PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

  1. Read our rules and policies.
  2. Download, install and run the BOINC software used by PrimeGrid.
  3. When prompted, enter the URL:
    http://www.primegrid.com/
  4. Select which subprojects to run via the PrimeGrid Preferences.

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
11) Message boards : Problems and Help : cmdline to force OCL instead of OCL4 (Message 113101)
Posted 2090 days ago by NeoMetal*
Wrong WU. The one you are referring to is a genefer 22 WU. This one is genefer 21. If I can't get this one running on OCL I might just abort it and start fresh but I'll give this one another try first.
This is the complete & correct stderr to the right one:

geneferocl 3.3.2-7 (Windows/OpenCL/32-bit) Copyright 2001-2017, Yves Gallot Copyright 2009, Mark Rodenkirch, David Underbakke Copyright 2010-2012, Shoichiro Yamada, Ken Brazier Copyright 2011-2014, Michael Goetz, Ronald Schneider Copyright 2011-2017, Iain Bethune Genefer is free source code, under the MIT license. Running on platform 'AMD Accelerated Parallel Processing', device 'Tahiti', vendor 'Advanced Micro Devices, Inc.', version 'OpenCL 1.2 AMD-APP (2527.7)' and driver '2527.7'. 32 computeUnits @ 955MHz, memSize=3072MB, cacheSize=16kB, cacheLineSize=64B, localMemSize=32kB, maxWorkGroupSize=256. Supported transform implementations: ocl ocl2 ocl3 ocl4 ocl5 Command line: projects/www.primegrid.com/primegrid_genefer_3_3_2_3.18_windows_intelx86__openclatiGFN21.exe -boinc -q 194778^2097152+1 Low priority change succeeded. Checking available transform implementations... A benchmark is needed to determine best transform, testing available transform implementations... Testing OCL transform... Testing OCL2 transform... Testing OCL3 transform... Testing OCL4 transform... Testing OCL5 transform... Benchmarks completed (95.641 seconds). Using OCL4 transform Starting initialization... Initialization complete (60.915 seconds). Testing 194778^2097152+1... Estimated time for 194778^2097152+1 is 53:30:00 maxErr exceeded for 194778^2097152+1, 1.0000 > 0.4500 maxErr exceeded while using OCL4; switching to OCL. maxErr exceeded may be caused by overclocking, overheated GPUs and other transient errors. Using OCL transform Resuming 194778^2097152+1 from a checkpoint (23560318 iterations left) Estimated time remaining for 194778^2097152+1 is 20:30:00 Successful computation progress with OCL; switching back to OCL4. Using OCL4 transform Resuming 194778^2097152+1 from a checkpoint (23498583 iterations left) Estimated time remaining for 194778^2097152+1 is 24:00:00 geneferocl 3.3.2-7 (Windows/OpenCL/32-bit) Copyright 2001-2017, Yves Gallot Copyright 2009, Mark Rodenkirch, David Underbakke Copyright 2010-2012, Shoichiro Yamada, Ken Brazier Copyright 2011-2014, Michael Goetz, Ronald Schneider Copyright 2011-2017, Iain Bethune Genefer is free source code, under the MIT license. Running on platform 'AMD Accelerated Parallel Processing', device 'Tahiti', vendor 'Advanced Micro Devices, Inc.', version 'OpenCL 1.2 AMD-APP (2527.7)' and driver '2527.7'. 