Message boards :
Number crunching :
NV GPU - AP bench test run (e.g. @ GT730)
Message board moderation
Author | Message |
---|---|
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
I added a NV GT730 VGA card (2 compute units) to my little J1900 PC. I would like to know and use the opti settings on this VGA card. Like in the last thread 'Intel® iGPU AP bench test run (e.g. @ J1900)', the default settings are: -unroll 4 -ffa_block 1024 -ffa_block_fetch 512 -hp ... right? New URLs with kwsn.info (the files for to use again): Windows AP bench 211 minimal '2LC67' WU of the Zblank shortened WUs What or how I need to adjust something, that the bench test run run on the NV GPU and not again on the Intel iGPU? It looks like that the GT730 can run 2 tasks simultaneously ... I let run the Lunatics_x41zc_win32_cuda50.exe. If I look to GPU-Z, I see: 78% GPU Load 80% Memory Controller Load So I can let run 2 tasks simultaneously? I did and saw: 99% GPU Load 100% Memory Controller Load How can I adjust the bench test run, that 2 bench test tasks are running simultaneously on the GT730? With which cmdline settings/values I should start? Thanks. |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
What or how I need to adjust something, that the bench test run run on the NV GPU and not again on the Intel iGPU? AP7_xxxxxxxxxx.exe --device 0 AP7_xxxxxxxxxx.exe --device 1 (just test which device # will fit) The above lines may be used in BenchCfg.txt or --device # may be added in APbench211.cmd to: set ref_app_args= set sci_app_args= How can I adjust the bench test run, that 2 bench test tasks are running simultaneously on the GT730? Maybe just run 2 bench tests simultaneously from 2 directories and use for both -instances_per_device 2 Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Thanks. I'll test it and will report back. BTW, it's possible to make a bench test run with AP app/task and SETI (MB) app/task simultaneously? Later in real life, if I set CUDA & OpenCL /0.5, then it can be, that 1 AP and 1 SETI task running simultaneously on the GT730. If I find fastest AP app settings (2 AP tasks simultaneously), maybe it's not the fastest settings if 1 AP and 1 SETI app/task simultaneously on GT730. BTW, this is the 'small' test, later I 'must' do such test also on the quad HD7990's machine. ;-) |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
BTW, it's possible to make a bench test run with AP app/task and SETI (MB) app/task simultaneously? The same answer: Maybe just run 2 bench tests simultaneously from 2 directories (to load CPU you may run even more tests simultaneously or use BOINC to load CPU by 'mode noBS') Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Thanks again. Which tool & test task I must use for to let run SETI (MB) bench test run (simultaneously) on NV (GT730) GPU? lunatics.kwsn.info/ Test Tools - MultiBeam - Thanks. BTW, I ask all first, before I make all bench test runs (I didn't have done this before, it's new for me (SETI (MB) runs - and 2 tasks simultaneously)). I could waste my time. ;-) My PC don't crunch AP/SETI tasks during the bench test runs ... - not good for the SETI project. ;-) |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
Which tool & test task I must use for to let run SETI (MB) bench test run (simultaneously) on NV (GT730) GPU? I think: MBbench 2.10 http://lunatics.kwsn.info/index.php?module=Downloads;sa=dlview;id=387 V7 MB test WUs 1 - WUs adapted for V7 testing (PG set) http://lunatics.kwsn.info/index.php?module=Downloads;sa=dlview;id=296 For apps - maybe compare: Lunatics_x41zc_win32_cuda42.exe Lunatics_x41zc_win32_cuda50.exe One in: ...\MBbench210\MBbench\Science_apps\Reference\ and the other in: ...\MBbench210\MBbench\Science_apps\ Â Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Example: I have one APbench211.cmd in: C:\bla\bla\bench01\APbench211.cmd and in: C:\bla\bla\bench02\APbench211.cmd If I would like to execute both simultaneously, I could make a .bat file? How should look it? Just like this in a .bat file (?) : "C:\bla\bla\bench01\APbench211.cmd" "C:\bla\bla\bench02\APbench211.cmd" ... and both bench test runs start simultaneously? Thanks. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Here: http://lunatics.kwsn.info/1-discussion-forum/loading-apu-to-the-limit-performance-considerations.0.html I put all stuff needed for such kind of testing additionally to standard KWSN bench pack. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
