SETI@home v8.19 Windows GPU applications support thread

Message boards : Number crunching : SETI@home v8.19 Windows GPU applications support thread
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 9 · Next

AuthorMessage
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1824856 - Posted: 17 Oct 2016, 8:13:18 UTC - in response to Message 1824768.  

I found the problem. It WAS -cpu_lock.

As stated earlier, issue is in inproper usage of this switch for more than single task per device config.
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1824856 · 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 1824857 - Posted: 17 Oct 2016, 8:16:07 UTC - in response to Message 1824816.  


You must think of the values you list as GUIDELINES, not absolutes.

Values from BOINC xml files are for BOINC usage, not app usage.
In particular, CPU consumption field tells BOINC how to schedule particular plan class, it doesn't tell the app how much CPU to use.
So, if they are guidlines - they are guidlines for BOINC client. And should be properly used as such.
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1824857 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1824883 - Posted: 17 Oct 2016, 10:36:38 UTC - in response to Message 1824822.  


0.2 was just a number I pulled out of my *ss, just wanted the 6 GPU WUs to pull 2 CPUs out of the available number. In the past, when I tried -use_sleep, I found that (on one of my other machines (Intel E5-2670)) that CPU/GPU WU went from 90%+ to well under 20%, even as elapsed time increased roughly 20-30%.
EDIT: IIRC


. . I think you will find that with this version, (r3528), with sleep on the WU will still use most of the GPU resources. If you are running multiple WUs per GPU then it will use almost all of it. I have a rig with a very outdated Pentium D CPU and running 2WU/GPU under r3528 on 2 GTX970s. It takes about 50% of a CPU core for each running WU but it uses 95% of each GPU, and the CPU is pretty much at it's limit. With a more powerful CPU you should be able to use almost all of the GPU resources even running singles. Well that is my observation anyway.

Stephen

.
ID: 1824883 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1824979 - Posted: 17 Oct 2016, 18:46:28 UTC - in response to Message 1824711.  

Thanks Raistmer,

I thought I remember needing that.

I think there is also 1 other command needed telling the total amount of all instances. Keith, you remember what the command is??

@Zalster, the other command is -total_GPU_instances_num N which gets used along with -cpu_lock and -instances_per_device N.

I ran those commands for a while but they ended up causing more issues than resolved. I also run ProcessLasso so that probably has an effect on BOINC. Running pretty much bare-bones now except for tuning.

@Mike, I don't run -cpu_lock on my FX processors and haven't seen any MB task processing degradation. Running two tasks per GPU and 1 CPU core per WU.

WU tasking solely controlled by app_config.xml entries.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1824979 · Report as offensive
Profile KC5VDJ - Jim the Enchanter
Avatar

Send message
Joined: 17 May 99
Posts: 81
Credit: 4,083,597
RAC: 0
United States
Message 1826166 - Posted: 22 Oct 2016, 20:36:13 UTC

I recently upgraded to the RX 480 and am wondering what I change to set the GPU to two concurrent tasks?

I was looking around in app_info.xml, and i'm not sure exactly where.

I am running the Lunatics 0.44 AVX and ATI5 clients.

I have the CPU where I want it with eight simultaneous threads, but since the GPU seems to not be maximum, I want to up that to two tasks.
Delidded i7-4790K (CLU/CLU) at 4.7GHz @ 1.310Vcore 24/7, 32GB DDR3-2400, Corsair H100i v2, Gigabyte Z97X-Gaming G1 WIFI-BK, MSI Radeon RX 480 Gaming 4G, HX-650 PSU, Corsair 750D

ID: 1826166 · 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 1826169 - Posted: 22 Oct 2016, 20:41:06 UTC - in response to Message 1826166.  

This GPU family has bugged driver that can't do correct context switching. So only 1 GPU task per device recommended.
To increase GPU throughput try to add recommended tuning line instead (look ReadMe for details).
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1826169 · 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 1829287 - Posted: 9 Nov 2016, 10:24:01 UTC

New v8.20 GPU apps was deployed recently on beta - please do massive testing to speedup their release to main.
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1829287 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1829316 - Posted: 9 Nov 2016, 12:51:43 UTC - in response to Message 1829287.  

New v8.20 GPU apps was deployed recently on beta - please do massive testing to speedup their release to main.


. . Hello again,

. . I have installed Beta 6 and will now see how it runs for my own edification. I hope there is an improvement in productivity rather than a reduction.

. . However I take it you require large scale participation on the Beta-Testing project. I will give it a couple of days then go through the pain of adding the Beta-testing project on La_Bamba, how long do I need to stay there for this time?

Stephen

.
ID: 1829316 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1829335 - Posted: 9 Nov 2016, 15:42:34 UTC

. . Hi Raistmer,

. . For what it is worth on La_Bamba (now with two GTX950s) I had been running :-

-high_prec_timer -use_sleep -sbs 384 -period_iterations_num 5 -tt 500

. . but I had been experiencing a little lag since the current run of BLC2 tasks began, it was fine before that.

. . When changing to V8.20/r3557 I upped the ante a little and now run :-

-high_prec_timer -use_sleep -sbs 512 -period_iterations_num 3 -tt 600

. . and expected the lag to get a little worse, but it has almost entirely cleared up. I will give it a day then up the stakes again, probably to -tt 800.

Stephen

.
ID: 1829335 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1829346 - Posted: 9 Nov 2016, 16:24:13 UTC - in response to Message 1829287.  

New v8.20 GPU apps was deployed recently on beta - please do massive testing to speedup their release to main.



Ok, will move a computer back over to beta here in a few minutes.
ID: 1829346 · 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 1829362 - Posted: 9 Nov 2016, 17:18:11 UTC - in response to Message 1829316.  