32 computeUnits @ 955MHz, memSize=3072MB, cacheSize=16kB, cacheLineSize=64B, localMemSize=32kB, maxWorkGroupSize=256. Supported transform implementations: ocl ocl2 ocl3 ocl4 ocl5 Command line: projects/www.primegrid.com/primegrid_genefer_3_3_2_3.18_windows_intelx86__openclatiGFN21.exe -boinc -q 194778^2097152+1 Low priority change succeeded. Using OCL4 transform Resuming 194778^2097152+1 from a checkpoint (17732637 iterations left) Estimated time remaining for 194778^2097152+1 is 15:50:00 geneferocl 3.3.2-7 (Windows/OpenCL/32-bit) Copyright 2001-2017, Yves Gallot Copyright 2009, Mark Rodenkirch, David Underbakke Copyright 2010-2012, Shoichiro Yamada, Ken Brazier Copyright 2011-2014, Michael Goetz, Ronald Schneider Copyright 2011-2017, Iain Bethune Genefer is free source code, under the MIT license. Running on platform 'AMD Accelerated Parallel Processing', device 'Tahiti', vendor 'Advanced Micro Devices, Inc.', version 'OpenCL 1.2 AMD-APP (2527.7)' and driver '2527.7'. 32 computeUnits @ 955MHz, memSize=3072MB, cacheSize=16kB, cacheLineSize=64B, localMemSize=32kB, maxWorkGroupSize=256. Supported transform implementations: ocl ocl2 ocl3 ocl4 ocl5 Command line: projects/www.primegrid.com/primegrid_genefer_3_3_2_3.18_windows_intelx86__openclatiGFN21.exe -boinc -q 194778^2097152+1 Low priority change succeeded. Using OCL4 transform Resuming 194778^2097152+1 from a checkpoint (17650385 iterations left) Estimated time remaining for 194778^2097152+1 is 16:20:00 Terminating because BOINC client requested that we should quit. geneferocl 3.3.2-7 (Windows/OpenCL/32-bit) Copyright 2001-2017, Yves Gallot Copyright 2009, Mark Rodenkirch, David Underbakke Copyright 2010-2012, Shoichiro Yamada, Ken Brazier Copyright 2011-2014, Michael Goetz, Ronald Schneider Copyright 2011-2017, Iain Bethune Genefer is free source code, under the MIT license. Running on platform 'AMD Accelerated Parallel Processing', device 'Tahiti', vendor 'Advanced Micro Devices, Inc.', version 'OpenCL 1.2 AMD-APP (2527.7)' and driver '2527.7'. 32 computeUnits @ 955MHz, memSize=3072MB, cacheSize=16kB, cacheLineSize=64B, localMemSize=32kB, maxWorkGroupSize=256. Supported transform implementations: ocl ocl2 ocl3 ocl4 ocl5 Command line: projects/www.primegrid.com/primegrid_genefer_3_3_2_3.18_windows_intelx86__openclatiGFN21.exe -boinc -q 194778^2097152+1 Low priority change succeeded. Using OCL4 transform Resuming 194778^2097152+1 from a checkpoint (17475424 iterations left) Estimated time remaining for 194778^2097152+1 is 14:40:00 Terminating because BOINC client requested that we should quit. geneferocl 3.3.2-7 (Windows/OpenCL/32-bit) Copyright 2001-2017, Yves Gallot Copyright 2009, Mark Rodenkirch, David Underbakke Copyright 2010-2012, Shoichiro Yamada, Ken Brazier Copyright 2011-2014, Michael Goetz, Ronald Schneider Copyright 2011-2017, Iain Bethune Genefer is free source code, under the MIT license. Running on platform 'AMD Accelerated Parallel Processing', device 'Tahiti', vendor 'Advanced Micro Devices, Inc.', version 'OpenCL 1.2 AMD-APP (2527.7)' and driver '2527.7'. 32 computeUnits @ 955MHz, memSize=3072MB, cacheSize=16kB, cacheLineSize=64B, localMemSize=32kB, maxWorkGroupSize=256. Supported transform implementations: ocl ocl2 ocl3 ocl4 ocl5 Command line: projects/www.primegrid.com/primegrid_genefer_3_3_2_3.18_windows_intelx86__openclatiGFN21.exe -boinc -q 194778^2097152+1 Low priority change succeeded. Using OCL4 transform Resuming 194778^2097152+1 from a checkpoint (16892736 iterations left) Estimated time remaining for 194778^2097152+1 is 15:30:00


As you see it starts in OCL4, fails then switches to OCL then switches back to OCL4 then starts and stays there from then on. TThrottle is set to 82C now. I've run numerous genefer 21s as high as 88C successfully so heat shouldn't be the problem now. I tried -x OCL and nothing. What should the <app> to <app> contain in app_config? or other ways to force OCL?

