives?

Are you using BOINC 6.6.20?

Best alternative, roll back to 6.5.0 ... a reported problem ... 6.6.23 is a potential alternative, but it looks to ME like the screwed up work fetch so you have to watch and reset debts (for me every 24-48 hours) which means that these later versions are not yet ready for prime time ... 6.6.24 will not use the second GPU if you have more than one ...

Waiting for 6.6.25 ... :)


I'm using 6.6.15. I've got no problem getting work and on my 2 X 260GTX machine both GPUs are crunching away. Nothing is crashing and there are no errors. It's just that WUs are taking two or three times longer to run - for the same credit.

You may be running into the bug we thought was in 6.6.20 ... roll back to 6.5.0 and try that and see if the times go back to normal. That is what I did when the problem hit me. I went from 16 tasks a day to 8-10. Rolled back to 6.5.0 and my 6 hour run times came back. (some tasks took as long as 24 hours thought the time step did not seem to change).

Seriously, you should consider it. If this changes things, then we know the long run bug was introduced earlier than 6.6.20 ...

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 9033 - Posted: 27 Apr 2009 | 21:51:22 UTC - in response to Message 9021.

Seriously, you should consider it. If this changes things, then we know the long run bug was introduced earlier than 6.6.20 ...


I second that. A rutime 2-3 times larger is not normal.

MrS
____________
Scanning for our furry friends since Jan 2002

Profile Megacruncher TSBT
Send message
Joined: 7 Aug 08
Posts: 8
Credit: 971,097,823
RAC: 7,770
Level
Glu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 9038 - Posted: 27 Apr 2009 | 23:54:08 UTC - in response to Message 9033.

I downgraded to 6.5.0 and it seems to be have speeded things up no end! Thanks very much for the advice.
____________
The Scottish Boinc Team

mscharmack
Avatar
Send message
Joined: 20 Aug 07
Posts: 18
Credit: 1,319,274
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwat
Message 9045 - Posted: 28 Apr 2009 | 4:27:58 UTC

Reverted back to 6.4.7 today. The system seems to be much more stable than under 6.6.20. I will keep it there for a few days and check out the results.

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 9051 - Posted: 28 Apr 2009 | 8:08:51 UTC

Thank you for the feedback.

I wish it were otherwise, but, sometimes the latest and greatest isn't ...

That is why, unless there is a compelling new feature you MUST have, it is usually best to wait at least 30 days before adopting a new version. The problems with a version don't always come out in the first few days, unless it is glaring and major.

The more subtle issues are harder and thus take longer to figure out what is going on.

The only reason I always suggest 6.5.0 over the 6.4.x versions is that there is no need for the configuration file hacks to get the GPU running.

Anyway, good you are back running.

uBronan
Avatar
Send message
Joined: 1 Feb 09
Posts: 139
Credit: 575,023
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 9066 - Posted: 28 Apr 2009 | 13:07:30 UTC
Last modified: 28 Apr 2009 | 13:08:36 UTC

Hmm now to be honest i see some units really have a much longer time to complete nowadays.
These GIANNI units really cost my little machine much more time to complete
the older units i did almost allways in less then 24 hours.
These newer units start with an estimate of 17 and by the looks of it is going to take more then 24 hours.
It is at the moment that i post at 21 hours 45 minutes and reports to have still 5 hours to go.
Even with good old boinc 6.5.0

jrobbio
Send message
Joined: 13 Mar 09
Posts: 59
Credit: 324,366
RAC: 0
Level

Scientific publications
watwatwatwat
Message 9073 - Posted: 28 Apr 2009 | 15:42:27 UTC - in response to Message 9066.

It would be good to have some repeatable test units through GPU grid which you could subscribe to for testing purposes (In CPDN Beta you can subscribe to different tests). Maybe 3 or 4 workunits that have been verified as clean and reliable could be repeatedly retrievable would make comparisons a lot easier.

It would stop me running out of units for the day when it starts erroring out and I cannot get any more.

Rob

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 9088 - Posted: 28 Apr 2009 | 21:22:11 UTC - in response to Message 9066.

