I attached to the GPUGrid project. Files were downloaded and work started on crunching numbers. Things went smoothly for about 1 hour. Then suddenly, BOINC manager reset the project and then detached me from it.

I reattached. Same smooth start. Same result. Reset and detached from the project. I'm running version 6.4.5 and an 8600 GT GPU. Any one else experience being rejected by the project for no apparent reason?

No, happily running on 3 systems ...

Can you post the relevant portions of the message log, say 20 lines above and below the detach?

No, happily running on 3 systems ...

Can you post the relevant portions of the message log, say 20 lines above and below the detach?

Thanks for your response. Here it is:

2/21/2009 10:25:03 PM|GPUGRID|Sending scheduler request: To fetch work. Requesting 45413 seconds of work, reporting 0 completed tasks
2/21/2009 10:25:08 PM|GPUGRID|Scheduler request completed: got 0 new tasks
2/21/2009 10:25:08 PM|GPUGRID|Message from server: No work sent
2/21/2009 10:25:08 PM|GPUGRID|Message from server: (reached per-CPU limit of 1 tasks)
2/21/2009 10:25:08 PM|GPUGRID|Message from server: (Project has no jobs available)
2/21/2009 10:27:33 PM|GPUGRID|Sending scheduler request: To fetch work. Requesting 45370 seconds of work, reporting 0 completed tasks
2/21/2009 10:27:38 PM|GPUGRID|Scheduler request completed: got 0 new tasks
2/21/2009 10:27:38 PM|GPUGRID|Message from server: No work sent
2/21/2009 10:27:38 PM|GPUGRID|Message from server: (reached per-CPU limit of 1 tasks)
2/21/2009 10:27:38 PM|GPUGRID|Message from server: (Project has no jobs available)
2/22/2009 12:51:34 AM|GPUGRID|Resetting project
2/22/2009 12:51:35 AM|GPUGRID|Detaching from project


The program got into a loop of requesting work every minute so I set it to "No New Tasks" which solved that problem. Sadly, the end result was the same.

@Gregg

Isn't a GeForce 8600 GT with only 32 shader units way too slow for this project? Please see this post.

@Gregg

Isn't a GeForce 8600 GT with only 32 shader units way too slow for this project? Please see this post.

I don't know. It is on the list as "It Works" although it is in white letters and the thread seems to directly address only gray, red and green letters. Maybe the 8600 GT was supposed to be in gray, not white letters.

If someone could state definitively that it doesn't have sufficient capacity, that would clear up the mystery.

Thanks.

Perhaps this discussion is helpful.

I think I'm seeing a pattern emerge. It appears that for GPUGRID purposes, my little old 8600 GT "may" work, but most likely won't. I will wait to reconnect to the project until I get an appropriate video card. Thanks to all for the help.

An 8600GT will complete work within the 4-day deadline, but the problem will come in as follows:

1. Though shorter workunits will complete in less than two-days, the 8600GT will take more than 48-hours to complete the longer variety of workunits which occur frequently (thus requiring a 24/7 runtime to begin with).

2. GPUGRID apps download based on the number of CPU cores, not GPU's. Thus, on your machine it will always try to keep two workunits downloaded.

3. If the combination is of two shorter units or one shorter unit and one longer unit, you should be fine completing both in under 4 days (or very close to that). The problem comes in when you get 2 longer workunits in a row (or more...3+ in a row is quite possible). When the latter happens, the amount of time that the card is late in returning work will increase, and you will have to abort one or more workunits to get back into an "on-time" cycle.

4. You can reduce this problem by overclocking somewhat (especially the shader speed), but no stable overclock will be fast enough on the 32 shader cards to be able to "set and forget" things...some degree of babysitting will always be the case.

A final note, a majority of the 8600 series cards have only 256mb (this is not the case with yours which has 512). Though not a problem now, GDF (a project Admin) has indicated that in the furture GPUGRID apps may require more memory.

@Gregg
If someone could state definitively that it doesn't have sufficient capacity, that would clear up the mystery.

My 8600GTS has the same number of shaders (32) as your 8600GT. Therefore you shouldn't have a problem... it won't be quick though, typical run time is around 36 hours but there are wu's of different lengths and credits around. Leave it 24 hours and make sure it is over 50% done. Check on it every day or so for the first few days if you like. As you have a core 2 duo Boinc will only let you have one gpu running and one queued so you should be able to do both within the 4 day deadline.

Phoneman1

A 8600GT can finish a WU within the deadline, but your PC will be almost unuseable when it runs and if you don't run 24/7 you might miss the deadlines.