EDIT: Even tried the following and still nothing: <cmdline>primegrid_genefer_3_3_2_3.18_windows_intelx86__openclatiGFN21.exe -boinc -q 194778^2097152+1 -x OCL</cmdline>
12) Message boards : Problems and Help : cmdline to force OCL instead of OCL4 (Message 113098)
Posted 2090 days ago by NeoMetal*
I have a 280x Tahiti and it normally runs OCL for both Genefer (genefer 21) and Genefer_WR (genefer 22) but is running OCL4 after a fan died and overheated some (TThrottle to the rescue again) but now runs OCL4 even though it tries OCL fine via this message:

Using OCL transform Resuming 194778^2097152+1 from a checkpoint (23560318 iterations left) Estimated time remaining for 194778^2097152+1 is 20:30:00 Successful computation progress with OCL; switching back to OCL4. Using OCL4 transform


It usually take 15-17 hrs to run genefer 21 but it's at 35 hrs with 20 to go on OCL4. What's the cmdline to force OCL instead of OCL4 to add to app_config?
13) Message boards : Number crunching : Happy New Year Challenge (Message 113069)
Posted 2091 days ago by NeoMetal*
+1 on TThrottle.

I've used it for many years and is an excellent program. It has saved my hardware's ass many times such as when a fan (CPU + GPU) dies while crunching, or an initial fan curve is set wrong by accident, etc.

I think most Windows users (if not all) should use it. I also recommend BOINC Tasks as well by the same author. There's a Linux version of that.

Link to TThrottle's manual to learn more: https://efmer.com/tthrottle-manual/

Homepage: https://efmer.com/

14) Message boards : Sophie Germain Prime Search : First Prime (Message 113068)
Posted 2091 days ago by NeoMetal*
Completed tasks 209232
Credit 7,415,906.60
Primes found 30 (18 as initial finder)
Percentage of tests resulting with a prime 0.0143

15) Message boards : Sieving : Link to sieving apps (and other PRP apps) (Message 112256)
Posted 2117 days ago by NeoMetal*
I've search for hours and can't find link to latest GFN app (is GFNSvCUDA-0_7-win32.exe the latest?) Can you post the link for it as well as all PRP apps and keep it updated with new versions (with explanation and changes made clear for new versions).

Maybe sticky post this if kept updated (remove this line if you do it)

16) Message boards : Problems and Help : GTX 1060 6GB not downloading tasks (Message 111265)
Posted 2151 days ago by NeoMetal*
Some of Primegrid CPU sub projects don't play with GPU sub projects well and can change what's called the DCF (duration correction factor) located (in Windows) C:\ProgramData\BOINC directory in a file called clientstate.xml near the beginning of the Primegrid section within it.

Sometimes the value gets wildly off when certain tasks' estimated FLOPs (in the metadata sent when downloading a task) are too far off. Ideally it should be close to 1.000000 as possible but if 2 or more different sub projects (CPU, GPU or both) are significantly off with their FLOPs estimate. This value can be too low or high causing too many, too few or none at all of 1 or more of the sub projects you've chosen to download and run at the same time.

To correct this just change the DCF value to default 1.000000 in the clientstate.xml file. Stop and exit BOINC first, open file and scroll down to the Primegrid section, find <duration_correction_factor>x.xxxxxx</duration_correction_factor> and change to default then save. Restart BOINC and update.

If this is what's causing you're problem this will fix it for now, BUT as you continue to run tasks the DCF will gradually go off kilter again until you change the combination of Primegrid sub projects which may be better (or worse) at keeping the DCF close to default. You may have to do this often until changing the combination of sub projects you run.

