Message boards :
Number crunching :
OpenCL MB v8.12 issues thread attempt 2
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 . . . 6 · Next
Author | Message |
---|---|
Phobyx Send message Joined: 15 Jan 16 Posts: 12 Credit: 36,234,378 RAC: 25 |
erm... how can I unhide ONE of many hosts? (security policy does not allow revealing the datacenter hosts). Masybe this helps: GenuineIntel Intel(R) Core(TM) i5-3450 CPU @ 3.10GHz [Family 6 Model 58 Stepping 9] 4 Cores Microsoft Windows 7 Professional x64 Edition, Service Pack 1, (06.01.7601.00) NVIDIA GeForce GTX 660 (2048MB) driver: 365.19 OpenCL: 1.2 BOINC 7.6.22 Sample WUs: http://setiathome.berkeley.edu/result.php?resultid=4937409828 http://setiathome.berkeley.edu/result.php?resultid=4938380792 on a side note, for some reasons -use_sleep worked for a few WUs, then stopped working (without me changing any settings). I had to disable GPU for now, got real work to do, sorry. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
yes, that's enough, thanks. Well, your GPU has 5 CU, little more than I considered as entry-level one but apparently not enough. To help improve default setting for next release please try next tuning lines sequentially: -sbs 256 -period_iterations_num 300 -use_sleep After each attempt please allow few tasks (preferably, with at least 1 VLAR/GUPPI task in them) complete - estimate lags. And then please report what config reduced lags in more degree. Next would be to try some combos of those settings. |
Phobyx Send message Joined: 15 Jan 16 Posts: 12 Credit: 36,234,378 RAC: 25 |
At a first glance: None of the settings improve much. CPU usage noticeably drops with use_sleep as expected (which really should be default I think), but lags are still hell, especially when running two in parallel. I recon this is a more basic problem. GPU, Frame Buffer and GPU memory usage is ordinary in all cases btw(~60-90%, 20-60% and ~33%) |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
At a first glance: I see only aborted tasks. What tasks with each choice, could you provide links to them? |
Phobyx Send message Joined: 15 Jan 16 Posts: 12 Credit: 36,234,378 RAC: 25 |
Please be patient, I need to do work here and did not manage to get through the process properly. Lags are so massive I cannot run it in the background while working. Will come back to it when I have some actual useful results |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Please be patient, I need to do work here and did not manage to get through the process properly. Lags are so massive I cannot run it in the background while working. Will come back to it when I have some actual useful results Worth to set BOINC to "don't use GPU while active" mode instead of task aborting. |
Sleepy Send message Joined: 21 May 99 Posts: 219 Credit: 98,947,784 RAC: 28,360 |
I also have a GTX 660. I found some peace with these settings: -sbs 128 -spike_fft_thresh 2048 -tune 1 32 1 2 -period_iterations_num 60 -cpu_lock -total_GPU_instances_num 2 I am not saying that these are optimal parameters. I am just saying that after some tests I found a reasonably good balance with these settings and had not much stamina to investigate further. The same parameters (and any other I have tried so far) on a 650 Ti in a PC with a better CPU 930, I7 Vs a Q8200, sometimes (10% ? But never happened on Guppi so far) produces infinite driver reset. Each time the WU restarts and reaches again the same point. Hence, I have to abort these tasks. Only solution I found so far. This does not happen on the 660 (I directly get BSODs on this PC... I still do not why). For Guppi VLAR tasks, though, lag and CPU load is slightly excessive with this configuration. I try to get on with it, but situation is far from optimal, like for everyone with VLARs. But this is also mixed with svchost called by Windows Update very hungry on CPU and RAM. Refer to the thread about WIN10 for this! ;-) Very stormy period this is! ;-) :-) Crunch on! Sleepy |
Phobyx Send message Joined: 15 Jan 16 Posts: 12 Credit: 36,234,378 RAC: 25 |
Please be patient, I need to do work here and did not manage to get through the process properly. Lags are so massive I cannot run it in the background while working. Will come back to it when I have some actual useful results this was related to something else, trying to get cuda apps to be productive while working. Unnecessary to say it was in vain :P BTW, doing the sbs 256 tests now. Lags are horrible, no difference. YOu need the results anyways? This one just finished, but had mixed settings: http://setiathome.berkeley.edu/result.php?resultid=4939332077 |
