Message boards : Graphics cards (GPUs) : Development BOINC 6.10.9 released
| Author | Message |
|---|---|
|
MarkJ  Volunteer moderator Volunteer tester Send message Joined: 24 Dec 08 Posts: 738 Credit: 200,909,904 RAC: 0 Level  Scientific publications       | |
|
Another new one. Currently just for Windows. Below is the un-offical change log. Changes for 6.10.8 ____________ BOINC blog | |
ID: 12762 | Rating: 0 | rate:
 /   | |
|
Very soon to be superseded, I hope. I just fired a patch at the dev mailing list. (Turn off coproc_debug if you run 6.10.9 and have more than a single GPU to avoid the startup crash!) | |
| ID: 12764 | Rating: 0 | rate:
/ | |
|
BarryAZ Send message Joined: 16 Apr 09 Posts: 163 Credit: 920,275,294 RAC: 0 Level  Scientific publications | |
|
Right, the .8 to .9 move -- I'll pass and fight with the .7 release for now. I REALLY don't expect the developers to fix the design flaw causing the sort of work fetch problem I posted on the 6.10.7 thread on this board. To approach correcting that requires a *policy* change by folks running the developers. Very soon to be superseded, I hope. I just fired a patch at the dev mailing list. (Turn off coproc_debug if you run 6.10.9 and have more than a single GPU to avoid the startup crash!) | |
| ID: 12767 | Rating: 0 | rate:
/ | |
To approach correcting that requires a *policy* change by folks running the developers. That's way too political for me! I'll stick to testing for bugs that cause crashes. ;) ____________ Crunching on Linux: Fedora 11 x86_64 / nVidia 185.18.36 driver / CUDA 2.2  | |
| ID: 12768 | Rating: 0 | rate:
/ | |
|
CTAPbIi Send message Joined: 29 Aug 09 Posts: 175 Credit: 259,509,919 RAC: 0 Level  Scientific publications | |
|
MarkJ, | |
| ID: 12769 | Rating: 0 | rate:
/ | |
 Paul D. BuckSend message Joined: 9 Jun 08 Posts: 1050 Credit: 37,321,185 RAC: 0 Level  Scientific publications | |
MarkJ, The only place you stand a chance is the Alpha or developer's mailing list. But there it is more 90/10 9or worse) you will be ignored anyway ... It is a free sign-up ... but, most posts seem to be ignored at best, belittled at worst ... And depending on the GPU science application you can see the "hangs" in windows ... I have been running GPU Grid on a dual GTX260 rig with never a hang ... there is a new GPU app for CUDA on MW and while it is running the computer is nigh unusable ... of course I have it to run even when I am using it ... but this is the first I have seen it this bad ... UCB I doubt will see either a bug ... the suggestion will be to not run GPU while computer is in use .. which would mean that every time I bump the mouse I would lose work ... but for someone that does not use BOINC, works fine ... :) | |
| ID: 12771 | Rating: 0 | rate:
/ | |
|
CTAPbIi Send message Joined: 29 Aug 09 Posts: 175 Credit: 259,509,919 RAC: 0 Level Scientific publications | |
|
...mmm...pity... | |
| ID: 12772 | Rating: 0 | rate:
/ | |
Paul, I hope I'm not speaking out of turn and you don't take offence, but it seems you are so disillusioned with BOINC it's going to give you a stomach ulcer. Maybe you need a break from it?