This is also one of the causes of why BOINC's estimated 'Time Left' can be so far off.
17) Message boards : Number crunching : Waiting for wingmen... (Message 111238)
Posted 2151 days ago by NeoMetal*
Mike, how about posting this data once a month for a while (a few months?) during the first week of each month until trends can be recognized (meaningful/meaningless?). Might provoke interesting off the cuff comments and/or other info (such as already) ...or not.

As an aside I have noticed that running MT on physical cores only while GPU and other software are on HT cores (and letting OS run on ALL cores as usual) does help improve throughput / time on what mackerel calls 'Goldilock' tests, by raising priorities (above normal usually, sometimes high) on LLR & GPU tests (including I/O & MEM Priorities) and lowering them on certain other processes (carefully/dangerously). I experimented on the settings in Process Lasso until I was getting 10-30% faster times on certain sub projects on my older 2600k running 1 WU on 4 threads vs. 4 single threaded Wus. On longer WUs I can get nearly break even sometimes. This is compared to my 3570k which doesn't have HT cores AND 2 MB less L3 also.



18) Message boards : Number crunching : Badges II (Message 111052)
Posted 2160 days ago by NeoMetal*
1 billion

19) Message boards : Number crunching : My List And Info For cc_config.xml (Message 110633)
Posted 2186 days ago by NeoMetal*
NeoMetal* will that information stop this annoyance:

4/10/2017 6:00:59 PM | PrimeGrid | Found app_config.xml
Deleted list

No. The cc_config.xml and the app_config.xml are different settings files. Those errors messages are usually just annoyances and not effect the actual app_config.xml input thread settings (or most other settings in the file). Until someone figures out why this happens (most likely a BOINC bug) most everyone will have to live with it.



20) Message boards : Number crunching : Looks like the newest version of BOINC 7.8.2 (Message 110632)
Posted 2186 days ago by NeoMetal*
I recall one discussion here finding that HT threads could drive the GPU while all cores were busy with LLR. So I run large LLR with -t 6 in app_config.xml for fast turnaround but trick BOINC into thinking LLR needs 12 threads to prevent scheduling anything else concurrently except GPU tasks. There might be a better way of doing that with BOINC, but I won't have the patience to try to optimize it. BOINC is a rather blunt instrument. With PRPNet I achieved maximum GPU throughput because I could see the rate of output in real time while fine-tuning CPU thread usage and transfer block sizes.

Intermediate and advance users should try Process Lasso where you can highly customize how individual apps run including CPU affinity, CPU, I/O and memory priorities, and so much more. Very customize-able. Free version has a nag screen but most settings/options work. Download from: https://bitsum.com/?inproduct

One of the best settings for LLR tasks is 'avoid non physical cores, no HT' and run GPU apps on HT threads only. This works well for most CPUs with HT. No HT? Then you can run LLR on all but one core, then set your GPU apps (as well as other CPU hungry apps) to the free core. It can be temporary or permanent so you don't have to keep changing affinity or priority settings in Windows task management every time a new task starts (settings there are not permanent). This will reduce cache misses.

There are other apps that can do some of what Process Lasso can do, but not like the umpteen thousands of custom combinations PL can do. And the best part is the free version gets you 90% of the settings that the Pro version ($) has.

Experiment on settings for Primegrid apps and other running apps until you find the best efficiency (progress speed, not power use) while still running other apps (like browsers, monitoring apps, background apps, etc.) if you like or need to. but leave critical OS and driver apps at default settings. This can take a lot of time but once you find the sweet spot you will rarely have to change settings anymore (unless you add new apps). In challenges you can reactivate all cores (and close all unneeded apps) easily and change back when the challenge ends. Good luck.



Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2023 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 2.63, 2.71, 2.59
Generated 29 Sep 2023 | 13:25:37 UTC