For testing purposes I installed 6.4.2. It had the problem that it never requested any GPU work voluntarily, not before I suspended the CPU projects. I accidently had a massive stockpile of CPU work, though.

Now I had a strange situation: I had 14 Milkyway WUs at 48 minutes each.. that's <3h of processing for my 4 cores. No other WUs were running except one GPU-Grid with an estimated runtime of 2h left. Cache setting is 1.25 days and ressource share 50% Milkyway and 37,5% GPU-Grid. Why the hell didn't BOINC request new work? I'm a little worried I'll still have to babysit it..

Is there any client which is more well-behaved? I did not really hear good stuff about the later ones.

MrS

I does not request new work because it make no distinction between LTD accumulated by the GPU vs the CPU ...

Which is why I tried to make a point about that in this latest fix cycle. We need to be able to properly establish the correct resource shares, THEN make accurate measurements of capabilities, THEN we can accumulate LTD correctly for each resource ... THEN we can fetch work properly ...

THe short answer ... set the R Share to 500 and try again ... I seem to be doing OK with RS of 100 on the i7 for the moment ... of course, I also don't mind going in and resetting LTD to 0 as I have to on that machine to make all things equal again ...

Oh, the other part of the short answer ... yes, you will need to babysit if for no other reason than the site does not issue work correctly ... but that is just my experience ... for you to get work it seems you have to be out or close to out of work before the scheduler issues work ... then you can get, assuming it is sending any at all, two or three tasks one after the other ... then ... tomorrow you get to bite your nails again ...

I've been using 6.5.0 for about two weeks. It seems to keep GPU tasks and CPU tasks in step (providing there is GPU work available).

Only downside I'm seeing (which may or may not be the Boinc client's fault) is the DCF drifts up to around 13.5 after a few units have been completed and stays around there. The estimated time in BoincView is currently 15 days+ ... but it still gets work when it is available. For the last 10 days or so I've stopped resetting the DCF.

I'm running resource shares of 100 / 100 / GPU / Cosmology with 0.5 days cache. I usually have 1 GPU task running and either 2 or three queued on the machine with the GTX260 card. The machine with the 8600GTS card normally has 1 or 2 GPU tasks queued and one running. Both machines are Vista HP 32 bit.

Boinc won't over-commit itself and at the moment treats GPU tasks like any other for work fetch. So your current work buffer must be able to be completed within the individual project deadlines before it will get any more. So, for example, if you had a 1 day cache and ran 4 other projects each of which had a four day deadline it wouldn't request a GPU task until there was time available and the ratio of LTD had shifted in GPU's favour. Paul's suggestion slants things more in GPU's favour but until time becomes available it won't request work. That is why suspending other projects and manually updating GPU brings a request for > 0 seconds of work. But even with a >0 seconds request you may not get any tasks....

A lot of the present work fetch problems though is because GPU does not seem to have work available. The web site doesn't help, it doesn't separate PS3 jobs ready to send from GPU jobs ready to send so people see the number ready to send and get very frustrated!

Phoneman1

Somehow my 6.4.2 is totally confused. It's got 24 Milkyways now, which amounts to 4.8h of work on my CPU and 9h when factoring in the 50% resource share. There's also a GPU-WU at estimated 10h cpu time. Adding the 10h of GPU-Grid I get a total time of 7.3h. Nevermind the one WU at 10h.. this tells me that all Milkyways will be done in less than 7.3h. They have a 3 days deadline (just got them today) .. but still BOINC runs them in high priority mode. WTF?! My on fraction etc. looks alright to me.

Guess I should try 6.5.0. But then BOINC will only have 1 GPU-Grid WU at a time due to the 15 days estimate, wouldn't it?

MrS

---

Scanning for our furry friends since Jan 2002

6.5.0 gave me two when I added a new (albeit slow - 9500GT) card, so at least initially you should get multiple workunits.

Somehow my 6.4.2 is totally confused. It's got 24 Milkyways now, which amounts to 4.8h of work on my CPU and 9h when factoring in the 50% resource share. There's also a GPU-WU at estimated 10h cpu time. Adding the 10h of GPU-Grid I get a total time of 7.3h. Nevermind the one WU at 10h.. this tells me that all Milkyways will be done in less than 7.3h. They have a 3 days deadline (just got them today) .. but still BOINC runs them in high priority mode. WTF?! My on fraction etc. looks alright to me.

Guess I should try 6.5.0. But then BOINC will only have 1 GPU-Grid WU at a time due to the 15 days estimate, wouldn't it?

MrS

I have not figured out the rules used ... my 9800 when it is allowed to pull work usually gets two ... at 0.5 buffer and only GPU Grid active the 280 pulls 3, at 1.0 it pulled 6 (giving me 5 spare) ...

But, I have also pulled those same numbers with all tasks in the buffer adding up to a couple days worth of work too ...