I overlooked the 'extended KWSN bench pack'? Because I couldn't find it. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
I overlooked the 'extended KWSN bench pack'? Because I couldn't find it. Just modify the scripts in the first post to use APbench211.cmd for now. APbench214.cmd does have some cosmetic improvements, but no functional improvements. Joe |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
First I used (because in the AP stderr the Intel iGPU is named first, before the NV GPU): AP7_xxxxxxxxxx.exe --device 1 ... in BenchCfg.txt But this didn't worked. Then I used: AP7_xxxxxxxxxx.exe --device 0 ... in BenchCfg.txt And it worked. Then I used no '--device N' ... in BenchCfg.txt And it worked also. I chose the GT730 in GPU-Z and looked to 'GPU Load'. The 2nd and 3rd try the GT730 was loaded. At the 1st try I didn't looked to the Intel iGPU, GT730 overview was also shown, but no matter the bench test run was stopped/finished immediately (it didn't worked). So I guess it tried on Intel iGPU. I tested in both BenchCfg.txt files: -cpu_lock -instances_per_device 2 In Task-Manager 1 app at Core#0 and 1 app at Core#1. OK, I made the following bench test run without '--device N'. I made two folders, in each APbench/APbench211.cmd. I tried a few .bat files, but I found not the correct entries (so that both bench test runs start to the same time). So I started both with the mouse. astropulse_7.10_windows_intelx86__opencl_nvidia_100.exe As I tested 1st time this app live, following was used at default: -unroll 2 -ffa_block 512 -ffa_block_fetch 256 So I tested following (both apps simultaneously): 1st app: -unroll 2 -ffa_block 512 -ffa_block_fetch 256 -use_sleep -cpu_lock -instances_per_device 2 -hp : Elapsed 578.466 secs CPU 15.547 secs -unroll 3 -ffa_block 512 -ffa_block_fetch 256 -use_sleep -cpu_lock -instances_per_device 2 -hp : Elapsed 480.758 secs CPU 16.703 secs -unroll 4 -ffa_block 512 -ffa_block_fetch 256 -use_sleep -cpu_lock -instances_per_device 2 -hp : Elapsed 471.458 secs CPU 14.141 secs -unroll 5 -ffa_block 512 -ffa_block_fetch 256 -use_sleep -cpu_lock -instances_per_device 2 -hp : Elapsed 479.567 secs CPU 5.406 secs -unroll 6 -ffa_block 512 -ffa_block_fetch 256 -use_sleep -cpu_lock -instances_per_device 2 -hp : Elapsed 479.016 secs CPU 5.188 secs 2nd app: -unroll 2 -ffa_block 512 -ffa_block_fetch 256 -use_sleep -cpu_lock -instances_per_device 2 -hp : Elapsed 566.403 secs CPU 8.078 secs -unroll 3 -ffa_block 512 -ffa_block_fetch 256 -use_sleep -cpu_lock -instances_per_device 2 -hp : Elapsed 480.212 secs CPU 3.719 secs -unroll 4 -ffa_block 512 -ffa_block_fetch 256 -use_sleep -cpu_lock -instances_per_device 2 -hp : Elapsed 483.566 secs CPU 9.031 secs -unroll 5 -ffa_block 512 -ffa_block_fetch 256 -use_sleep -cpu_lock -instances_per_device 2 -hp : Elapsed 480.005 secs CPU 4.531 secs -unroll 6 -ffa_block 512 -ffa_block_fetch 256 -use_sleep -cpu_lock -instances_per_device 2 -hp : Elapsed 478.938 secs CPU 4.281 secs I was curious and made a quick test if it work. What should I test now? Each settings a 2nd and 3rd run? I should test higher values for -unroll? What I should do now? Thanks. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