I'd say +/-10% is normal. It could also be that you'll get more credits for WUs which lasted longer.

MrS
____________
Scanning for our furry friends since Jan 2002

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1947
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 9090 - Posted: 28 Apr 2009 | 22:08:33 UTC - in response to Message 9088.

There have been not substantially changes in the size of WUs.
In particular the one I am running have the same length of many of the past.

Best way to check speed is to look for another computer which runs the same group of WUs. The tag of the workunit gives it up:

name-group-etc-etc

gdf

fractal
Send message
Joined: 16 Aug 08
Posts: 87
Credit: 1,248,879,715
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 9096 - Posted: 29 Apr 2009 | 2:40:45 UTC - in response to Message 9051.

TI wish it were otherwise, but, sometimes the latest and greatest isn't ...

That is why, unless there is a compelling new feature you MUST have, it is usually best to wait at least 30 days before adopting a new version. The problems with a version don't always come out in the first few days, unless it is glaring and major.

The more subtle issues are harder and thus take longer to figure out what is going on.

The only reason I always suggest 6.5.0 over the 6.4.x versions is that there is no need for the configuration file hacks to get the GPU running.


This is why I am still running 6.3.21.

I can not find any indication that any newer release is any better, and most appear to be worse.

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 9100 - Posted: 29 Apr 2009 | 6:36:30 UTC - in response to Message 9096.

This is why I am still running 6.3.21.

I can not find any indication that any newer release is any better, and most appear to be worse.

Well, 6.5.0 is good enough that I have found no issue with it yet that other versions don't also have.

Or, to put it another way, there are no new and unique issues with that version. But, if yours works for you ... :)

uBronan
Avatar
Send message
Joined: 1 Feb 09
Posts: 139
Credit: 575,023
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 9106 - Posted: 29 Apr 2009 | 8:45:58 UTC
Last modified: 29 Apr 2009 | 8:47:52 UTC

this is old :

522140 369932 16 Apr 2009 1:49:24 UTC 17 Apr 2009 8:02:26 UTC Over Success Done 1,320.89 2,883.44 4,613.50

then the newer came :


592920 416850 27 Apr 2009 21:49:19 UTC 28 Apr 2009 17:24:51 UTC Over Success Done 1,664.50 3,843.30 4,804.12
587989 414037 26 Apr 2009 23:11:14 UTC 27 Apr 2009 21:45:17 UTC Over Success Done 1,683.81 3,843.30 4,804.12
587883 414011 26 Apr 2009 23:11:14 UTC 27 Apr 2009 16:13:23 UTC Over Redundant result Cancelled by server 0.00 --- ---
583337 410986 25 Apr 2009 23:43:01 UTC 27 Apr 2009 1:12:17 UTC Over Success Done 1,307.28 3,946.78 4,933.48
580325 407937 25 Apr 2009 9:41:06 UTC 26 Apr 2009 5:31:13 UTC Over Success Done 1,661.56 3,946.78 4,933.48
574507 407246 24 Apr 2009 7:18:54 UTC 25 Apr 2009 2:52:59 UTC Over Success Done 1,503.06 3,844.58 4,805.73
570199 405183 23 Apr 2009 11:18:40 UTC 24 Apr 2009 7:18:14 UTC Over Success Done 1,760.33 3,946.78 4,933.48
561277 400709 22 Apr 2009 11:43:26 UTC 23 Apr 2009 8:13:11 UTC Over Success Done 1,964.63 3,843.30 4,804.12

So in general it went from 1300-1500 seconds of cpu time to 1600-1960 seconds cpu time as you can see the newer units only get marginaly more. but less then the when it returns faster.
I assume i get less bonus because the return time is now over 24 hours, or something like that.
Since the largest units gets less then the one running for 1370 seconds (4933 points) 1600- 1900 seconds work get 4800.
As i am guessing is that 100 seconds more time here in GPU time about an hour of work more or less, and thats about what i see
The last unit ended after 26 hours of crunching and the new one i received is going to pass that.
Since it shows more hours to go after it allready run for the same time what i noted on the previous.
This is unit p1020000-IBUCH_pYIpYVk52804-1-10-RND3781_0 the previous was a GIANNI with also RND on the end.

