Message boards :
Number crunching :
Lunatics Windows Installer v0.41 Release Notes
Message board moderation
Previous · 1 . . . 7 · 8 · 9 · 10 · 11 · 12 · 13 . . . 17 · Next
Author | Message |
---|---|
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
It's Josef's CPU applications which do the double-wisdom jig. Raistmer's ATI applications do all they're going to do the first time. For these reasons, and others, there has been a move in (computer science) literature, over the last couple of years, from dynamic run time optimisation ( as per fftw planning & wisdom ) toward offline (install time) optimisation. This is similar to what LLVM based compilers do on mobile devices, 'optimise your applications' before you use them (after app & system updates too). With Cuda builds I currently use old-fashioned compile-time optimisation (multiple embedded GPU binaries)with run time planning (as dictated by CUFFT library), dividing the first run convergence on 'known GPUs'. When an Unknown GPU is encountered (such as GK110 recently with x41zc), the (PTX) code embedded in the app gets compiled & optimised on the first run.... IOW, You could go so far as to (optionally) run a short bench/regression test immediately after install, but prior to first live run. This would ensure the offline optimisation enabled in LLVM based first run, or FFTW dynamic wisdom based, gets populated fairly before tests or live running. On one hand raised priorities during these learning period might be best, or on the other hand, conditions representative of real running (which I prefer). In either case the goal is optimise both the running, and the initial 'user experience'. In most cases telling the user something like "I'm doing something good now, this could take some time" would be acceptable to me, though both fftw's dynamic and Cuda's new card first run approaches violate that, and still work very well. Jason "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Rockhount Send message Joined: 29 May 00 Posts: 34 Credit: 31,935,954 RAC: 29 |
Sorry, my fault. On this machine I don't have the user rights to change the graphic driver. greetings Regards from nothern Germany Roman SETI@home classic workunits 207,059 SETI@home classic CPU time 1,251,095 hours |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Since optimizing with the installer, it seems I am reported errors on some AP tasks. Have to run off to work, but could somebody have a look at my recent AP errors and see if you can offer any clues? They have happened on at least 3 different rigs. I'll check back in tonight. Thanks. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
QSilver Send message Joined: 26 May 99 Posts: 232 Credit: 6,452,764 RAC: 0 |
This question from yesterday probably got overlooked: One quick question: Can the NV 4200M on this laptop handle 2 units at a time? Any thoughts? |
JohnDK Send message Joined: 28 May 00 Posts: 1222 Credit: 451,243,443 RAC: 1,127 |
Since optimizing with the installer, it seems I am reported errors on some AP tasks. Not that I think I could help but your link gives "No access" :) |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14649 Credit: 200,643,578 RAC: 874 |
This question from yesterday probably got overlooked: A bit of a marginal case, I'd say. Your computer listing here says: NVIDIA NVS 4200M (482MB) driver: 268.83 where the Wiki comparison table for Mobility NVS Series gives us a GF119 GPU with 'up to 1024' MiB of memory and 48 shaders. The 'GF' series GPU can certainly handle two tasks at once, but that 482 MB of memory feels a little tight. I run a 420M GPU with 1023MB of memory: that could certainly handle two at a time, but I run it with a single task, because the 48 shaders (same as yours) make it substantially slower than the venerable 9800GTs I run in desktop machines. |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13720 Credit: 208,696,464 RAC: 304 |
The 'GF' series GPU can certainly handle two tasks at once, but that 482 MB of memory feels a little tight. Running 2 at a time on my systems, and with a couple of browsers open it's using 670MB of Video RAM. Grant Darwin NT |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
Fred E.: I'm puzzled by the problem I'm having. The problem was identified and solved in dialog between Fred E. and me. This is not caused by error in the installer. (aimerge.cmd included in the installer uses the well known command 'find' which is supposed to be present and working on any Windows) The cause: It turns out that on some systems the Path environment variable is not set correctly (probably some program install/uninstall deleted it). (find.exe used by aimerge.cmd is in \WINDOWS\system32\ and this directory need to be in Path= (which is the case on fresh installed Windows)) To check: Start command prompt (Win+R and type in the 'Run' window: CMD) Type (in the command prompt window): set pa See if somewhere in the Path= there is: C:\WINDOWS\system32 (drive letter may be different, you see my is not C: but D:): If you don't have this directory in the Path= some commands will not work on your system. Fix: Add this text to the beginning of Path environment variable (copy from here, change drive letter if needed): ;C:\WINDOWS\system32;C:\WINDOWS; To get to the window for editing: Windows XP: Win+Pause Windows 7/Vista: Win+Pause and at top-left click 'Advanced system settings' ... and follow the picture bellow. In 'Edit System Variable' window press Home to go to the beginning and Ctrl+V to paste the ;C:\WINDOWS\system32;C:\WINDOWS; (I think you need to be 'Administrator' to do the above change) - Reboot the computer - See if the set command now shows ;C:\WINDOWS\system32;C:\WINDOWS; in the Path= (tested on Windows XP 32-bit, you may see some difference on other systems) P.S. On Fred E. computer the Path was only this: Path=C:\Program Files\NVIDIA Corporation\PhysX\Common We fixed it to: Path=C:\WINDOWS\system32;C:\WINDOWS;C:\WINDOWS\System32\Wbem;C:\Program Files\NVIDIA Corporation\PhysX\Common ... and the installer's aimerge.cmd correctly created app_info.xml Fred E.: "The installer ran fine, no visible problems, now running BOINC with the app_info.xml it created (with usual edits for count)." Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
