Message boards :
Number crunching :
SETI@home v8.12 Windows GPU applications support thread
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 17 · Next
Author | Message |
---|---|
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
here is the link on updated builds (r3480): https://cloud.mail.ru/public/4g4V/Gt278t2ku SETI apps news We're not gonna fight them. We're gonna transcend them. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Yes, I see you decreased the CPU usage.. except near the end, same effect around 70% complete then there is a huge upswing in CPU usage Non_guppi time to complete have gone up slightly Is this due to a sleep command? If so, can a -no_use_sleep be placed in the commandline to counter it? edit.. Also noticed this change LowPerformanceGPU path: no Is there a way to access HighPerformanceGPU path? Edit 2... GUPPI work units time to complete also prolonged about 90-120 sec per work unit I'll test it more tomorrow. I"m assuming you asked 8.13 to release on Beta? No kernal was sent along with work units, but I think we figured out a work around. Going to run those for a while |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Yes, I see you decreased the CPU usage.. except near the end, same effect around 70% complete then there is a huge upswing in CPU usage Versus what app/config? Can you post few links to results?
sleep active by default only in low-performance path.
ReadMe.
Same. Versus what and links to result.
Yep, Eric is informed already so await 8.14 soon. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
I'll answer these when I can. The machine I was testing with is on beta for the 8.14, I'll have to return it to main so I can restart it with tasks and get you the information. Thou, I was commenting on your r3480 that was downloaded from your mail.ru address. Of course that was before the release of 8.13 then 8.14, so I'm guessing those are the same as the downloaded app? I did find the high performance command, it decrease time to complete by 15 sec. |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
I'll answer these when I can. . . From what original runtime? |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
That was from the Beta site on a single work units (ie only 1 work unit on each GPU rather than running 2 or 3 at a time on it. Ran them overnight). 4 min 30 sec to 4 min 15 secs. Edit.. I believe the high performance pathway commanline is not yet available on 8.12 although a -low_perf is listed in the stderrs. May have to wait until the next version comes out. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Exactly the same. So you can run either, just provide -v 6 switch in command line and feed me with data from your GPU :) SETI apps news We're not gonna fight them. We're gonna transcend them. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
I suppose you compare with switch and w/o switch. That's good info. It means I correctly assume what would be better for high-end cards. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
[quote] It awailable on 8.14+ SETI apps news We're not gonna fight them. We're gonna transcend them. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
I'm going to change one of my current computers over to your r3480 version and install your -v 6 switch along with the -high_perf and see how it does here on main. Edit... Spoke too soon.. AP run is happening. After the APs are done, I'll convert the system. I need to run out the cache. The last time I tried this, it "ghosted" 400 work units since I didn't notice the change in nomenclature in the exe file |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Look also if app's working set increase as for iGPU flavour. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Raistmer, I've installed your -v 6 and -high_pref in my commandline The stderr report is very long which I guess is what you are after? I'll post link to some Non-guppi, but don't ask me to give good examples of ar http://setiathome.berkeley.edu/result.php?resultid=5017830894 http://setiathome.berkeley.edu/result.php?resultid=5017830952 http://setiathome.berkeley.edu/result.php?resultid=5017830276 http://setiathome.berkeley.edu/result.php?resultid=5017830845 There's a lot more but I'm only posting 4 as I don't know what you are looking for. Edit.. Running GUPPIs now and will list in a new post |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Ok, here are some GUPPI results http://setiathome.berkeley.edu/result.php?resultid=5017843460 http://setiathome.berkeley.edu/result.php?resultid=5017843394 http://setiathome.berkeley.edu/result.php?resultid=5017842916 http://setiathome.berkeley.edu/result.php?resultid=5017836911 http://setiathome.berkeley.edu/result.php?resultid=5017830144 Ok, compared to r3430, (these in groups of 3 per card) are running 2-4 minutes longer. Which is opposite of what is happening on Beta. I've noticed that the apps on beta don't behave the same way on Main, why is that? This isn't the first time we've brought this up. When you first introduced sah, they ran much faster on Beta but then dragged out once they were released on Main. I would almost say some sort of bloating of the apps but I know that isn't true so any idea why that happens? |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Thanks, I'll look into. Regarding different beta/main speeds: provided same ARs compared, the single difference that could explain such behavior is difference in FFTW wisdom files. Both binaries and CL files should be identical between beta and main after release. One can check this with some hash (md5) generating utility. Look for FFTW wisdom files in both project directories (*.wisdom). How they differ? You can just rename wisdom file from beta to main one and look if speed will change. But don't forget that main and beta can generate different sets of tasks (different AR distribution) so comparing averages it could be AR-related issue still. EDIT: and of course even don't look into run times with any -v N switch besides -v 0 or 1 - increased verbosity definitely will extend elapsed time. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
All of these have been running with -v 6 since you asked for that. Didn't know it was going to slow down the computation, lol If you want I can remove the -v 6 and run them without it or do you want them run with -v 1? |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
-v N enables additional printing in stderr. in Case of 6 it's info about timing. Few tasks of different ARs would be enough. Then it can be removed again (and no need to keep -v N in tuning line for just production runs). BOINC cuts stderr so maybe I'll prepare special test pack instead of such online tests. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
WU true angle range is : 0.009756 BLC work unit http://setiathome.berkeley.edu/result.php?resultid=5017930097 WU true angle range is : 0.009039 BLC work unit http://setiathome.berkeley.edu/result.php?resultid=5017863265 WU true angle range is : 0.382198 non-vlar http://setiathome.berkeley.edu/result.php?resultid=5017843415 WU true angle range is : 0.382198 non-vlar http://setiathome.berkeley.edu/result.php?resultid=5017843420 Here is the work units and wingmen WU true angle range is : 0.009319 http://setiathome.berkeley.edu/workunit.php?wuid=2198133437 WU true angle range is : 0.382198 http://setiathome.berkeley.edu/workunit.php?wuid=2198629361 WU true angle range is : 0.428523 http://setiathome.berkeley.edu/workunit.php?wuid=2200050457 WU true angle range is : 0.010348 http://setiathome.berkeley.edu/workunit.php?wuid=2199467715 If you need more, I can run more. For now going to run without any -v and see how the times do. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
As reminder: currently not only new 8.14 set of OpenCL apps in test on beta but also new data format testing continues. So I would recommend to increase share/completely switch to beta for few next weeks. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
No more for now, thanks for provided results. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
ok, since my last post, all work units without -v 6 Time to complete same as when it had -v6 Both GUPPI and non-guppi have increased time to complete. I run 3 at a time and total increase is about 4 minutes for both groups. I'm going to switch back now to r3430 once my cache runs out. |
©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.