Long Run-Times? Optimize.
log in

Advanced search

Message boards : News : Long Run-Times? Optimize.

Previous · 1 · 2
Author Message
Profile Overtonesinger
Avatar
Send message
Joined: 16 Jul 10
Posts: 21
Credit: 142,646,892
RAC: 1,941
Message 18540 - Posted: 14 Feb 2014, 17:38:30 UTC - in response to Message 18486.

OK, going in!

Thanx! :-)
____________
Melwen - child of the Fangorn Forest

Profile Overtonesinger
Avatar
Send message
Joined: 16 Jul 10
Posts: 21
Credit: 142,646,892
RAC: 1,941
Message 18543 - Posted: 14 Feb 2014, 17:47:57 UTC - in response to Message 18486.

Dear Slicker.

There is no "items_per_iteration" nor "items_per_reduction".

S, what shall I change. Or is it already correct? Thanx. :)

-----------------
This is the config I just downloaded from this thread.: For powerful x64 Collatz on AMD GPUs:

verbose=1
threads=8
items_per_kernel=22
kernels_per_reduction=9
sleep=1

____________
Melwen - child of the Fangorn Forest

Profile Zydor
Avatar
Send message
Joined: 19 Aug 09
Posts: 364
Credit: 840,811,292
RAC: 0
Message 18548 - Posted: 14 Feb 2014, 19:20:51 UTC - in response to Message 18543.

One step at a time ...... use the one you posted, but bring down from 22 to 18. 22 is likely way too high for your cards.

Get it running at the lower values ie 18 not 22, then step up - don't do it the other way round in one giant leap, that's a whole world of hurt :)

Don't forget you need a properly configured cc_config and app_config as well.

Profile eXtreme Warhead
Avatar
Send message
Joined: 18 Nov 12
Posts: 15
Credit: 229,064,135
RAC: 100,934
Message 18616 - Posted: 17 Feb 2014, 13:46:15 UTC

does someone have a functionally optimized config for running collatz only on one 660ti so that the os isn't laggy as hell?

i took the optimizations from above on a system with a hd5850 and was able to improve the time per wu from 50min for one unit to 62min for two units. the problem on my normal pc is, that with the same settings the os is laggy as i couldn't be worse. even the cursor only jumps around :?

actually i use only the appconfig for two wu at the same time and be able to run 2wu in 68min. days ago my 660ti has needed ~35min and my 5850 about 50min, but now my 5850 is faster...

Profile Zydor
Avatar
Send message
Joined: 19 Aug 09
Posts: 364
Credit: 840,811,292
RAC: 0
Message 18620 - Posted: 17 Feb 2014, 14:39:17 UTC - in response to Message 18616.
Last modified: 17 Feb 2014, 14:39:55 UTC

does someone have a functionally optimized config for running collatz only on one 660ti so that the os isn't laggy as hell?


Have you set values into all three config files, not just cc_config and app_config ?

Examples of the third config file are below. If the third is not yet set, use items_per_kernel=18 not 22 to start with, step up in increments as each WU goes through to the max of 22

Profile eXtreme Warhead
Avatar
Send message
Joined: 18 Nov 12
Posts: 15
Credit: 229,064,135
RAC: 100,934
Message 18623 - Posted: 17 Feb 2014, 17:28:29 UTC - in response to Message 18620.
Last modified: 17 Feb 2014, 17:38:35 UTC

cc_config is empty, appconfig only set for two wu parallel. i only used the original version from the opt_config folder for the nvidia x64

tested now with items_per_kernel=18 and the mouse isn't laggy, but the video has some frequently skipped frames in it

tested a bit around and even with

threads=7
items_per_kernel=18
kernels_per_reduction=7

it has partially skipped frames. doesn't look like i can anything optimize about the original defautl values which are

threads=6
items_per_kernel=20
kernels_per_reduction=6

beecause without a nvidia config file it is perfectly smooth?

edit: checked it with
threads=6
items_per_kernel=20
kernels_per_reduction=6
and even there skipped frames. then cleared the config file and everything runs fine again. so what are the standardvalues?

Profile Zydor
Avatar
Send message
Joined: 19 Aug 09
Posts: 364
Credit: 840,811,292
RAC: 0
Message 18625 - Posted: 17 Feb 2014, 18:01:29 UTC - in response to Message 18623.
Last modified: 17 Feb 2014, 18:03:35 UTC

so what are the standardvalues?......


Go to this thread:

http://boinc.thesonntags.com/collatz/forum_thread.php?id=1009&postid=16503#16503

Start at post one from Slicker, and work your way up. Don't be put off by the post dates, the miss-understanding over the use of three config files (not just two) has long been misunderstood/not applied correctly by a large number of people. Initially don't get into detail, just scan read the posts, then when back to latest post, flick back to post one, and read in detail - you'll quickly see what I am trying to get at. When changing the values (or creating the file in the first place) make sure Notepad saves with the xml extensions - not txt file extension. The application software will ignore any of those saved as .txt files.

The investment of time in understanding the three configuration files (three - not just the two classic BOINC config files) has massive pay back.

Once you have read through that thread, please come back with questions, not a problem, however I think that thread will answer most of your questions, and solve 99% of your issues.

