AMD Dual Graphics. Logical choice?...

Message boards : Number crunching : AMD Dual Graphics. Logical choice?...
Message board moderation

To post messages, you must log in.

Previous · 1 · 2

AuthorMessage
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1762274 - Posted: 4 Feb 2016, 19:44:06 UTC - in response to Message 1762253.  

но как будто стало работать медленее

Ну не все задания одинаковые - обрати внимание на "angle range"
http://setiathome.berkeley.edu/results.php?hostid=7488485&offset=0&show_names=0&state=2&appid=

WU true angle range is : 2.188299 - Run time 58 min 38 sec
http://setiathome.berkeley.edu/result.php?resultid=4703167560

WU true angle range is : 0.374034 - Run time 1 hours 30 min 58 sec
http://setiathome.berkeley.edu/result.php?resultid=4703455742
 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1762274 · Report as offensive
Profile Jesiros
Avatar

Send message
Joined: 3 Aug 08
Posts: 68
Credit: 997,827
RAC: 0
Russia
Message 1762281 - Posted: 4 Feb 2016, 21:35:18 UTC
Last modified: 4 Feb 2016, 21:36:10 UTC

Ну то дураку понятно что они не одинаковые. Я про то что до сбоя, до 18% дошло относительно быстро, а вот после прописания команд стало несколько медленнее. Может я и придираюсь, тестировать надо, а сие занятие неблагодарное и малополезное )).
ID: 1762281 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1762416 - Posted: 5 Feb 2016, 9:04:40 UTC - in response to Message 1762281.  

после прописания команд стало несколько медленнее

Если толъко "несколько" (<5%) - это известно:

From ReadMe_MultiBeam_OpenCL_ATI.txt (written by Raistmer and others)
-period_iterations_num N : Splits single PulseFind kernel call to N calls for longest
                           PulseFind calls. Can be used to reduce GUI lags or to
                           prevent driver restarts. Can affect performance.
                           Experimentation required. The default value for v6/v7/v8
                           tasks is N=20. N should be positive integer.


My note:
The default value for new apps seem to be N=50
 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1762416 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1762417 - Posted: 5 Feb 2016, 9:29:44 UTC - in response to Message 1762209.  

Я сейчас над новым билдом работаю - будет переключаться для слабых видюх в щадящий режим так сказать.

(Raistmer: "I am now working on a new build - will be switching for the weak GPUs in "gentle mode" so to speak.")

Will this only amend -period_iterations_num (different for different GPUs) or will be some more?

Even with -period_iterations_num 900 I still feel big mouse lag every few minutes (for ~10 seconds) if the task is Not VHAR (VHAR is OK even on default 50)
(so it's kernel that runs only on Non-VHAR tasks)

AMD Radeon HD 6570 - Catalyst 11.12 - Windows XP
 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1762417 · Report as offensive
Profile Jesiros
Avatar

Send message
Joined: 3 Aug 08
Posts: 68
Credit: 997,827
RAC: 0
Russia
Message 1762418 - Posted: 5 Feb 2016, 9:31:00 UTC
Last modified: 5 Feb 2016, 10:23:21 UTC

Как то так и есть.
upd: мне вот эти все английские ридми и описания почти ничего не говорят, я их тупо не понимаю.
ID: 1762418 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1762425 - Posted: 5 Feb 2016, 10:36:55 UTC - in response to Message 1762417.  

Я сейчас над новым билдом работаю - будет переключаться для слабых видюх в щадящий режим так сказать.

(Raistmer: "I am now working on a new build - will be switching for the weak GPUs in "gentle mode" so to speak.")

Will this only amend -period_iterations_num (different for different GPUs) or will be some more?

Even with -period_iterations_num 900 I still feel big mouse lag every few minutes (for ~10 seconds) if the task is Not VHAR (VHAR is OK even on default 50)
(so it's kernel that runs only on Non-VHAR tasks)

AMD Radeon HD 6570 - Catalyst 11.12 - Windows XP


Well, I think those lags correspond to PulseFind part of processing algorithm.
Currently it divided to 3 separate paths of processing.
For tail elements of search (low number of iterations inside loop) it done in reduced kernel and in one pass.
For midrange number of iterations full kernel uses GPU memory buffer (where -sbs N option can change smth)(3 calls of that kernel for single icfft search) and for most long serial loops (most hard part for GPU cause we have very little parallelism there but number of iterations huge) processed via so called Partial PulseFind. Number of Partial PulseFind kernel calls governed by -period_iterations_num N.

Currently I did next modifications:
shifted threshold from Partial to full PulseFind switch;
added sync points between 3 sequental PulseFind calls for low-end GPUs (it seems that even smaller kernels running in sequence w/o sync can cause driver restarts on some configs);
changed default number of iterations for low-end GPUs (and whole LowPerformanceGPU path detection introduced)

My C-60 experiences driver restarts with some tasks with N==200. But N=300 seems enough to pass. Lags are huge indeed but my biggest concern to avoid driver restart for now.
Additional profiling needed to better understand where exactly biggest kernel occurs (maybe with N>100 it's full PulseFind so further N increase just meaningless).

For now I provided "full-sync" build (see link above). It resembles iGPU build that uses full-sync too but mostly for CPU time reduction purposes.

