Message boards :
Number crunching :
Test, 6.09 -> x41g, 190.38 -> 285.58
Message board moderation
Author | Message |
---|---|
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
This thread is only for advanced user! I made this test on my system*, the result could be different on your system. This thread is not for to recommend a CUDA application. You could read it, think about and decide what you could do. [* Intel Core2 Duo E7600 @ 3.06 GHz, DDR2 800/5-5-5-18, GIGABYTE GTX260 SOC @ 680/1500/1250 MHz (manufacturer OCed), WinXP Home 32bit] I used KWSN Knabench V1.81r and made my own bench WUs**. This're copied original WUs. So this test is realistic. Between all test runs a reboot of the system. No other acitivity/work during the tests. [** AR= 0.30x, 0.39x, 0.43x and 2.37x] [green fastest] [nVIDIA driver 190.38] setiathome_6.09_windows_intelx86__cuda23.exe (with CUDA V2.3) TaskName: PG030429731429144.wu - 797.688 secs Elapsed - 66.516 secs CPU time TaskName: PG039172342915667.wu - 605.141 secs Elapsed - 56.547 secs CPU time TaskName: PG043184314852926.wu - 524.516 secs Elapsed - 51.781 secs CPU time TaskName: PG23752934904573.wu - 131.094 secs Elapsed - 26.516 secs CPU time MB_6.08_mod_CUDA_V11_VLARKill_refined.exe (with CUDA V2.3) TaskName: PG030429731429144.wu - 820.797 secs Elapsed - 62.734 secs CPU time TaskName: PG039172342915667.wu - 618.047 secs Elapsed - 55.422 secs CPU time TaskName: PG043184314852926.wu - 535.422 secs Elapsed - 49.297 secs CPU time TaskName: PG23752934904573.wu - 129.531 secs Elapsed - 24.125 secs CPU time MB_6.08_CUDA_V12_VLARKill_FPLim2048.exe (with CUDA V2.3) TaskName: PG030429731429144.wu - 835.766 secs Elapsed - 65.313 secs CPU time TaskName: PG039172342915667.wu - 621.172 secs Elapsed - 52.734 secs CPU time TaskName: PG043184314852926.wu - 562.000 secs Elapsed - 48.781 secs CPU time TaskName: PG23752934904573.wu - 129.531 secs Elapsed - 23.891 secs CPU time [nVIDIA driver 197.45] setiathome_6.09_windows_intelx86__cuda23.exe (with CUDA V2.3) TaskName: PG030429731429144.wu - 805.953 secs Elapsed - 68.344 secs CPU time TaskName: PG039172342915667.wu - 612.609 secs Elapsed - 60.063 secs CPU time TaskName: PG043184314852926.wu - 529.594 secs Elapsed - 53.266 secs CPU time TaskName: PG23752934904573.wu - 133.344 secs Elapsed - 27.672 secs CPU time setiathome_6.09_windows_intelx86__cuda23.exe (with CUDA V3.0) TaskName: PG030429731429144.wu - 804.797 secs Elapsed - 67.094 secs CPU time TaskName: PG039172342915667.wu - 611.047 secs Elapsed - 57.422 secs CPU time TaskName: PG043184314852926.wu - 528.703 secs Elapsed - 51.953 secs CPU time TaskName: PG23752934904573.wu - 133.000 secs Elapsed - 26.813 secs CPU time MB_6.09_CUDA_V12b_FERMI.exe (with CUDA 3.0) TaskName: PG030429731429144.wu - 984.359 secs Elapsed - 65.422 secs CPU time TaskName: PG039172342915667.wu - 712.281 secs Elapsed - 51.641 secs CPU time TaskName: PG043184314852926.wu - 675.281 secs Elapsed - 46.578 secs CPU time TaskName: PG23752934904573.wu - 133.281 secs Elapsed - 22.203 secs CPU time setiathome_6.10_windows_intelx86__cuda_fermi.exe (with CUDA V3.0) TaskName: PG030429731429144.wu - 835.375 secs Elapsed - 67.719 secs CPU time TaskName: PG039172342915667.wu - 635.016 secs Elapsed - 58.719 secs CPU time TaskName: PG043184314852926.wu - 550.906 secs Elapsed - 52.625 secs CPU time TaskName: PG23752934904573.wu - 133.375 secs Elapsed - 27.234 secs CPU time I know that a few members don't like that I make tests. I don't understand why. This members don't need to post here. If members like this test and know it to appreciate, please let a message here in this thread. I know that a few members read this forum and never posted here. Don't be shy. You could start with your first message here. It would be nice to know, if it's appreciated. |
-= Vyper =- Send message Joined: 5 Sep 99 Posts: 1652 Credit: 1,065,191,981 RAC: 2,537 |
Nice test Sutaru.. That more or less concludes what i've thought also. Older drivers with older cards gets better performance.. Please try the 191.07 driver also while you test this if you have the necessary time to do that. Kind regards Vyper _________________________________________________________________________ Addicted to SETI crunching! Founder of GPU Users Group |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
Also useful to confirm that setiathome_6.10_windows_intelx86__cuda_fermi.exe runs on a GTX260 - though I presume it failed with all drivers before 197.45 The project could simplify its application despatch mechanism, at the expense of losing support for those efficient older drivers. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Good work!. (Don't fear me Sutaru! I like this one too, as there is no confusing or made up numbers.) As soon as the NDA's lift I hope you'll test the new builds. Best regards, 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. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Nice test Sutaru.. Thanks. I don't need to do, because Raistmer did this already last year.. Message 953054. A few messages up, I made a dummie test with real WUs with ~ same AR and it look like 190.38 is the winner on my system. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Also useful to confirm that setiathome_6.10_windows_intelx86__cuda_fermi.exe runs on a GTX260 - though I presume it failed with all drivers before 197.45 It'll be like this, that all GPUs with nVIDIA driver 197.45+ will get the 6.10_fermi app automatically (if stock)? |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Good work!. Thanks. I fear nothing and noone. If you would explain what NDA mean, then maybe I'll make the test. What build will it be? A mod of the stock 6.09 or 6.10 ? With VLAR kill, or a CPU/GPU hybrid app without VLAR kill? |