Testing continues up to release to main or explicit end of testing notice.
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1829362 · Report as offensive
Profile tullio
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 8797
Credit: 2,930,782
RAC: 1
Italy
Message 1829365 - Posted: 9 Nov 2016, 17:22:17 UTC

I am getting both 8.20 and 8.01 on my Windows PC with GTX 750 and they both run OK.
Tullio
ID: 1829365 · 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 1829366 - Posted: 9 Nov 2016, 17:22:29 UTC - in response to Message 1829335.  

probably to -tt 800.

Stephen

.


-tt N per ReadMe means the targeted kernel time so app will attempt to run kernel as long as 800 ms. That is, 0.8s. Of course such long running kernel will create quite noticeable lags. Such big -tt values recommended only for high-end cards that doesn't experience (or operator doesn't care) any lags to improve performance up to possible limit w/o any load balancing for usability.
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1829366 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1829450 - Posted: 9 Nov 2016, 22:22:53 UTC - in response to Message 1829415.  

New v8.20 GPU apps was deployed recently on beta - please do massive testing to speedup their release to main.

Hard to do Beta testing when the whole Beta project went belly up.
Ah well, maybe later.

Beta is back up.
ID: 1829450 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1829477 - Posted: 10 Nov 2016, 0:25:44 UTC - in response to Message 1829366.  
Last modified: 10 Nov 2016, 0:28:25 UTC

probably to -tt 800.

Stephen

.


-tt N per ReadMe means the targeted kernel time so app will attempt to run kernel as long as 800 ms. That is, 0.8s. Of course such long running kernel will create quite noticeable lags. Such big -tt values recommended only for high-end cards that doesn't experience (or operator doesn't care) any lags to improve performance up to possible limit w/o any load balancing for usability.


. . Hi Raistmer

. . Am I wrong in believing that this is the maximum window for kernel runtime. That is, if the kernel does not need to run so long it will end in less time. So if the kernels created run in less than 800mS on the GPU then the effects will be less noticeable.

. . Before Blc2s I was running triples with it set to -tt 800 and it was running pretty well. For Blc2s I had to drop back to running doubles, increase -period_iterations_num to 5 and reduce -tt to 500. It seems that Blc2s require longer kernels than the other greenbank tasks, like the Blc6s, or am I misunderstanding that point?

Stephen

.
ID: 1829477 · 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 1829568 - Posted: 10 Nov 2016, 9:55:25 UTC - in response to Message 1829477.  
Last modified: 10 Nov 2016, 9:57:22 UTC

It's desired time. If kernel took long it will be split to few calls if possible, if it took less next kernel calls will be merged into one if possible.
Possibility of merging/splitting depends on particular data pattern so yes, different tasks will induce different app behavior and different lags if any.
Also, initial kernel duration depends on initial data pattern and startup splitting coefficient that -period_iterations_num N provides. So this option still can reduce (or introduce) lags.
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1829568 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1829887 - Posted: 11 Nov 2016, 19:48:14 UTC - in response to Message 1829287.  

New v8.20 GPU apps was deployed recently on beta - please do massive testing to speedup their release to main.


. . Hi Raistmer,

. . Murphy has struck again, after confirming AOK running Lunatics 0.45 Beta 6 on La_Bamba I have run my cache dry and tried to join the Beta-Test project only to be told the project is not available, is it down again ..........?

Stephen

.
ID: 1829887 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1829889 - Posted: 11 Nov 2016, 19:55:16 UTC - in response to Message 1829887.  

New v8.20 GPU apps was deployed recently on beta - please do massive testing to speedup their release to main.


. . Hi Raistmer,

. . Murphy has struck again, after confirming AOK running Lunatics 0.45 Beta 6 on La_Bamba I have run my cache dry and tried to join the Beta-Test project only to be told the project is not available, is it down again ..........?

Stephen

.

Beta seems to be working fine for me at the moment.
If your host isn't liking one of the URLs maybe try the other?
https://setiathome.berkeley.edu/beta/
http://setiweb.ssl.berkeley.edu/beta/
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1829889 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1829911 - Posted: 11 Nov 2016, 21:28:39 UTC - in response to Message 1829889.  
Last modified: 11 Nov 2016, 21:31:41 UTC

New v8.20 GPU apps was deployed recently on beta - please do massive testing to speedup their release to main.


. . Hi Raistmer,

. . Murphy has struck again, after confirming AOK running Lunatics 0.45 Beta 6 on La_Bamba I have run my cache dry and tried to join the Beta-Test project only to be told the project is not available, is it down again ..........?

Stephen

.

Beta seems to be working fine for me at the moment.
If your host isn't liking one of the URLs maybe try the other?
https://setiathome.berkeley.edu/beta/
http://setiweb.ssl.berkeley.edu/beta/


. . Hi Hal,

. . Sorry but it was a false alarm. I had been e-mailed instructions on how to join Beta-Test some time ago, so I referred back to that e-mail. But you are correct about the problem being the URL. I had forgotten that there was a typo in the URL given in that e-mail and used the URL as given. This gave the result message "this project is not available" and with the earlier problems I foolishly presumed it was down again. I have now saved that e-mail as a text file, with the URL corrected, for future reference.

. . Thanks for the response.

Stephen

.
ID: 1829911 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1830642 - Posted: 15 Nov 2016, 1:14:23 UTC

. . Hi Raistmer,

. . FWIW I have had a couple of inconclusives in Beta, I posted the links in the Beta message board under V8.20 inconclusives.

. . The first wingman was running on an i7 using the HD4600 GPU and the second was running Apple Darwin. {biting my tongue hard}

. .

Stephen

.
ID: 1830642 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 9 · Next

Message boards : Number crunching : SETI@home v8.19 Windows GPU applications support thread


 
©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.