Just wondering if there's any good reason to upgrade anything?


log in

Advanced search

Message boards : Number crunching : Just wondering if there's any good reason to upgrade anything?

Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · Next
Author Message
KB7RZF
Volunteer tester
Avatar
Send message
Joined: 15 Aug 99
Posts: 9463
Credit: 3,107,734
RAC: 2,376
United States
Message 1217808 - Posted: 13 Apr 2012, 21:15:12 UTC - in response to Message 1217661.

Al, don't forget, we have a new type of work unit coming out very soon so you will have to change things for it. I believe all of us running anonymous platform will have to change but I'm not sure what will happen for those running stock.

When they released the updated MB work, stock app's will automatically update. Those running Lunatic's opp apps will have to wait till they released an updated optimized app to run the new work.
____________

AlProject donor
Send message
Joined: 3 Apr 99
Posts: 481
Credit: 52,040,624
RAC: 23,713
United States
Message 1218284 - Posted: 14 Apr 2012, 15:51:35 UTC

LOL that figures, but I suppose it's a constant arms race, progress and all that. What is the new type of WU called, and is there a thread on it to learn more? Is this something that is supported by v7 natively? I presume Lunatics is working on it frantically... ;-)

Has there been any talk of making bigger WU's (bundle a bunch together?), like the AP ones, to reduce traffic on the server? It may be a wash in terms of strain on capacity of the connection through the campus to the servers, but if fewer bigger units were used, it may help out with throughput on the server end. Maybe something that is flagged for systems that can handle them in a timely manner, and have the smaller ones reserved for older systems that may have an issue with them? Not sure the mechanics behind it, but if there is a way to do that, it may help. Good idea or not?
____________

Josef W. SegurProject donor
Volunteer developer
Volunteer tester
Send message
Joined: 30 Oct 99
Posts: 4229
Credit: 1,042,592
RAC: 335
United States
Message 1218339 - Posted: 14 Apr 2012, 18:04:40 UTC - in response to Message 1218284.

LOL that figures, but I suppose it's a constant arms race, progress and all that. What is the new type of WU called, and is there a thread on it to learn more? Is this something that is supported by v7 natively? I presume Lunatics is working on it frantically... ;-)
...

BOINC 7.x does not process S@H tasks, it's merely an Infrastructure for Network Computing. BOINC 7.x will be needed on hosts running stock if/when this project has stock OpenCL applications. Earlier BOINC versions do not send back the OpenCL properties.

The new SETI@home application, however, will have the official name setiathome_v7 and the "user friendly" name SETI@home v7. The added type of signal searching is Autocorrelation. There's an old thread, SETI@home v7 Beta is coming.
Joe

Profile betregerProject donor
Avatar
Send message
Joined: 29 Jun 99
Posts: 2232
Credit: 4,713,638
RAC: 10,242
United States
Message 1218340 - Posted: 14 Apr 2012, 18:15:25 UTC - in response to Message 1218339.

LOL that figures, but I suppose it's a constant arms race, progress and all that. What is the new type of WU called, and is there a thread on it to learn more? Is this something that is supported by v7 natively? I presume Lunatics is working on it frantically... ;-)
...

BOINC 7.x does not process S@H tasks, it's merely an Infrastructure for Network Computing. BOINC 7.x will be needed on hosts running stock if/when this project has stock OpenCL applications. Earlier BOINC versions do not send back the OpenCL properties.

The new SETI@home application, however, will have the official name setiathome_v7 and the "user friendly" name SETI@home v7. The added type of signal searching is Autocorrelation. There's an old thread, SETI@home v7 Beta is coming.
Joe

This is very good news, I have been wondering for some time about when the autocorrelation feature would be implemented.
____________

AlProject donor
Send message
Joined: 3 Apr 99
Posts: 481
Credit: 52,040,624
RAC: 23,713
United States
Message 1219788 - Posted: 18 Apr 2012, 3:28:14 UTC - in response to Message 1218339.

Josef, thanks for the links, and the clarification. I knew that BOINC and SETI were different entities, but it's good to put it out there from time to time. It was nice to read up on autocorrelation, as well as the v7 info. Keep up the good work!
____________

AlProject donor
Send message
Joined: 3 Apr 99
Posts: 481
Credit: 52,040,624
RAC: 23,713
United States
Message 1220914 - Posted: 21 Apr 2012, 2:34:51 UTC
Last modified: 21 Apr 2012, 2:38:12 UTC

Guys, could someone please take a sec and look at my multi GPU client, and see if there is anything they can tell about my invalid/error results. I know there aren't a lot of them, but before I upgraded, I had 0 of either, and now I have some. If there's an issue, I'd like to nip it in the bud before it pollutes any more. Also, it appears that my RAC is dropping slightly, by about 2k or so. That may be a standard variation, or maybe the way it's doing VLR or AP tasks since I upgraded. Can you tell by my recent results if I should be running them on my GPU or CPU exclusively? Thanks!
____________

Profile arkaynProject donor
Volunteer tester
Avatar
Send message
Joined: 14 May 99
Posts: 3622
Credit: 48,536,020
RAC: 35,300
United States
Message 1221126 - Posted: 21 Apr 2012, 6:52:36 UTC - in response to Message 1220914.

Guys, could someone please take a sec and look at my multi GPU client, and see if there is anything they can tell about my invalid/error results. I know there aren't a lot of them, but before I upgraded, I had 0 of either, and now I have some. If there's an issue, I'd like to nip it in the bud before it pollutes any more. Also, it appears that my RAC is dropping slightly, by about 2k or so. That may be a standard variation, or maybe the way it's doing VLR or AP tasks since I upgraded. Can you tell by my recent results if I should be running them on my GPU or CPU exclusively? Thanks!