perryjay Send message Joined: 20 Aug 02 Posts: 3377 Credit: 20,676,751 RAC: 0 |
Hi Sutaru, The NDA means the non-disclosure agreement. He can't talk about it until NVidia tells him he can. PROUD MEMBER OF Team Starfire World BOINC |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
300+ thread views and only 3 replies.. Hmm.. not much appreciated.. but I guess much used.. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Hi Sutaru, Ahh.. thanks.. So my questions will be unanswered until this time. ;-) |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
...If you would explain what NDA mean, then maybe I'll make the test. Gladly. NDA stands for 'Non-disclosure agreement', which the Cuda 3.1 beta libraries are subject to. That just means 3.1 based builds can't be released yet, or chat about Cuda 3.1 in specifics etc... What build will it be?6.09 code base originally, though fermi & tesla compatible in the same build & heavily modified to reduce CPU time. BTW stock fermi 6.10 is binary identical to the 6.09 fermi build, but they decided they needed some new number when juggling plan classes (for some reason, so just a number.). With VLAR kill, or a CPU/GPU hybrid app without VLAR kill? Probably neither for initial testing, so resheduling still be recommended for the time being. Fermi's extra hardware has already helped VLAR 'a bit', and more progress for older cards is looking possible. V13 hybrid project has been temporarily shelved due to the GPU/cuda codebase being not quite mature enough yet for integration with AKv8b ... and fermi's new capabilities mean that code will probably change a lot over the next few months. 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. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
...they decided they needed some new number when juggling plan classes (for some reason, so just a number.). They made a mistake when installing the Fermi application at Beta (got the DLL names wrong). BOINC makes it very hard to change an installed program, but easy to install an updated version - like good financial record-keeping programs. So Ferni ended up being installed as v6.10, and they changed the executable name to match - good practice, makes it much easier to understand what's going on. So they ended up with
It'll be like this, that all GPUs with nVIDIA driver 197.45+ will get the 6.10_fermi app automatically (if stock)? I doubt it. As far as the project is concerned, all these semi-compatible cards are just a bloody nuisance. They don't actually need all the extra computing power: as we are seeing, it just outstrips the work supply chain and breaks the servers. Having started down the GPU road, they feel obligated to provide a token service for all classes, but once they have a complete list (as they now have - as above - at Beta), I don't imagine they'll change anything. I merely stated that it would be possible. |
MadMaC Send message Joined: 4 Apr 01 Posts: 201 Credit: 47,158,217 RAC: 0 |
Thank you for this, very useful, be intersting to see if the 257 drivers are any better. Appreciated by me :-) |
Skywalker66 @ Berlin Send message Joined: 31 Jan 01 Posts: 78 Credit: 27,692,349 RAC: 0 |
Vielen Dank Sutaru Das hat schon lange gefehlt. Klasse Arbeit !!!!! Habe diesen Test ansatzweise auch schon selber probiert, aber 190.38 bzw. 191,07 wird von meinen Karten nur ein einziges Mal akzeptiert. Mit deiner Arbeit hast du allen geholfen, die es wirklich interessiert Gruß mache weiter so und bleibe wie du bist |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
300+ thread views and only 3 replies.. How long did it take to do these tests? I think changing the video drivers back and forth is not easy thing. Do you use some special tool, e.g.: Driver Sweeper http://www.guru3d.com/category/driversweeper/ to completely uninstall the video drivers before installing the next? (many reboots, eh) Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
I know that a few members don't like that I make tests. I don't understand why. Very clear test, speaks for itself, good job Sutaru. Please do test and don't let anyone scare you . |
IFRS Send message Joined: 21 May 99 Posts: 1736 Credit: 259,180,282 RAC: 0 |
Good work Sutaru, ty for this very usefull input. Hard to have the time to do it myself, so tyvm. |
Biffa Send message Joined: 27 Oct 99 Posts: 41 Credit: 22,750,323 RAC: 0 |
Or a clean install, do all tests with oldest set of drivers, update drivers run next set of tests, etc etc. With Nvidia and Ati drivers its not necessary to use anything like driver cleaner, in fact if you are using Vista or Windows 7 using a driver cleaner program will cause more problems than using the correct upgrade path. (install over top) |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
And what about changing the video drivers version backwards? Sutaru is now using 190.38 (after using for the test 197.45): http://setiathome.berkeley.edu/hosts_user.php?userid=8619447 Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
Terror Australis Send message Joined: 14 Feb 04 Posts: 1817 Credit: 262,693,308 RAC: 44 |
When going backwards, the NVidia install program will uninstall the newer version first, then after a reboot, continues with the installation of the older driver. |
©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.