... You might want to consider my http://setiathome.berkeley.edu/forum_thread.php?id=75185&postid=1542393#1542393. That test was on a GT 630 rev 2 which is essentially the same as your GT 730. I haven't done multiple instance testing, etc., but these Kepler GPUs which can do 384 simultaneous calculations are clearly much stronger than the J1900 iGPU. Some other testing on my GT 630 seemed to indicate much larger -ffa_block settings also offered improvement, something in the 6400 to 10240 range maybe. Again, that was single instance testing... You now have a fairly complex processing scenario with 4 CPUs, an iGPU, and two half GT 730's. The interactions of those 7 processors will be tough to figure out. I'm not going to suggest specific tests this time, but the sequence I took you through for the iGPU still makes sense to me. Joe |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
If I would like to execute both simultaneously, Not exactly, use "Start CMD /C ": (but in fact a few seconds difference between starts don't matter so manual start is OK) Put the file in: C:\bla\bla\ @Echo OFF CD bench01 Start CMD /C "APbench211.cmd" CD .. CD bench02 Start CMD /C "APbench211.cmd" CD .. Â Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
I chose the GT730 in GPU-Z and looked to 'GPU Load'. The 2nd and 3rd try the GT730 was loaded. Use SIV instead - it shows all GPUs in the main window, no need to choose http://rh-software.com/ Direct: http://rh-software.com/downloads/siv.zip If 'Version 5.00' do not work use older: http://rh-software.com/index_arc.html 'Version 5.00' may not work because of: "Migrated from the SHA-1 to SHA-2 code signing. This has no real impact on most systems, but to use SIV V5 on W7 you need KB3033929 installed and on Vista x64 to disable Driver Signature Verification." Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Thanks. Maybe someone have a hint about what would be better (to reach max whole performance (RAC) of PC) on a J1900 CPU with GT730 GPU (?): [assumed on CPU also AP tasks] * AP GPU app, cmdline without -use_sleep 1 CPU-Core reserved for AP GPU calculation (<*_ncpus>1</*_ncpus> in app_info.xml file) just 1 AP GPU app on GT730 vs. * AP GPU app, cmdline with -use_sleep 2 AP GPU apps simultaneously on GT730 (It would be nice to know this, before I make the bench test runs ;-) Thanks. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Not exactly, use "Start CMD /C ": Only for to be sure ... You mean this in a .bat file (I added 'exit'(?))? (it looks like now it's named .cmd file, since Win7? - after searching the web) - - - - - - - - - - @Echo OFF CD bench01 Start CMD /C "APbench211.cmd" CD .. CD bench02 Start CMD /C "APbench211.cmd" CD .. exit - - - - - - - - - - Thanks. |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
Thanks. My 2 cents worth, on a 2.2 GHz CPU I don't think it is worth shutting down 1 core to feed the GPU (unless it's a 980). The core will likely do more work than you gain on GPU work. But definitely use -use_sleep |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Not exactly, use "Start CMD /C ": For my test to run all processors at the same time I made bench_CPU_4_iGPU.bat like this: pushd %~dp0 cd APbench_iGPU\ start APbench214.cmd cd.. cd APbench_CPU.0\ start APbench214.cmd cd.. cd APbench_CPU.1\ start APbench214.cmd cd.. cd APbench_CPU.2\ start APbench214.cmd cd.. cd APbench_CPU.3\ start APbench214.cmd cd.. The .bat or .cmd extension is acceptable in Win 7 for things like this. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
KWSN bench re-enables BOINC on completion. Hence separate handling of this required for multi-instance test. Again, look provided scripts, they handle this. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
KWSN bench re-enables BOINC on completion. Hence separate handling of this required for multi-instance test. Right. I forget that since I use non-standard installation method for BOINC I can just manually stop BOINC & bench scripts treats my machine like there is no BOINC installed. Since registry entry it checks for does not exist. I copy program directory to machine instead of using installer. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
©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.