RIght now the i7 has buffer of 1.0 and is running SIMAP for which it has a least a days worth of work. Also AI, and Cosmology have a full days work ... along with WCG's days work ... giving me 5 days? I don't know ... heck I just want to keep it busy ...

I thought the memory leak had something to do with the drivers and only on XP 64? On Linux and Vista 64 I never saw something about memory leaks...

Dukzilla, since you use XP 64bit, could you please install the 180.84 Nvidia drivers and try again?

I don't know what I did right or wrong but its working now. I woke up this morning to find my 2nd monitor wouldn't accept a setting above 1024x768. The order was also backwards. I rebooted (for the 3rd time in 24 hours). The 2nd monitor came up normally and in the right position. I decided to do a manual update for the project. I watched the message log. It gave me the same initial error messages BUT it downloaded 4 work units. Lo and Behold, its running. I don't know what the heck is going on. I don't care. Its working. Mostly. Shouldn't I see a total of 5 processes running. I have an AMD 9850 quad core with 4GB RAM and the GPU with 512MB.

Is there a setting I need to adjust to run all 5? I've checked all my projects and they are set to 8 CPU. Also, should I see the message (0.05 CPUs, 1CUDA) at the end of the line for the one running GPUGRID work unit?

I'll be really happy when I actually complete a work unit.

Please bear with the newbie questions. I want to get the most out of this project. It will help both of us.

Thanks again for all of your help.

Greg

BTW - I went through the automatic update process and it said I already have the latest drivers. I'm going to do a manual update and see if that helps.

Depending on the version of BOINC you are running you will see just the same number as your cores or one more ... There are ways to get the earlier version to use the cores plus one or you could try 6.5.0 which will run plus one out of the box.

There are still server issues and the BOINC Version from my observation makes no difference as to that ...

The message (0.05...) is what you should see for the one task running on the GPU...

There are reports on SaH where the death of the task can cause the GPU to go nuts ... Yours is the first report of that for GPU Grid that I have seen ... so, something unusual is going on ...

The best version I have reports on and that *I* personally are using for the Nvidia drivers is 180.48 you may want to snag those and install them. Just like with games, the latest video drivers got real important with BOINC.

I too will be happy when you complete a task ...

The best version I have reports on and that *I* personally are using for the Nvidia drivers is 180.48 you may want to snag those and install them...

Yea, it's good that this driver is working fine for you, but with XP 64bit he should really use 180.84. See this forum thread...

---

pixelicious.at - my little photoblog

Depending on the version of BOINC you are running you will see just the same number as your cores or one more ... There are ways to get the earlier version to use the cores plus one or you could try 6.5.0 which will run plus one out of the box.

There are still server issues and the BOINC Version from my observation makes no difference as to that ...

The message (0.05...) is what you should see for the one task running on the GPU...

There are reports on SaH where the death of the task can cause the GPU to go nuts ... Yours is the first report of that for GPU Grid that I have seen ... so, something unusual is going on ...

The best version I have reports on and that *I* personally are using for the Nvidia drivers is 180.48 you may want to snag those and install them. Just like with games, the latest video drivers got real important with BOINC.

I too will be happy when you complete a task ...

I finally completed a task and manually updated so it would post. The remaining tasks all say ~20 hours to complete. This is up from the 2 hours is was waying yesterday. I updated the drivers to 180.48 when I was going through this process with SETI. I did it again last night just to make sure.

My next update will be to 6.5.0. Maybe that will make a difference.

Greg

Hi there,
everything went fine for me, until i tried to OC my 9800GTX+ 1GB...
From that time, not a single ping. I try all mentioned and go to defauls... :-(
I have one more 9800GTX 512 in my bag now, If there will be more troubles than poins in this projec, I will sell it and buy some second hand QUAD instead of it...

---

I went to default clocks on my card, stopped other projects, restarted computer, then it loaded from GPUGRID some tests and now i have 4 new WU... :-D
Great. Let us see the future...
(running 6.4.5)

