There is a known BOINC work fetch issue on machines with multiple GPUs of a single type (ATI, nVidia, etc) when exclusion statements are present in a cc_config.xml file. that is to say that it generally happens when one tries to run different projects on different GPUs of the same type. the issue is that neither project will maintain a work buffer like its supposed to - rather it will not contact the server for new work until the entire queue has been crunched...this obviously defeats the purpose of a work buffer. to my knowledge, the problem was never fixed, or even addressed, despite there having been a thread about it some time ago. because that thread is well over 90 days old, it has been locked for quite some time...and yet no resolution was ever reached.

well i'm here to say that i upgraded to BOINC v7.2.28 just the other day, and upon trying to run Milkyway@Home on one of my HD 7970s and Einstein@Home on the other (in the same machine of course), the problem still persists - both projects crunch through their entire queues before they actually fetch any new work, making it impossible to "maintain" a constant work buffer. because there was quite a bit of testing involved in the original thread on the topic (WHICH CAN BE FOUND HERE), i propose that it be reopened, since i'm not entirely sure the issue has been fixed in BOINC v7.2.28.

TIA,
Eric

*EDIT* - it appears that i may be mistaken, and that it just looks like no resolution to the issue was reached b/c the original thread i referenced above ends quite abruptly with no explanation of a solution. that said, i just noticed that Milkyway@Home is fetching ~16 tasks at a time now...this is more tasks per fetch, and it fetches less often, but it appears that it may be maintaining a cache now...i'll have to confirm Einstein@Home still.

---

OK, after further scrutiny, i've found that both Milkyway@Home and Einstein@Home are maintaining work buffers. so i guess i raised a bit of a false alarm here with regard to the work fetch issue not having been fixed yet.

that said, perhaps the thread that i suggested should be resurrected should at least be appended by a moderator with a final post stating that the issue has in fact been fixed. i understand that those who just recently started crunching different projects on two or more GPUs of the same type/brand (ATI or nVidia), and therefore most likely have a BOINC version that alleviates the aforementioned work fetch bug, wouldn't have even known about the bug in the first place. but those who were around and already trying to crunch different projects on two or more GPUs of the same type before the bug got fixed have been left wondering 1) if the the bug has in fact been fixed, and 2) which version of BOINC finally fixed it. a simple appended post to the original (and now locked) thread addressing the work fetch bug would help those that are familiar with it and are still wondering if its been fixed...sure, those folks could also incrementally update BOINC until they see the work fetch bug disappear, but that's a ton of unnecessary work to solve a problem that's already been solved...

TIA,
Eric

---

I don't mind opening the thread --I have just done so-- the problem here is that you're then required to post in it within 6 hours, or else it'll get locked again by the autolock script.

---

thanks so much Ageless...

i guess the other downside to the fact that the thread will automatically lock again in 6 hours is that i'm not privy to the exact version of BOINC that truly contained the fix to the work fetch bug. i know for certain that the fix was first implemented into a version of BOINC that resides between 7.0.64 and 7.2.28, but that's as much as i know...

in the mean time, i will pot a brief comment to the original thread that says what's essentially been said here. that way anyone who searches for the work fetch bug will at least find that thread and have an explanation readily available at the very end of the thread.

thanks again,
Eric

---