Profile Lazarus-uk
Send message
Joined: 16 Nov 08
Posts: 29
Credit: 122,821,515
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 9109 - Posted: 29 Apr 2009 | 10:03:54 UTC

Just to let you all know that there are some much longer WUs out there. I just finished this one:

88-KASHIF_HIVPR_dim_ba5-0-100-RND6765_0

It took a little under 14hrs on an overclocked GTX260 (1512 shader) and one of the finish files was 50MB+. It all took 20mins to upload.

Claimed credit 8076.97800925926
Granted credit 10096.2225115741


Mark

uBronan
Avatar
Send message
Joined: 1 Feb 09
Posts: 139
Credit: 575,023
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 9111 - Posted: 29 Apr 2009 | 12:42:36 UTC
Last modified: 29 Apr 2009 | 12:54:28 UTC

Wow me scared if i ever got one of those :D
But i probably won't receive such units :D, because i think it will pass my deadline for at least 200 hours ;)
You have a powercard i have the slowest possible one xD
Btw i have also the problem to get my cpu's to run MW and get units on my VC,
i need to set the get work cache to 10 days to get enough work for my 4 cpu's or it runs dry for about 12 hours.
But on the other hand have the problem that i can handle only 1 unit at about 22 to 27 hours, when i leave it gpugrid ofcourse loads new and cancels them some hours later.
So i have to babysit all the time to prevent this from happening i have no clue if the often cancelling of units has any negative effect for the project.
And if you look at your stats it looks kinda weird so many cancelled out units.

Clownius
Send message
Joined: 19 Feb 09
Posts: 37
Credit: 30,657,566
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwat
Message 9115 - Posted: 29 Apr 2009 | 18:18:51 UTC

I have some of the longer KASHIF_HIVPR and TONI_HIVPR Wu's on my system now. They list at 10+ hours instead of 7-8 in boincs guestimate of runtime. I have also noticed in general my sytems slowing down with time. Wu's are creeping up. Im guessing its something to do with how long my system is on for between reboots. This is Vista im sure it has plenty of ways to spoil things. Really must get Linux dual booting on this thing and keep Vista for games.

Joe
Send message
Joined: 1 Sep 08
Posts: 37
Credit: 5,864,088
RAC: 0
Level
Ser
Scientific publications
watwatwatwatwat
Message 9117 - Posted: 29 Apr 2009 | 19:33:53 UTC - in response to Message 9115.

I have a few KASHIF_HIVPR_xxx WUs... It seems that they need triple time - but I earn only the double points... Is this possible?

Kind regards

Joe

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 9124 - Posted: 29 Apr 2009 | 20:18:30 UTC

@uBronan

Looking at the cpu time does not tell you everything. Go to "your account/my computer/pc xxxx/results" where you can click each of them. Near the bottom is the line "# Approximate elapsed time for entire WU: xxxxx s", which is what you should be looking at. The cpu time could depend on many things.
For example your current last WU lists 70870s, which is 19:41h - a nice time for a 9600GT. Which, by the way, is not the slowest card ;)

And you have problems to get cpu WUs for Milkyway because the ATI GPUs rip every single WU out of their hands, as soon as it's generated. It actually doesn't make much sense to run such WUs on cpus any more. What do you mean by VC?

To reduce your scheduling problems you should probably reduce your GPU-Grid ressource share to not more than 25% of what the cpus get over all projects or 20% of the entire share.

@GDF & all

I also had a "KASHIF_HIVPR" unit. It did run for 25h instead of the usual ~12h (108% longer). The credits do reflect this increase in time: 110% more. So things seem to be fine, except that I'm not sure if this was an intended change (maybe to reduce up- and download traffic and server load). I thought the previous balance of aiming for a runtime of 12h on an 9800GT was quite nice.