Phobyx Send message Joined: 15 Jan 16 Posts: 12 Credit: 36,234,378 RAC: 25 |
I also have a GTX 660. Gave those a try for a few minutes. No difference. Lags are unbearable, no matter what setting I try :-( (and yes, the params do go to the app, I double checked) Erm... do you also use all 4 card heads like I do? I can't see the bus interface or frame buffer having noticeable peaks, but... as it feels to me that the app settings have quite no impact something more rare and/or strange could be the issue. |
Sleepy Send message Joined: 21 May 99 Posts: 219 Credit: 98,947,784 RAC: 28,360 |
Erm... do you also use all 4 card heads like I do? No, that is the only card (of any kind) in that (oldish) PC. Sleepy |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
do you also use all 4 card heads like I do? You mean 4 monitors connected to single GPU device? Another possible difference: sleepy's driver: 356.43 your: 365.19 |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
All OpenCL builds at stock version 8.12 (rev 3430) now. This thread can be used for all device types until very specific vendor-cases will be discovered. |
Phobyx Send message Joined: 15 Jan 16 Posts: 12 Credit: 36,234,378 RAC: 25 |
do you also use all 4 card heads like I do? Correct, 4 Monitors (I'm a Programmer :D). I found out that lags are widely independet of the settings I pass to the apps. It's hard for me to track which WU had which settings, but as the settings apparently make no difference for me, do you actually need the WU results? I also noticed: Lags are apparently related to changes on the desktop only(for example moving windows, typing text to web forms, redrawing icons, scrolling and the like) but probably not other process (at least I can't see it when inspecting several processes. That's just an indicator so far, no proof). They're massive in count (frequency 1 to many per second), and regularly some have long duration (regularly up to 1 second, sometimes 3 seconds especially when moving a window crossing screen boundaries) So it looks to me like it's a locking issue somewhere (Bus, Memory or the like). I deactivated direct draw (=hardware acceleration) to see if that avoids lockups, but that's not it. Having no insight how the apps work exactly I am bound to speculations. if it helps, I do have some debugging options on this machine (C++/C# VS2008,2013,2015,WDK). So if there's a reasonable way to dig deeper, just tell me |
Chris Adamek Send message Joined: 15 May 99 Posts: 251 Credit: 434,772,072 RAC: 236 |
Missed Gaussians starting to show up on other people's Mac Pros. I should finally be able to run some offline tests tonight/this weekend. http://setiathome.berkeley.edu/workunit.php?wuid=2164018667 non-SoG: http://setiathome.berkeley.edu/workunit.php?wuid=2163984105 Chris |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Try to disconect additional monitors - will it improve performance on last one? |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Missed Gaussians starting to show up on other people's Mac Pros. I should finally be able to run some offline tests tonight/this weekend. It belongs to OS X situation, not Windows 8.12 issue. AFAIK that app should stay on beta so far. |
Chris Adamek Send message Joined: 15 May 99 Posts: 251 Credit: 434,772,072 RAC: 236 |
Missed Gaussians starting to show up on other people's Mac Pros. I should finally be able to run some offline tests tonight/this weekend. Oops, I didn't read well enough to see this was just 8.12.:). Btw, why did the 8.10 app get pushed to main, have you emailed Eric about it yet? FWIW, it might be helpful to be in the wild for a bit to see if we can see which systems might be affected... Thanks, Chris |
Phobyx Send message Joined: 15 Jan 16 Posts: 12 Credit: 36,234,378 RAC: 25 |
no difference wether 1 or 4 monitors... |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
no difference wether 1 or 4 monitors... Then it seems only driver difference with working w/o lags system remains important. Worth to try downgrade from 36x.xx to 35x.xx driver. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Btw, why did the 8.10 app get pushed to main Perhaps he decided that quite small fraction affected. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.