6.1.2009 11:48:26||Starting BOINC client version 6.4.5 for windows_intelx86
6.1.2009 11:48:26||log flags: task, file_xfer, sched_ops
6.1.2009 11:48:26||Libraries: libcurl/7.19.0 OpenSSL/0.9.8i zlib/1.2.3
6.1.2009 11:48:26||Data directory: C:\Documents and Settings\All Users\Data aplikací\BOINC
6.1.2009 11:48:26||Running under account Garfield
6.1.2009 11:48:27||Processor: 4 GenuineIntel Intel(R) Core(TM)2 Quad CPU Q9550 @ 2.83GHz [x86 Family 6 Model 23 Stepping 7]
6.1.2009 11:48:27||Processor features: fpu tsc pae nx sse sse2 mmx
6.1.2009 11:48:27||OS: Microsoft Windows XP: Home x86 Editon, Service Pack 3, (05.01.2600.00)
6.1.2009 11:48:27||Memory: 3.25 GB physical, 3.09 GB virtual
6.1.2009 11:48:27||Disk: 3.72 GB total, 2.02 GB free
6.1.2009 11:48:27||Local time is UTC +1 hours
6.1.2009 11:48:27||Not using a proxy
6.1.2009 11:48:28||CUDA devices found
6.1.2009 11:48:28||Coprocessor: GeForce 9800 GTX/9800 GTX+ (1)
6.1.2009 11:48:28|Milkyway@home|URL: http://milkyway.cs.rpi.edu/milkyway/; Computer ID: 42149; location: (none); project prefs: default
6.1.2009 11:48:28|QMC@HOME|URL: http://qah.uni-muenster.de/; Computer ID: 131092; location: (none); project prefs: default
6.1.2009 11:48:28|SETI@home|URL: http://setiathome.berkeley.edu/; Computer ID: 4738967; location: (none); project prefs: default
6.1.2009 11:48:28|GPUGRID|URL: http://www.gpugrid.net/; Computer ID: 21614; location: (none); project prefs: default
6.1.2009 11:48:28||No general preferences found - using BOINC defaults
6.1.2009 11:48:28||Reading preferences override file
6.1.2009 11:48:28||Preferences limit memory usage when active to 2994.28MB
6.1.2009 11:48:28||Preferences limit memory usage when idle to 3160.62MB
6.1.2009 11:48:28||Preferences limit disk usage to 1.86GB
6.1.2009 11:48:29|GPUGRID|Sending scheduler request: To fetch work. Requesting 207360 seconds of work, reporting 0 completed tasks
6.1.2009 11:48:32||Project communication failed: attempting access to reference site
6.1.2009 11:48:52|GPUGRID|Scheduler request failed: Couldn't resolve host name
6.1.2009 11:48:53||Internet access OK - project servers may be temporarily down.
6.1.2009 11:49:52|GPUGRID|Sending scheduler request: To fetch work. Requesting 207360 seconds of work, reporting 0 completed tasks
6.1.2009 11:49:57|GPUGRID|Scheduler request completed: got 2 new tasks
6.1.2009 11:50:00|GPUGRID|Started download of VsX9714-SH2_US-11-40-SH2_US240000-LICENSE
6.1.2009 11:50:00|GPUGRID|Started download of VsX9714-SH2_US-11-40-SH2_US240000-COPYRIGHT
6.1.2009 11:50:02|GPUGRID|Finished download of VsX9714-SH2_US-11-40-SH2_US240000-LICENSE
6.1.2009 11:50:02|GPUGRID|Finished download of VsX9714-SH2_US-11-40-SH2_US240000-COPYRIGHT
6.1.2009 11:50:02|GPUGRID|Started download of VsX9714-SH2_US-11-40-SH2_US240000-VsX9714-SH2_US-10-40-SH2_US240000_1
6.1.2009 11:50:02|GPUGRID|Started download of VsX9714-SH2_US-11-40-SH2_US240000-VsX9714-SH2_US-10-40-SH2_US240000_2
6.1.2009 11:50:09|GPUGRID|Finished download of VsX9714-SH2_US-11-40-SH2_US240000-VsX9714-SH2_US-10-40-SH2_US240000_1
6.1.2009 11:50:09|GPUGRID|Finished download of VsX9714-SH2_US-11-40-SH2_US240000-VsX9714-SH2_US-10-40-SH2_US240000_2
6.1.2009 11:50:09|GPUGRID|Started download of VsX9714-SH2_US-11-40-SH2_US240000-VsX9714-SH2_US-10-40-SH2_US240000_3
6.1.2009 11:50:09|GPUGRID|Started download of VsX9714-SH2_US-11-40-SH2_US240000-complex_full.sol.ionized.pdb
6.1.2009 11:50:13|GPUGRID|Finished download of VsX9714-SH2_US-11-40-SH2_US240000-VsX9714-SH2_US-10-40-SH2_US240000_3
6.1.2009 11:50:13|GPUGRID|Started download of VsX9714-SH2_US-11-40-SH2_US240000-complex_full.sol.ionized.psf
6.1.2009 11:50:31|GPUGRID|Finished download of VsX9714-SH2_US-11-40-SH2_US240000-complex_full.sol.ionized.pdb
6.1.2009 11:50:31|GPUGRID|Started download of VsX9714-SH2_US-11-40-SH2_US240000-parameters
6.1.2009 11:50:31|GPUGRID|Sending scheduler request: To fetch work. Requesting 205510 seconds of work, reporting 0 completed tasks
6.1.2009 11:50:42|GPUGRID|Finished download of VsX9714-SH2_US-11-40-SH2_US240000-parameters
6.1.2009 11:50:42|GPUGRID|Started download of VsX9714-SH2_US-11-40-SH2_US240000-SH2_US240000
6.1.2009 11:50:42|GPUGRID|Scheduler request completed: got 2 new tasks

---

My next update will be to 6.5.0. Maybe that will make a difference.
Greg

The update (and several reboots) seems to have resolved most of the issues with S@H and GPUGRID. I can now process work units for both. Sometimes. I'm still getting periods where all the remaining CUDA work units fail with Computation Error. A reboot seems to fix this. I have to reboot not just restart the BOINC client. I'll keep monitoring the boards for both projects and see if a solution presents itself. Thanks for all of your help so far.