MrS
____________
Scanning for our furry friends since Jan 2002

uBronan
Avatar
Send message
Joined: 1 Feb 09
Posts: 139
Credit: 575,023
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 9132 - Posted: 29 Apr 2009 | 23:24:44 UTC
Last modified: 29 Apr 2009 | 23:43:27 UTC

@E.T. Ape VC me abit lazy means VideoCard and is almost the slowest then ;)
Hmmm i tend to forget to check inside the results indeed and seem allways to forget that the given times in boinc seems inaccurate.
Sadly the boincview program can't handle the gpu projects also and i can't find any other program which is able to make a nice overview like boincview did.
Browsing every result through these pages is not very handy either.

About this last unit boinc reports it took 24 hours to complete.
So i guess between boinc and the project seem to be a huge difference in reported time needed to complete the unit but i saw some restarts so i guess my system had made some not planned restarts.

MarkJ
Volunteer moderator
Volunteer tester
Send message
Joined: 24 Dec 08
Posts: 738
Credit: 200,909,904
RAC: 1
Level
Leu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 9142 - Posted: 30 Apr 2009 | 12:20:22 UTC

I did this KASHIF work unit. Took approx 7.5 hours on a GTX260+. The output file was 29Mb though, so they take a while to upload.
____________
BOINC blog

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1947
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 9143 - Posted: 30 Apr 2009 | 12:27:25 UTC - in response to Message 9142.

Some of the KASHIF_HIVPR came out a bit too large. There are two types of them. One is of normal length. Resubmitting would mean lose quite a bit of results, so we try to keep going for that batch.

gdf

mscharmack
Avatar
Send message
Joined: 20 Aug 07
Posts: 18
Credit: 1,319,274
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwat
Message 9144 - Posted: 30 Apr 2009 | 15:29:21 UTC
Last modified: 30 Apr 2009 | 15:30:53 UTC

Yes, I just completed one of them long ones, almost 54 hours of continuous processing on a 9600 GSO video card. No hope of any bonus because of the length of time to complete (still I got 8000+ for it). Looks like I'm going to have to update my video cards.

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 9153 - Posted: 30 Apr 2009 | 19:43:41 UTC - in response to Message 9144.

Looks like I'm going to have to update my video cards.


You don't have to.. the beautiful thing about mistakes is that there's always a finite chance they'll not be repeated regularly ;)

MrS
____________
Scanning for our furry friends since Jan 2002

Profile robertmiles
Send message
Joined: 16 Apr 09
Posts: 500
Credit: 576,513,997
RAC: 27,747
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 9224 - Posted: 2 May 2009 | 21:06:44 UTC - in response to Message 8788.

Forgive the dumb question, but just checking the obvious, easy stuff first:

You didn't accidently leave SLI turned on, right? That would reduce the number of GPUs available to CUDA down to 1.

Mike


How do you check the SLI setting and change it if needed? I've never seen it mentioned before, and thought I bought this computer with 2 GPUs included.

Profile robertmiles
Send message
Joined: 16 Apr 09
Posts: 500
Credit: 576,513,997
RAC: 27,747
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 9226 - Posted: 2 May 2009 | 21:26:50 UTC - in response to Message 9143.

Some of the KASHIF_HIVPR came out a bit too large. There are two types of them. One is of normal length. Resubmitting would mean lose quite a bit of results, so we try to keep going for that batch.

gdf


Have you thought of extending the deadline for the longer type in order to let participants return them after the previous deadline and still get credit? From what I've seen on other projects, telling the machine running the workunit about this change after the workunit was sent is not necessary - just changing it at your end.

Another idea is to allow participants to return workunits and get credit if EITHER of the following is true:

1. The results are valid and returned before the deadline.

2. The results are valid and returned before anyone the workunit was reassigned to manages to return valid results.

This will sometimes mean getting more sets of valid results from a workunit than you need; you can then discard any unwanted sets.

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 9247 - Posted: 3 May 2009 | 11:59:11 UTC - in response to Message 9226.

