Advanced search

Message boards : Graphics cards (GPUs) : Why don't my Work Units ever finish?

Author Message
garycgreen
Send message
Joined: 14 Feb 09
Posts: 2
Credit: 0
RAC: 0
Level

Scientific publications
wat
Message 11162 - Posted: 17 Jul 2009 | 4:11:27 UTC
Last modified: 17 Jul 2009 | 4:15:06 UTC

I've tried at least six or seven times now over the past three months and they always run way out of time. I have only a GeForce 8500GT, driver version 18250,compute capability 1.1, 512 MB, est 5GFlops. BOINC 6.6.36.

Is it time for me to give up?
____________

Betting Slip
Send message
Joined: 5 Jan 09
Posts: 670
Credit: 2,498,095,550
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 11165 - Posted: 17 Jul 2009 | 10:06:10 UTC - in response to Message 11162.

Firstly 8500GT is not a fast card and while capable of running this project is not a recommended card.

However to address your issue. Do have your computer on 24/7 if not there isn't a hope in hell of you ever completing on time without upgrading.

Do you allow GPU use when computer is in use? If no, how much do you use the computer and how long does it have to be idle before GPU can be used?

I have an 8600GT (at least three times as fast as your 8500GT) which I don't attach to this project because of the fear it won't complete in time and I do have computers on 24/7.

This not a criticism of your graphics card just a painful fact.
____________
Radio Caroline, the world's most famous offshore pirate radio station.
Great music since April 1964. Support Radio Caroline Team -
Radio Caroline

Bruce Downing
Send message
Joined: 20 Jul 09
Posts: 14
Credit: 263,684,863
RAC: 103,259
Level
Asn
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 11205 - Posted: 20 Jul 2009 | 17:32:11 UTC

I also have work units which take forever. I am new to CUDA, so bear with me!
I started with a 9500, way too slow. So, I returned it to Dell and got a 9800GT, but it's no better. The first task I got from GPUGRID has been running for 8 hours , 42% completed, and 8 more hours remaining. Is there a card out there which will do the job AND be compatible with my Dell 530? Here are the messages:

7/20/2009 8:56:50 AM Processor: 4 GenuineIntel Intel(R) Core(TM)2 Quad CPU Q9300 @ 2.50GHz [Intel64 Family 6 Model 23 Stepping 7]
7/20/2009 8:56:50 AM Processor features: fpu tsc pae nx sse sse2 pni mmx
7/20/2009 8:56:50 AM OS: Microsoft Windows Vista: Home Premium x64 Edition, Service Pack 2, (06.00.6002.00)
7/20/2009 8:56:50 AM Memory: 6.00 GB physical, 12.10 GB virtual
7/20/2009 8:56:50 AM Disk: 696.57 GB total, 604.31 GB free
7/20/2009 8:56:50 AM Local time is UTC -4 hours
7/20/2009 8:56:50 AM CUDA device: GeForce 9800 GT (driver version 18618, compute capability 1.1, 1024MB, est. 55GFLOPS)
7/20/2009 8:56:50 AM Not using a proxy

Help! (and thanks)

Bruce D.

Betting Slip
Send message
Joined: 5 Jan 09
Posts: 670
Credit: 2,498,095,550
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 11209 - Posted: 20 Jul 2009 | 20:28:22 UTC - in response to Message 11205.
Last modified: 20 Jul 2009 | 20:31:13 UTC

I also have work units which take forever. I am new to CUDA, so bear with me!
I started with a 9500, way too slow. So, I returned it to Dell and got a 9800GT, but it's no better. The first task I got from GPUGRID has been running for 8 hours , 42% completed, and 8 more hours remaining. Is there a card out there which will do the job AND be compatible with my Dell 530? Here are the messages:

7/20/2009 8:56:50 AM Processor: 4 GenuineIntel Intel(R) Core(TM)2 Quad CPU Q9300 @ 2.50GHz [Intel64 Family 6 Model 23 Stepping 7]
7/20/2009 8:56:50 AM Processor features: fpu tsc pae nx sse sse2 pni mmx
7/20/2009 8:56:50 AM OS: Microsoft Windows Vista: Home Premium x64 Edition, Service Pack 2, (06.00.6002.00)
7/20/2009 8:56:50 AM Memory: 6.00 GB physical, 12.10 GB virtual
7/20/2009 8:56:50 AM Disk: 696.57 GB total, 604.31 GB free
7/20/2009 8:56:50 AM Local time is UTC -4 hours
7/20/2009 8:56:50 AM CUDA device: GeForce 9800 GT (driver version 18618, compute capability 1.1, 1024MB, est. 55GFLOPS)
7/20/2009 8:56:50 AM Not using a proxy

Help! (and thanks)

Bruce D.