Above all - please make sure you use all three configuration files, don't leave any blank. It can be daunting for some not used to messing around with configuration files, but it is easy once the *lightbulb* goes off in the head :)

Rule One ..... use all three configuration files, that's your baseline, then adjust their values as time goes on and you get into what each line does. One key line in the "third" config file that most have ignored:

items_per_kernel=21

For now use:

items_per_kernel=18

That line has a maximum value of 22, don't leap to 22, get all three config files in place, and do changes to them one at a time. Once all three files are tweeked and your happy, tweek it back to 21 (22 maybe too fast for yours - try if you like, but watch failed WUs carefully, may have to go back to 21)

But ....... please, please ...... use all three configuration files, not using them is the largest cause of grief there is on Collatz, and the largest single cause of WUs running way too slowly.

Don't hesitate to ask questions as a follow on post, if I don't know I'll soon say so, and Slicker will soon step in if we get it wrong.

Joe
Send message
Joined: 11 Jan 10
Posts: 26
Credit: 2,230,136,979
RAC: 2,159,915
Message 18699 - Posted: 22 Feb 2014, 5:23:13 UTC - in response to Message 18463.

Have played with your optimized instructions..but you forgot one important variable. XP on an nvida card.I have added countless useless lines of code in god knows where

For many projects XP on a a nvida card is the best PPD I've stuck your code in every file I could locate and nothing but 4800 second WU's that were 1200 seconds before.

I understand I'm not a cool lkinix programer..just a guy looking for points.

Stopped by again because you seem to me to really be trying to make everyone happy. I don't pretend to speak for anyone but myself.

If this project doesnt realize that there are folks out there that have been doing this for 10 plus years over multiple projects, which all are still work today problem free in spite of multiple upgrades/changes over that time, then I can only conclude this is a project that has evolved beyond the scope average user.

I have no problem with the fact you have to make choices and you have seemingly chosen to exclude the average XP user using nivida cards giving maxium PPD on other projects..there are Linix only projects which I obviously don't participate in that have also made this choice.

Again speaking for myself, just label this as a project for advanced users using Win 7 and above or whatever linix version works.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 18701 - Posted: 22 Feb 2014, 14:22:01 UTC - in response to Message 18699.

I can't give you the specific filename(s) without knowing more about your setup and you have your computers hidden, haven't told us whether running XP 32 or 64-bit, which GPU you are using, or which settings you have tried, but the following three simple steps should do the trick.

Step 1. Find the location of the solo_collatz config file(s).
They should be located in a folder named "c:\document and settings\all users\boinc\projects\boinc.thesonntags.com_collatz". (That assumes I am remembering the correct location from an operating system that is 12 years old which I haven't used in a number of years so I'm a little foggy on the folder names.)

Step 2. Find the config file with the exact same name from http://boinc.thesonntags.com/collatz/download/opt_config

Step 3. Download the file from step 2 and overwrite the file from step 1. Now be patient as the settings won't be used until the NEXT WU starts.

Joe
Send message
Joined: 11 Jan 10
Posts: 26
Credit: 2,230,136,979
RAC: 2,159,915
Message 18706 - Posted: 23 Feb 2014, 4:47:57 UTC - in response to Message 18701.

That worked thanks.

Profile Zydor
Avatar
Send message
Joined: 19 Aug 09
Posts: 364
Credit: 840,811,292
RAC: 0
Message 18779 - Posted: 6 Mar 2014, 17:06:02 UTC

Been getting a drastic mix of file lengths coming through - all the various sizes - and only a few of the larger ones, increasing local activity.

Its ok for bigger Beasts, but may mean smaller PCs - might - be getting the larger ones, don't know.

Acutely aware lots going on in the background .... therefore a gentle request :) .... any chance of returning to a consistent large sized WU for the bigger machines?

Profile skildude
Send message
Joined: 12 Sep 09
Posts: 13
Credit: 206,024,670
RAC: 8
Message 19976 - Posted: 24 Oct 2014, 5:10:32 UTC

I have an R9 290X GPU

the Optimized settings have cut my run times almost in half.

I'm currently using these optimized settings

verbose=1
threads=8
items_per_kernel=22
kernels_per_reduction=9
sleep=1
build_options=-Werror

I notice that the GPU down clocks and the CPU usage is high.

I have 2 cores idle for processing but see that the GPU is using a great deal of CPU time. In fact, the CPU time is almost constant. Is there a way to make sure that BOINC is getting the work to the idle CPU's and isn't competing for time on a CPU.

What could be causing the excessive CPU use even though the WU's seem to be running much faster than before. ~6 hours for the large WU's ~21 minutes for solos and the minis run about 90 seconds

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 20010 - Posted: 31 Oct 2014, 4:39:52 UTC
Last modified: 31 Oct 2014, 4:42:16 UTC

There's been a lot of discussion about CPU utilization but thus far, how each vendor decides to implement certain methods/functions seems to affect the utilization even when using asynchronous calls. nVidia actually documented that unless there are more OpenCL command queues than physical processors/cores that the asynchronous calls won't really be asynchronous. There's no way that I know of for the developer to force it to be asynchronous as there is with CUDA.


In the mean time, you can try running multiple WUs on a fast GPU via an app_config.xml if the GPU is downclocking. There are some fixes in the next version that will help AMD GPUs perform better (less memory bottleneck and more GPU stream processor load) but that won't be ready until after Thanksgiving or maybe even next year.

Previous · 1 · 2
Post to thread

Message boards : News : Long Run-Times? Optimize.


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.