Greg

My next update will be to 6.5.0. Maybe that will make a difference.
Greg

The update (and several reboots) seems to have resolved most of the issues with S@H and GPUGRID. I can now process work units for both. Sometimes. I'm still getting periods where all the remaining CUDA work units fail with Computation Error. A reboot seems to fix this. I have to reboot not just restart the BOINC client. I'll keep monitoring the boards for both projects and see if a solution presents itself. Thanks for all of your help so far.

Greg

I recall notes on SaH that this is a common problem with their application, that once it crashes in some particular way, which may not bring down the system, all remaining work will be CE ...

Just a suggestion, try running only one project at a time. I have had a couple tasks here that died when I first started and zero since ... not conclusive, but nearly two solid weeks I think it has been since my last crash ... so, at the moment, I think the weakness is still on the SaH end ... YMMV

Geez, 6.4.2 is so nuts! I'll try 6.5.0 now..

Twice this week BOINC decided it would be a good idea not to fetch any CPU work and just run dry until I return from work and discover the disaster.

MrS

---

Scanning for our furry friends since Jan 2002

Geez, 6.4.2 is so nuts! I'll try 6.5.0 now..

Twice this week BOINC decided it would be a good idea not to fetch any CPU work and just run dry until I return from work and discover the disaster.

MrS

6.5.0 *MAY* work better, or the same ...

The problem you are likely running into is the work fetch policy not making a distinction between work done on the CPU vs work done on the GPU ...

I configured for resource share of 200 (high, but not highest) and 0.4 esxtra buffer ... and I still watch it like a hawk ... at times I have gone in and reset LTDs to zero to force more work fetch ...

Not sure when we may expect the next version of BOINC Manager that begins to address this ... rumor has mid next month which likely means later than that ...

{edit}Of course I am at home all the time and I have the luxury of being able to hover over my systems ... and not being able to sleep means that I do it at all hours of the day and night ... :) {/edit}

I'm happy with the change: 6.5.0 behaves just as expected for me!

I don't have to suspend all cpu tasks any more to make it request new work and it actually tries to establish a cash of the size I request, not only a few hours like 6.4.2 did.

Edit:

The problem you are likely running into is the work fetch policy not making a distinction between work done on the CPU vs work done on the GPU

Yes, that was definitely a reason, coupled with buggy code.

MrS

---

Scanning for our furry friends since Jan 2002

after i had the same issues as many here i found myself with help of search engine copy of boinc 6.5.0 and yes all works fine again
version 6.4.5 gave so many problems with other applications that i even tried the newest but those did not really run nicely as well including 6.6.7
i am amazed to see this version running so smooth

after i had the same issues as many here i found myself with help of search engine copy of boinc 6.5.0 and yes all works fine again
version 6.4.5 gave so many problems with other applications that i even tried the newest but those did not really run nicely as well including 6.6.7
i am amazed to see this version running so smooth

Here is a BOINC "tip", download a version of BOINC and try it ... if it works, KEEP A COPY OF THE INSTALLER ...

Keep at least 2 versions back ... 6.5.0 works well, so does 5.10.45 and so I keep copies of both ...

When I find another stable copy and it has run for several months I will think about deleting my old copies of 5.10.45 ...

But, I don't expect a stable version of BOINC for awhile ... and it is more than likely to NOT be 6.7.0 ...

I was thinking that 6.4.5 was a stable version untill i tried to install the sse optimized clients.
All optimized apps are being totally ignored by this version, when i placed the files in the maps where they must be they did not work.
I was thinking it was the applications who where faulty but after i downloaded other optimized applications found that the issue was with them also.
So i took my old 4.2.19 and installed it and direct after i was done all the optimized applications started :)
Ofcourse i reinstalled the 6.4.5 version to check and poef gone optimized applications and back to originals.
So yes i agree with you i am gonna keep a bit more older versions since newer ones and also the recommended ones not work like they should.
For now i think 6.5.0 runs the best :D

Wow, a 4-series client?! Milkyway optimized certainly works with 6.5.0, so 6.4.5 looks even worse if it doesn't support the anonymous platform properly.

MrS

---

Scanning for our furry friends since Jan 2002