Erm, last time I looked at this code (at least for 6.10 branch), I don't see how moving the mouse should be interpreted as a another graphics app is running on the GPU. CTAPbIi, please start a new thread. What is hanging, the entire desktop, manager app, or something else. What distro, kernel, CUDA version, boinc version, GPU hardware, etc. etc. I'll jump in and help if I can. ____________ Crunching on Linux: Fedora 11 x86_64 / nVidia 185.18.36 driver / CUDA 2.2 | |
| ID: 12773 | Rating: 0 | rate:
/ | |
|
MarkJ Volunteer moderator Volunteer tester Send message Joined: 24 Dec 08 Posts: 738 Credit: 200,909,904 RAC: 0 Level Scientific publications | |
MarkJ, Actually i'm not. As it says the best place to report issues are to the BOINC Alpha mailing list. If you wish to join the mailing list you can do so here I'm pissed off with the freezes while running GPUGRID. I tried to take off "use GPU while in use" either via web either in the client itself. it continues to freeze. from other hand, I asked this question here and buddy said that this stuff works fine in windows, but i'm running linux. As you can see from the change log there is a change for Linux priorities, however they haven't released a Linux build yet. ____________ BOINC blog | |
| ID: 12774 | Rating: 0 | rate:
/ | |
|
MarkJ Volunteer moderator Volunteer tester Send message Joined: 24 Dec 08 Posts: 738 Credit: 200,909,904 RAC: 0 Level Scientific publications | |
And depending on the GPU science application you can see the "hangs" in windows ... I have been running GPU Grid on a dual GTX260 rig with never a hang ... there is a new GPU app for CUDA on MW and while it is running the computer is nigh unusable ... of course I have it to run even when I am using it ... but this is the first I have seen it this bad ... Now hold on a minute. If the MW app has issues like that its clearly not BOINC's fault. I would suggest you report it to the MW developers so they can look into it. ____________ BOINC blog | |
| ID: 12776 | Rating: 0 | rate:
/ | |
As you can see from the change log there is a change for Linux priorities, however they haven't released a Linux build yet. I actually wrote and submitted that patch to the dev list and it should help make the desktop on an interactive (rather than headless) cruncher appear more responsive under heavy load, by setting the kernel scheduler policy for CPU tasks to BATCH. It does not affect the priority of GPU tasks. CTAPbIi, please start another thread and give a more detailed description of your problem(s), hardware and software versions. ____________ Crunching on Linux: Fedora 11 x86_64 / nVidia 185.18.36 driver / CUDA 2.2 | |
| ID: 12777 | Rating: 0 | rate:
/ | |
|
BarryAZ Send message Joined: 16 Apr 09 Posts: 163 Credit: 920,275,294 RAC: 0 Level Scientific publications | |
|
Understood -- I do wish that addressing some design changes and issues was not a case of being 'too political' -- but it is what it apparently is. To approach correcting that requires a *policy* change by folks running the developers. | |
| ID: 12778 | Rating: 0 | rate:
/ | |
The work fetch sequence I posted over on the 6.10.7 thread is so wrong that the denial of it being a problem that I've seen from the client development side is really troublesome. It is an example sort of thing that has Paul so bothered. I so don't want to get involved in this conversation, but here goes.... boinc (or at least the core) is open source! That means you get to see the code, warts and all. You can change it. If the scheduling code is so broken, I'd expect someone to at least have a stab at releasing a forked version of boinc with a fixed scheduler. The 'forked' codebase (or at least the client core) gains traction with more and more people using this version if it is so much better. The 'original' boinc developers will see the new scheduler out there in the wild and working better than their implementation for the majority of users. There is the old saying that talk is cheap. It really does help to show that their is a better way. Not just expend hot air talking about it, but actually do it. People tend to listen more to your ideas when you show a working implementation of them. Now let me state that I do not agree with forks. They tend to be more destructive than constructive, but if you're not being listened to and you can code. (Just about any change made