Hi
Your 9800GT will do fine on this project as finishing a WU in 24hrs is great. However because of this project (due to an insuffiency in BOINC apparently) giving you a WU per CPU core you have 3 waiting while you crunch the first and when you finish that and report it, it will download another one so that you have 3 waiting again, a recipe for missing the deadline to say nothing of your 25% credit bonus for returning within 48hrs.
The best thing you can do with your existing card is set GPU Grid in BOINC Manager to NO NEW TASKS until you finish all 4 that you have already. It should still report them as each one finishes (if it doesn't do it manually). When the last one finishes set project to ALLOW NEW TASKS and it will download another one as soon as it starts downloading a WU set to NO NEW TASKS again because if you don't you will get 4 again...rinse and repeat.
To answer your question about a new card fast enough to avoid this problem you will need a GTX275 or maybe get away with GTX 260 and set to crunch while computer in use as these cards can handle it (especially 275 as long as your NOT gaming). The only problem that I can see is that you have a Dell.
Dell sometimes use NON STANDARD cases and more importantly PSU's so be careful your PSU is powerfull enough for these new cards with right connectors and that they will fit in the case. Next time don't buy Dell (my opinion)
Hope this helps.
____________
Radio Caroline, the world's most famous offshore pirate radio station.
Great music since April 1964. Support Radio Caroline Team -
Radio Caroline

Profile Hydropower
Avatar
Send message
Joined: 3 Apr 09
Posts: 70
Credit: 6,003,024
RAC: 0
Level
Ser
Scientific publications
watwatwatwatwatwat
Message 11210 - Posted: 20 Jul 2009 | 21:13:53 UTC - in response to Message 11209.

You might try adding
<ncpus>2</ncpus>
to your cc_config.xml file while BOINC is not running and then restart BOINC. If you're lucky it will reduce the number of perceived CPUs to 2 (or whatever number you type under 5) so your queue size would be reduced.

Betting Slip
Send message
Joined: 5 Jan 09
Posts: 670
Credit: 2,498,095,550
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 11217 - Posted: 21 Jul 2009 | 8:44:36 UTC - in response to Message 11210.

You might try adding
<ncpus>2</ncpus>
to your cc_config.xml file while BOINC is not running and then restart BOINC. If you're lucky it will reduce the number of perceived CPUs to 2 (or whatever number you type under 5) so your queue size would be reduced.


Good solution but if he is running other BOINC projects that use CPU he will only utilize 2 of his 4 cores.
____________
Radio Caroline, the world's most famous offshore pirate radio station.
Great music since April 1964. Support Radio Caroline Team -
Radio Caroline

Profile Hydropower
Avatar
Send message
Joined: 3 Apr 09
Posts: 70
Credit: 6,003,024
RAC: 0
Level
Ser
Scientific publications
watwatwatwatwatwat
Message 11235 - Posted: 21 Jul 2009 | 21:20:16 UTC - in response to Message 11217.

I know, but at least he will have a better chance at finishing GPUGrid packets this way. I just thought of another solution, setting the CPU %usage for each project in such a way that there is a 50/50 share between this project and other projects. That will allow him to use the other cores for other projects and for example two cores for GPUGrid. This may need some testing, tweaking and experimenting.

Thanks for making me think creatively.
____________
Join team Bletchley Park, the innovators.

garycgreen
Send message
Joined: 14 Feb 09
Posts: 2
Credit: 0
RAC: 0
Level

Scientific publications
wat
Message 11267 - Posted: 23 Jul 2009 | 4:11:27 UTC - in response to Message 11165.

Yes, I was afraid that the card is just too tiny. The machine is on 24x7 and the card is "allowed" all the time, but I'm also running SETI projects on the GPU, which halves the time available for GPUGRID.

I guess I'll just have to give up on GPUGRID...the SETI workunits just fly by, but these are a lot bigger, I guess.

Thanks, guess I'll be back when I upgrade the card!

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 11762 - Posted: 9 Aug 2009 | 14:25:44 UTC

Guys.. a 9800GT is fine for the project and all you have to do to get the bonus credits is to adjust ressource share and cache size.

MrS
____________
Scanning for our furry friends since Jan 2002

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 11812 - Posted: 11 Aug 2009 | 22:12:45 UTC - in response to Message 11162.

I've tried at least six or seven times now over the past three months and they always run way out of time. I have only a GeForce 8500GT, driver version 18250,compute capability 1.1, 512 MB, est 5GFlops. BOINC 6.6.36.

Is it time for me to give up?


Yes!
When I first attached to this project I struggled to get one result with an 8600GT and that had a Boinc rating of 14GFlops. You are wasting your time. If your card has <30GFlops you will struggle. I expect the tasks are longer now than when I started, so you might even struggle with 40GFlops cards.

For anyone who is struggling to finish tasks because Boinc has downloaded 4 tasks, it’s likely to be because you are using out of date software or your boinc setting allow you to cache too much work. A few hours of cached tasks is plenty. If you have 4 GPUGRID tasks running and they keep hopping between each, and your system is not always on, you may not going to get them all finished unless you have a very high end card! If you are also crunching other GPU tasks such as Seti or Aqua you have less chance of finishing tasks.
For anyone in this predicament I would recommend you select no more tasks (for the CUDA projects) and let the existing work units finish. Try to use suspend task and then enable it when the other task is complete (if it can finish on time). Then upgrade Boinc, your card drivers and configure your settings for say 0.2 days or 1 day of cached work.

Post to thread

Message boards : Graphics cards (GPUs) : Why don't my Work Units ever finish?

//