Mike Send message Joined: 17 Feb 01 Posts: 34253 Credit: 79,922,639 RAC: 80 |
Since optimizing with the installer, it seems I am reported errors on some AP tasks. I`m almost certain its a driver issue. I see you are running the same driver on all GPU`s 304.79. I suggest to install a newer driver at least on one host to see if that changes anything. With each crime and every kindness we birth our future. |
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
I've had 3 validation inconclusives with my ATI cards. All 3 were against Cuda cards. 2 of the Cuda card results were overflows. In a rich man's house there is no place to spit but his face. Diogenes Of Sinope |
Juha Send message Joined: 7 Mar 04 Posts: 388 Credit: 1,857,738 RAC: 0 |
Fix: I'm sure Richard, William and the rest of the gang are pleased to hear you figured out why the install didn't work. However, if you are going to fix something fix it right. In this case, don't hardcode the paths like that. The correct form is: %SystemRoot%\system32;%SystemRoot%;%SystemRoot%\System32\Wbem; (taken from an almost vanilla installation of XP.) |
Mike Send message Joined: 17 Feb 01 Posts: 34253 Credit: 79,922,639 RAC: 80 |
I've had 3 validation inconclusives with my ATI cards. All 3 were against Cuda cards. 2 of the Cuda card results were overflows. That might be host dependent. I more worry about those just missing one signal. With each crime and every kindness we birth our future. |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
Fix: Yes, this looks like the correct way. Then it is strange that on my system (Win XP) it is hardcoded (copy is from 'Edit System Variable' window, not from CMD, and I didn't edited it manually ever): %CommonProgramFiles%\Microsoft Shared\Windows Live;D:\Program Files\AMD APP\bin\x86;D:\WINDOWS\system32;D:\WINDOWS;D:\WINDOWS\System32\Wbem;H:\Programs.XP\Samsung;H:\ASRAPI\BIN;D:\Program Files\Common Files\Teleca Shared;D:\Program Files\QuickTime\QTSystem\;D:\Program Files\ATI Technologies\ATI.ACE\Core-Static  - ALF - "Find out what you don't do well ..... then don't do it!" :)  |
Juha Send message Joined: 7 Mar 04 Posts: 388 Credit: 1,857,738 RAC: 0 |
Yes, this looks like the correct way. Without any proof whatsoever I'd put the blame on some lousy installer reading the variable in its expanded form and then writing it back like that. |
QSilver Send message Joined: 26 May 99 Posts: 232 Credit: 6,452,764 RAC: 0 |
This question from yesterday probably got overlooked: Running 2 at a time on my systems, and with a couple of browsers open it's using 670MB of Video RAM. Thanks for taking a look, Richard and Grant. I'll leave it at one task. |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Since optimizing with the installer, it seems I am reported errors on some AP tasks. I may try some driver updating tomorrow. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
William Send message Joined: 14 Feb 13 Posts: 2037 Credit: 17,689,662 RAC: 0 |
I've had 3 validation inconclusives with my ATI cards. All 3 were against Cuda cards. 2 of the Cuda card results were overflows. False overflows are usually flaky cards - too hot or just half dead. A person who won't read has no advantage over one who can't read. (Mark Twain) |
JPP Send message Joined: 31 May 99 Posts: 18 Credit: 59,436,360 RAC: 47 |
hi / just a stupid beginer's request i recently performed upgrade to latest setiathome version + w32 041 optimizer after many many attempts (remove/reinstall); i can see the time to compute a wu with my gpu (650ti) simply doubled; compared to ""before; Even when CPU's are idle; it simply takes twice than before I tried also to update to the latest nvidia driver i can see i m using cuda32/ x41zc ; so all looks correct to me and i face a pretty stable situation therefore i come with 2 questions 1 would you mind to tell me where to dig first ? 2 Is there anywhere info/bench on expected crunch time for with various gpu which coud help to figure out roughtly if on track or not i suspect i have something wrong but miss the point txs in advance cheers:jpp |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13720 Credit: 208,696,464 RAC: 304 |
after many many attempts (remove/reinstall); i can see the time to compute a wu with my gpu (650ti) simply doubled; compared to ""before; v7 processing takes much longer than with v6. Sometimes the processing time is more than doubled, others less than double. Grant Darwin NT |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14649 Credit: 200,643,578 RAC: 874 |
Yes, this looks like the correct way. That looks plausible. On an almost vanilla Windows XP with BOINC-related additions, the path is: PATH=C:\Program Files\NVIDIA Corporation\PhysX\Common;C:\WINDOWS\system32;C:\WINDOWS;C:\WINDOWS\System32\Wbem;C:\CUDA\bin So guess which installer we blame ;P OK, I've got the message - and many, many thanks for continuing to explore this while I caught my breath after Monday. OK, now we know where the problem lies, I've got two or three possibilities for a workround - I'll let them marinade for a day or two. There was an important update to the BOINC source code yesterday, which affects all BOINC applications (across all projects, not just us) - especially GPU apps. So it's just possible that we may have to update the installer with another full release, sooner rather than later. If so, that would be an ideal opportunity to deploy whatever solution we come up with. |
©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.