to the scheduler behaviour will have a consequence in another set of circumstances. The question is whether it is the majority or minority that is having most problems in the first place. You can't please everyone.) b) It occurs to me that some of the the people complaining (or at least those shouting loudest in public) at the moment aren't necessarily the right people. Perhaps if a few of the individual project owners put forth a case for scheduler changes, they'd have more chance of being listened to. ie. the people who are really affected by some of the issues when less science is produced for their projects, lots of WU's are queued on the client, never to be completed, deadlines missed.... This are the sorts of things that yes, they affect the user. But they really affect the project that the user is contributing to! PS. Me, I don't see some of the issues that others see. But I only run a simple config with 2 projects. GPUGRID on the GPU's and WCG on the CPU's. ____________ Crunching on Linux: Fedora 11 x86_64 / nVidia 185.18.36 driver / CUDA 2.2 | |
| ID: 12785 | Rating: 0 | rate:
/ | |
|
Paul D. Buck Send message Joined: 9 Jun 08 Posts: 1050 Credit: 37,321,185 RAC: 0 Level Scientific publications | |
Paul, I hope I'm not speaking out of turn and you don't take offence, but it seems you are so disillusioned with BOINC it's going to give you a stomach ulcer. Maybe you need a break from it? Been there, done that... I used to be 141 in the world ... I am now 180 I think ... something about 2 years without running BOINC ... see the world position graph in the middle And truth never gives me offense... OOPs; 177 out of 1,801,419 ... I am still signing my checks as 1.7 million ... JackOfAll Not going to quote, too long ... but ... I agree with the observation about forks and synecdoche already exists as an alternative ... there are many more political and practical reasons that I do not think that a fork like this is likely to succeed ... but it is entirely possible that a project such as WCG could continue to grow and eventually decide that going their own way makes sense ... in many ways WCG is actually doing that ... their integration with BOINC and many of the external "tools" like BAM is tenuous at best ... and their connection seems to me to be getting more tenuous as they get more individualistic each passing year ... As to the projects applying pressure, well, we have kinda been there and done that too historically ... and UCB can be very resistent to the needs and desires of projects too ... and with projects tending to be only interested in the welfare of their project and to care little or nothing for the greater good of the community ... well there is that old Unix aphorism about users being a renewing resource ... As to the last point, over time you may start to see some of the issues that some of the SaH Fanatics complain about ... it has been so long since I have been there I forget the kinds of things that bug them. But, lest we forget, a lot of the classic SaH users did not make the transition to BOINC because they did not like the changes from their simple set-ups that to that point worked so well for them. So we lost those guys because we could not figure out how to satisfy their concerns. At the other end, are the nuts like me that tend (though I am down to I think 5 projects till I get my WCG badges to where I want them) to support any project if it is issuing work. And I can assure you that BOINC does not handle multiple projects gracefully ... and it is getting less graceful each generation. More critical, the faster and "wider" your system gets the less well BOINC uses it ... | |
| ID: 12787 | Rating: 0 | rate:
/ | |
|
BarryAZ Send message Joined: 16 Apr 09 Posts: 163 Credit: 920,275,294 RAC: 0 Level Scientific publications | |
|
And I am 123 out of that same 1801049... Paul, I hope I'm not speaking out of turn and you don't take offence, but it seems you are so disillusioned with BOINC it's going to give you a stomach ulcer. Maybe you need a break from it? Been there, done that... I used to be 141 in the world ... I am now 180 I think ... something about 2 years without running BOINC ... see the world position graph in the middle And truth never gives me offense... OOPs; 177 out of 1,801,419 ... I am still signing my checks as 1.7 million ... ...[/quote] | |
| ID: 12801 | Rating: 0 | rate:
/ | |
|
BarryAZ Send message Joined: 16 Apr 09 Posts: 163 Credit: 920,275,294 RAC: 0 Level Scientific publications | |
|
At a certain point, I may submit to the client foolishness and support no more than four clients per workstation (two CPU and two GPU) -- providing support to the current live state of projects I have by varying the mix from workstation to workstation. At the other end, are the nuts like me that tend (though I am down to I think 5 projects till I get my WCG badges to where I want them) to support any project if it is issuing work. And I can assure you that BOINC does not handle multiple projects gracefully ... and it is getting less graceful each generation. More critical, the faster and "wider" your system gets the less well BOINC uses it ... | |
| ID: 12802 | Rating: 0 | rate:
/ | |
|
Paul D. Buck Send message Joined: 9 Jun 08 Posts: 1050 Credit: 37,321,185 RAC: 0 Level Scientific publications | |
And I am 123 out of that same 1801049... Cool ... I lost my place and though I am trending down it is not likely I will ever be back in that range again. Too many people have access to work machines to build a farm on and all I have is what I can run in my home with my spouse's negative feedback on my hobby... Though maybe the next round of upgrades may put me back in contention ... rumors of 12 CPU chips (6 cores with HT = 12 virtual CPUs) and maybe 3 PCI-e slots filled with GPUs ... The thing is, unlike the developers, in addition to have a lot of workstations doing BOINC representing a fairly broad range of hardware and running from Win2K to Windows 7, I also am of have participated in a broad range of projects: My problem is that I can see merit in almost all of the projects extant. And if they are responsive to the users (The Lattice Project or Sztaki with their contentment with high failure rates on tasks don't qualify in my book, PG and YoYo are also not high on my list for similar reasons though they at least seem to try to be responsive) ... So, I tend to run the "bigger" and more rewarding projects at a resource share of 100 and scale that down to 5 for those that I have little interest in ... or leave them out of the mix entirely because ... like SaH, if for no other reason than they really don't need my help ... far too many idiots over there ... I have often wondered what is going to happen to some of those people when SaH closes its doors ... | |
| ID: 12806 | Rating: 0 | rate:
/ | |
|
BarryAZ Send message Joined: 16 Apr 09 Posts: 163 Credit: 920,275,294 RAC: 0 Level Scientific publications | |
|
I still run SETI on a number of systems -- generate something just over 1% of my current credit activity. For me, with the extra GPU and broader GPU support, my big three over the past month have been Milkyway (CPU only due to their double precision requirement for GPU), GPUGrid, and Collatz (the new guy in my group). Behind them have been the 'steadies' (CPU only) of Spinhenge, POEM and Climate. Then at the low end, it's Einstein, Malaria, and SETI and WorldGrid.
| |
| ID: 12809 | Rating: 0 | rate:
/ | |
|
Paul D. Buck Send message Joined: 9 Jun 08 Posts: 1050 Credit: 37,321,185 RAC: 0 Level Scientific publications | |
I still run SETI on a number of systems -- generate something just over 1% of my current credit activity. For me, with the extra GPU and broader GPU support, my big three over the past month have been Milkyway (CPU only due to their double precision requirement for GPU), GPUGrid, and Collatz (the new guy in my group). Behind them have been the 'steadies' (CPU only) of Spinhenge, POEM and Climate. Then at the low end, it's Einstein, Malaria, and SETI and WorldGrid. My normal big are: ABC CPDN Einstein, SIMAP WCG Now that MW is more universal on GPUs (color me happy) I am moving to the following on GPU Collatz GPU Grid, MW I had tried to add a faster Nvida card and it died in less than a week so ... I just ordered another ATI card and I will take the system with the GTX280 and 9800GT and put two ATI cards in it and move the 280 to the system that has just the one ATI card in it (only one slot) ... At the moment I am stating to lean towards, if I can figure out the settings, running all three GPU projects on the Nvidia side and the two on the ATI side and see where the chips fall ... I keep hoping that Einstein will get the lead out and get a good GPU app version out with low CPU load ... if and when that happens I am not sure what I will do ... probably hope I can afford to up-engine some systems so I can add more GPU power. NOt all bad in any case, the more projects that have GPU applications means more CPU power left for the rest ... win-win ... | |