Try that build, see how it respond on N and sbs changes and report here. It will provide additional info.
ID: 1762425 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1762435 - Posted: 5 Feb 2016, 11:46:45 UTC - in response to Message 1762425.  

For now I provided "full-sync" build (see link above). It resembles iGPU build that uses full-sync too but mostly for CPU time reduction purposes.

Try that build, see how it respond on N and sbs changes and report here. It will provide additional info.

So you ask me to try MB8_win_x86_SSE3_OpenCL_ATi_APU_r3366_sync on:
AMD Radeon HD 6570 - Catalyst 11.12 - Windows XP

Under BOINC?
In "production server" (here, at SETI@home) or at Beta?

Or in offline test?
Which WUs to use for SETI@home v8 offline test?

(I now use MB8_win_x86_SSE2_OpenCL_ATi_HD5_r3330 at default)
 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1762435 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1762491 - Posted: 5 Feb 2016, 15:43:32 UTC - in response to Message 1762435.  
Last modified: 5 Feb 2016, 15:44:44 UTC

For now I provided "full-sync" build (see link above). It resembles iGPU build that uses full-sync too but mostly for CPU time reduction purposes.

Try that build, see how it respond on N and sbs changes and report here. It will provide additional info.

So you ask me to try MB8_win_x86_SSE3_OpenCL_ATi_APU_r3366_sync on:
AMD Radeon HD 6570 - Catalyst 11.12 - Windows XP

Under BOINC?
In "production server" (here, at SETI@home) or at Beta?

Or in offline test?
Which WUs to use for SETI@home v8 offline test?

(I now use MB8_win_x86_SSE2_OpenCL_ATi_HD5_r3330 at default)


Just as your abilities allow. Better in all mentioned modes (beta or main at your choice I'm confident in app enough to not insist on only-for-beta status).
But in "user besides the screen" mode. To estimate lags.

And for offline test try PG0009 from PG set. It's VLAR, that is, hardest case for GPU PulseFind.
ID: 1762491 · Report as offensive
Profile tullio
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 8797
Credit: 2,930,782
RAC: 1
Italy
Message 1762510 - Posted: 5 Feb 2016, 16:12:38 UTC

Since all my SETI@home GPU tasks on RADEON were failing, I uninstalled the Lunatics package and changed GPU to a Geforge GTX 750 which seems to work well. Can I reinstall the Lunatics package changing the GPU setting or is it best to wait for the new package?
Tullio
ID: 1762510 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1762999 - Posted: 6 Feb 2016, 22:39:45 UTC - in response to Message 1762491.  

And for offline test try PG0009 from PG set. It's VLAR, that is, hardest case for GPU PulseFind.

Is if safe for the state of video driver to kill the offline run of app (any GPU app)?
By "kill" I mean in Windows Task Manager, Process Explorer, ... (I also have System Explorer, ProcessHacker, Process Lasso)
In Process Lasso there is "Terminate" and "Force terminate" but I don't know the difference

Is it possible outside of BOINC to "nicely" tell the app "Exit"?
 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1762999 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1763001 - Posted: 6 Feb 2016, 22:57:58 UTC - in response to Message 1762999.  

After driver restart app's process hangs so should be killed by TaskManager or similar tool.
ID: 1763001 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1763012 - Posted: 7 Feb 2016, 0:30:37 UTC - in response to Message 1763001.  

I don't mean "After driver restart"
For ~3.5 years using Radeon HD 6570 I don't remember any driver restart caused by apps.
During that time I had only 2-3 driver restarts caused by very old or misconfigured games.

I mean if the app runs normally but I want to stop the offline test before it ends (e.g because of too much lag)

What undesirable effects killing the running GPU app may have?
Probability if those bad effects may happen? - if you can estimate "by eye"

(E.g. if you stand on the ground and touch an AC 230 V wire for 1 ms there is Probability of maybe 10-15% that you May not be electrocuted because at that moment the voltage may be around +50...-50 V - I didn't calculate it, just looking by eye at sinusoid; and considered my skin/body resistance - not some kid wet by salt water

Of course I don't intend to try this "Russian Roulette" ;) - 17 volts shown as dangerous in worst scenario:
http://www.allaboutcircuits.com/textbook/direct-current/chpt-3/ohms-law-again/
)


Process Hacker have "Terminator" - do you see any method that is "nice"?
http://processhacker.sourceforge.net/


 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1763012 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1763098 - Posted: 7 Feb 2016, 8:27:12 UTC - in response to Message 1763012.  
Last modified: 7 Feb 2016, 8:34:53 UTC

Ough, you approached to this matter much fundamentally i ever did , loL :)
To break offline test i usually just hit Ctr-Shift-Esc and kill corresponding process in TaskManager. Perhaps this will correspond to very first checkbox in app you listed - NtTerminateProcess.

And usually process killing via TaskManager handles leftovers well enough - I did not observe any issues after such action. Maybe it combinates NtTerminateProcess with WM_QUIT message... i would stick with TaskManager way instead of some "selective termination" such app could provide.

P.S. Nice tool to add to collection, BTW, thanks!
ID: 1763098 · Report as offensive
Previous · 1 · 2

Message boards : Number crunching : AMD Dual Graphics. Logical choice?...


 
©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.