MrS

---

Scanning for our furry friends since Jan 2002

Thanks Scott and Phoneman. Your comments were helpful. It appears that my card is in the gray area of being able to handle the load.

That said, I still don't understand why, after only a few hours of computing, the server detaches me from the project. It's not waiting a few days to see the status of the WU. Maybe it is keeping tabs on my progress and sees that I can't keep up.

That said, I still don't understand why, after only a few hours of computing, the server detaches me from the project.

I've never seen such behaviour before. Which BOINC version do you use? I'm a little worried about BOINC deciding to reset a project itself. Do that in a wrong moment and one may loose many WUs. That's not something I'd not (yet) trust a program to decide.

MrS

---

Scanning for our furry friends since Jan 2002

That said, I still don't understand why, after only a few hours of computing, the server detaches me from the project.

I've never seen such behaviour before. Which BOINC version do you use? I'm a little worried about BOINC deciding to reset a project itself. Do that in a wrong moment and one may loose many WUs. That's not something I'd not (yet) trust a program to decide.

MrS

GPUGRID is the only project that has done that to me in the many years that I have been involved with BOINC. In only started running GPUGRID yesterday.I was running 6.4.5 and it did it to me twice. I researched the forum and didn't find anything resembling my problem, but did see that some people were having trouble with aspects of 6.4.5 and it was suggested that 6.5.0 might fix their problem. I realize it is a pre-release version but decided to see if that would fix my problem. Just like it did with 6.4.5, 6.5.0 ran the project for about 2-3 hours and then detached me.

How did you install BOINC? in "protected mode", etc.?

Were you doing anything on the machine when it detached or was it just running?

What I find most interesting is that there is a communication and then *2 HOURS LATER* the detach message ...

Why the two hour delay? What else might have happened, what else may be running on the system?

I also wonder why there is no other messages for two hours ... I know I have busier systems, but I don't think I have two hour gaps in the messages ... ever ... even my slowest systems ...

How did you install BOINC? in "protected mode", etc.?

Were you doing anything on the machine when it detached or was it just running?

I've never installed anything in "protected mode." I just click the .exe file and let installer do its thing. It has flushed me a total of four times. I was probably just doing usual internet browsing during that time.

What I find most interesting is that there is a communication and then *2 HOURS LATER* the detach message ...

Why the two hour delay? What else might have happened, what else may be running on the system?

I also wonder why there is no other messages for two hours ... I know I have busier systems, but I don't think I have two hour gaps in the messages ... ever ... even my slowest systems ...

Regarding the lack of messages for two hours, I only posted the messages that were specific to the GPUGRID project. I was receiving normal message traffic regarding the other projects I was running during the two hours between attaching and getting detached. And, BOINC Manager showed that GPUGRID was running the entire time.

One last note. As you know, when the GPU is working on a project, its temperature jumps significantly (but stays well below a critical level). I noticed the last time I tried to get the project to run, that after running for about 20 minutes, the temperature on the GPU dropped back to its normal low-usage level. However, the BOINC Manager showed that the project was still running. It could not have been running given the magnitude of the temperature change of the GPU.

One last note. As you know, when the GPU is working on a project, its temperature jumps significantly (but stays well below a critical level). I noticed the last time I tried to get the project to run, that after running for about 20 minutes, the temperature on the GPU dropped back to its normal low-usage level. However, the BOINC Manager showed that the project was still running. It could not have been running given the magnitude of the temperature change of the GPU.

Now this sounds familiar as a similar thing has happened to me with my 9500GT. On it, the GPUGRID work would show running in BOINC manager, but the temperature monitor showed it as being basically idle and the progress bar stopped counting. A reboot got the machine working correctly again briefly, but something else went wrong and it dumped a full day's 8 workunits yesterday. Haven't checked it this evening yet to see if it has self corrected...though mine was just hung and not resetting or detaching??? It is a 32-bit Vista machine and the hang-ups happened when my kids switched users in an attempt to guess a password.

@Gregg,

Well, it is not supposed to happen ... but, the reason we wanted the messages, all the messages is to see what is happening on your system. What is happening on some other project may be causing the detach on GPU Grid. in other words, a cross-project bug ... project A does something that triggers the detach from project B ... it should not happen, and I am not saying it happens here ... but ... that gap is important ... what happened to GPU Grid does not look to me like something done by the GPU Grid project ...

OK, just a long shot here. Any chance these gentlemen/women are attached to a BOINC account manager like BAM?

If GPUGRID or any other project for that matter is not yet part of the projects they subscribed to in the relevant host, it will be automatically detached after the daily (or manually requested) update contacting the account manager.

This happened to me in the past for other projects.

Kind regards

Alain