| ID: 12812 | Rating: 0 | rate:
/ | |
|
CTAPbIi Send message Joined: 29 Aug 09 Posts: 175 Credit: 259,509,919 RAC: 0 Level Scientific publications | |
CTAPbIi, please start a new thread. What is hanging, the entire desktop, manager app, or something else. What distro, kernel, CUDA version, boinc version, GPU hardware, etc. etc. I'll jump in and help if I can. I'm done. http://www.gpugrid.net/forum_thread.php?id=1394#12815 ____________  | |
| ID: 12816 | Rating: 0 | rate:
/ | |
|
MarkJ Volunteer moderator Volunteer tester Send message Joined: 24 Dec 08 Posts: 738 Credit: 200,909,904 RAC: 0 Level Scientific publications | |
|
Well this one didn't last very long either. Its now been superceeded by 6.10.10. See seperate message thread for details. | |
| ID: 12897 | Rating: 0 | rate:
/ | |
|
Richard Haselgrove Send message Joined: 11 Jul 09 Posts: 1576 Credit: 5,876,811,851 RAC: 10,273,931 Level  Scientific publications | |
Well this one didn't last very long either. Its now been superceeded by 6.10.10. See seperate message thread for details. Largely because of pressure from this board to introduce a mechanism to prevent unwanted requests for inappropriate work! | |
| ID: 12901 | Rating: 0 | rate:
/ | |
|
MarkJ Volunteer moderator Volunteer tester Send message Joined: 24 Dec 08 Posts: 738 Credit: 200,909,904 RAC: 0 Level Scientific publications | |
Well this one didn't last very long either. Its now been superceeded by 6.10.10. See seperate message thread for details. And about time too! Dr A seems to like you at the moment, you've managed to convince him to change a couple of things now. ____________ BOINC blog | |
| ID: 12903 | Rating: 0 | rate:
/ | |
|
Richard Haselgrove Send message Joined: 11 Jul 09 Posts: 1576 Credit: 5,876,811,851 RAC: 10,273,931 Level Scientific publications | |
Well this one didn't last very long either. Its now been superceeded by 6.10.10. See seperate message thread for details. Others (Paul, Barry, Nicolas) provided the impetus and the clues about where to look, I just provided the final summary in a form of words which tickled his fancy. ;-) | |
| ID: 12904 | Rating: 0 | rate:
/ | |
|
BarryAZ Send message Joined: 16 Apr 09 Posts: 163 Credit: 920,275,294 RAC: 0 Level Scientific publications | |
|
Very interesting stuff -- I have encountered a problem though with 6.10.x over on Collatz on a new system running Win 7 RTM on an ATI 785 chipset (embedded ATI GPU 4200). 6.10.7 detected the installed Win7 specific version of Catalyst and downloaded ATI GPU work, all of which comp errored out quickly. 6.10.9 and 6.10.10 would not download any work as neither version of the client detected the installed Catalyst driver (the client did detect the 4200GPU, just denied the existence of Catalyst). I then installed an 8400GS I had hanging around -- it is processing GPU's with that combination. Well this one didn't last very long either. Its now been superceeded by 6.10.10. See seperate message thread for details. | |
| ID: 12911 | Rating: 0 | rate:
/ | |
|
Richard Haselgrove Send message Joined: 11 Jul 09 Posts: 1576 Credit: 5,876,811,851 RAC: 10,273,931 Level Scientific publications | |
Well this one didn't last very long either. Its now been superceeded by 6.10.10. See seperate message thread for details. And now there's a v6.10.11 (Windows only, until Charlies gets his hands on it). | |
| ID: 12915 | Rating: 0 | rate:
/ | |
|
Paul D. Buck Send message Joined: 9 Jun 08 Posts: 1050 Credit: 37,321,185 RAC: 0 Level Scientific publications | |
Well this one didn't last very long either. Its now been superceeded by 6.10.10. See seperate message thread for details. And it is not clear to me that the server side is properly updated as the options are still the same though the notes seem to indicate that they should be different. I just hope that I can get work when I need it from there even though I am not updating past 6.10.7 till it looks like I will not have to enter version a day mode ... | |
| ID: 12918 | Rating: 0 | rate:
/ | |
Message boards : Graphics cards (GPUs) : Development BOINC 6.10.9 released