Another idea is to allow participants to return workunits and get credit if EITHER of the following is true:

1. The results are valid and returned before the deadline.

2. The results are valid and returned before anyone the workunit was reassigned to manages to return valid results.[/quote]

Thanks for the suggestion, it's already done this way :)

MrS
____________
Scanning for our furry friends since Jan 2002

Profile robertmiles
Send message
Joined: 16 Apr 09
Posts: 500
Credit: 576,513,997
RAC: 27,747
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13206 - Posted: 17 Oct 2009 | 4:21:40 UTC - in response to Message 9111.

Wow me scared if i ever got one of those :D
But i probably won't receive such units :D, because i think it will pass my deadline for at least 200 hours ;)
You have a powercard i have the slowest possible one xD
Btw i have also the problem to get my cpu's to run MW and get units on my VC,
i need to set the get work cache to 10 days to get enough work for my 4 cpu's or it runs dry for about 12 hours.
But on the other hand have the problem that i can handle only 1 unit at about 22 to 27 hours, when i leave it gpugrid ofcourse loads new and cancels them some hours later.
So i have to babysit all the time to prevent this from happening i have no clue if the often cancelling of units has any negative effect for the project.
And if you look at your stats it looks kinda weird so many cancelled out units.


Just noticed your message from months ago. If you haven't tried it yet, you might want to try my method of setting your computer location differently on some BOINC projects that work better with a short queue, and setting the preferences for that location to maintain a shorter queue. I'm currently using all four locations on the normal list of locations, even though all my computers are in one room.

[ESL Brigade] eimer
Avatar
Send message
Joined: 3 Feb 09
Posts: 62
Credit: 2,017,436
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwat
Message 13252 - Posted: 22 Oct 2009 | 3:13:28 UTC
Last modified: 22 Oct 2009 | 3:14:15 UTC

hiho,

when I request new WU's, i get 4 of a runtime about 10 hrs and 1 with an runtime of 5 hrs, whats about the long runtime WU's, from what i seen/think their credits aint that good as the short ones!

Richard Haselgrove
Send message
Joined: 11 Jul 09
Posts: 1405
Credit: 3,492,659,449
RAC: 386,566
Level
Arg
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13260 - Posted: 22 Oct 2009 | 22:11:43 UTC - in response to Message 13252.

When I request new WU's, i get 4 of a runtime about 10 hrs and 1 with an runtime of 5 hrs, whats about the long runtime WU's, from what i seen/think their credits aint that good as the short ones!

Likewise, my host 43404 got two tasks in the same request for work: the GIANNI-BIND (task 1410010) was estimated to take twice as long as the KASHIF-HIVPR (task 1409994). Yet, as you can see, they finished within 5% of the same time, and the credits were pretty closely in the same proportion - to answer the question in the other new thread.

The key word here, for newly-downloaded tasks, is "estimated". When I first joined this project, those initial estimates were far too low. Gradually, the estimates are being corrected: I think there must have been another correction between 8 October (when my KASHIF wu was created), and 16 October (when my GIANNI wu was created).

Don't fuss too much about the estimates. It's the final run time which matters.

Profile robertmiles
Send message
Joined: 16 Apr 09
Posts: 500
Credit: 576,513,997
RAC: 27,747
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13261 - Posted: 23 Oct 2009 | 2:00:32 UTC - in response to Message 13260.

I've found that upgrading to new BOINC version has a significant effect on the runtime estimates also, especially if you switch from the 6.6.36 version to any 6.10.* version.

Profile Jet
Send message
Joined: 14 Jun 09
Posts: 25
Credit: 5,835,455
RAC: 0
Level
Ser
Scientific publications
watwatwatwatwat
Message 13265 - Posted: 24 Oct 2009 | 8:18:54 UTC - in response to Message 13252.

100% agree with you. WU's becomes longer & given credit is far away from any proportion with shorter WU's.
Have complete this WU 881205, its processing was consumed 37K+seconds (well over 10 hours). Was prized by traditional 5369 credits.
It was discussed several time somewhere here, but wasn't any reasonable reply, what happens with credit system in GPUGRID. Looks like that almost clear: ratio GPU time \ Given credits is artificially deflated, w\o any explanation for participants.
This is volunteer project, so , no obligations...

Post to thread

Message boards : Graphics cards (GPUs) : Have work units gotten longer????

//