Most of them are -12 triplet errors and nothing to worry about, but 3 of them are -131 errors which were the oversize WU from this thread.
http://setiathome.berkeley.edu/forum_thread.php?id=67680
Also nothing on your end.
____________

AlProject donor
Send message
Joined: 3 Apr 99
Posts: 481
Credit: 52,040,624
RAC: 23,713
United States
Message 1223786 - Posted: 27 Apr 2012, 0:56:53 UTC

Well, I've been watching some more, and am still trying to decide weather to continue to do standard AP v6.01 tasks or not. I notice that the recent ones run 14+ hours, but looking at the same ones but labeled cuda_fermi, they run about 2 hours. Does the RAC you receive for the standard ones equal 7 of the fermi ones? If not, would it make sense to just do the cuda ones, and pass on the std ones? Is that even possible, or ethical? I really haven't paid close attn to the RAC/task type ratio, could someone please try to explain it to me?
____________

Josef W. SegurProject donor
Volunteer developer
Volunteer tester
Send message
Joined: 30 Oct 99
Posts: 4229
Credit: 1,042,592
RAC: 335
United States
Message 1223829 - Posted: 27 Apr 2012, 3:48:23 UTC - in response to Message 1223786.

Well, I've been watching some more, and am still trying to decide weather to continue to do standard AP v6.01 tasks or not. I notice that the recent ones run 14+ hours, but looking at the same ones but labeled cuda_fermi, they run about 2 hours. Does the RAC you receive for the standard ones equal 7 of the fermi ones? If not, would it make sense to just do the cuda ones, and pass on the std ones? Is that even possible, or ethical? I really haven't paid close attn to the RAC/task type ratio, could someone please try to explain it to me?

There are no AP v6 tasks labeled cuda_fermi, so it's a complex question what to run to get the highest RAC on any given system. The AP tasks generally average roughly 700 cobblestones, MB tasks like the cuda_fermi ones may over fairly long periods average about 100 cobblestones but the amount of work and credit varies with angle range.

One factor to consider may be that even though one AP task may on average yield 7 times as much credit, it's probably easier to download 7 MB WUs than one AP WU. And there are times when there are no AP tasks being produced.
Joe

AlProject donor
Send message
Joined: 3 Apr 99
Posts: 481
Credit: 52,040,624
RAC: 23,713
United States
Message 1223926 - Posted: 27 Apr 2012, 11:57:32 UTC - in response to Message 1223829.

Josef, thanks for the explanation, I misspoke when I said task, I meant application. Right now for example I have 2 tasks going, one is ap_15jn11ac_B3_P0_00121_20120423_28241.wu_0 using the AP 6.01 (cuda_fermi) app, & the 2nd task is ap_15jn11aa_B1_P1_00102_20120422_20571.wu_0 using the AP 6.01 app. The fermi labeled task is at 71% and has ran for 1:14:00 with 0:31:00 till completion, while the standard task is at 63%, has ran for 7:57:20 and has 4:55:05 remaining. I think you're saying that those are both the same type of tasks, just using different apps to do them? When they are done, would they yield the same amount of credit? If so, it seems that the standard app isn't designed to do them nearly as efficiently as the AP app, and it would be better to leave them to that app, letting the CPU chew on tasks using the standard setiathome_enhanced 6.03 app? Thoughts on my reasoning?
____________

Profile jason_geeProject donor
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 4963
Credit: 73,068,376
RAC: 14,767
Australia
Message 1223928 - Posted: 27 Apr 2012, 12:18:41 UTC - in response to Message 1223926.
Last modified: 27 Apr 2012, 12:27:31 UTC

ap_15jn11ac_B3_P0_00121_20120423_28241.wu_0 using the AP 6.01 (cuda_fermi)


No such thing. If you don't know what it is or how it got there, I would recommend aborting it. (as a start, prior to investigation as to why you'd be getting tasks assigned to non-existent applications)
____________
"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change."
Charles Darwin

LadyL
Volunteer tester
Avatar
Send message
Joined: 14 Sep 11
Posts: 1679
Credit: 5,230,097
RAC: 0
Message 1223933 - Posted: 27 Apr 2012, 12:38:39 UTC - in response to Message 1223928.

ap_15jn11ac_B3_P0_00121_20120423_28241.wu_0 using the AP 6.01 (cuda_fermi)


No such thing. If you don't know what it is or how it got there, I would recommend aborting it. (as a start, prior to investigation as to why you'd be getting tasks assigned to non-existent applications)


it's the openCL app. What it identifies itself as in the () depends on what you have put in the <plan_class> field.

Since that app is beta and not installer supplied, that part of the app_info is self-written. The string could be anything. it's a bit tricky to get that relabled and not lose work.
____________
I'm not the Pope. I don't speak Ex Cathedra!

LadyL
Volunteer tester
Avatar
Send message
Joined: 14 Sep 11
Posts: 1679
Credit: 5,230,097
RAC: 0
Message 1223938 - Posted: 27 Apr 2012, 12:50:33 UTC - in response to Message 1223926.
Last modified: 27 Apr 2012, 12:54:56 UTC

Josef, thanks for the explanation, I misspoke when I said task, I meant application. Right now for example I have 2 tasks going, one is ap_15jn11ac_B3_P0_00121_20120423_28241.wu_0 using the AP 6.01 (cuda_fermi) app, & the 2nd task is ap_15jn11aa_B1_P1_00102_20120422_20571.wu_0 using the AP 6.01 app. The fermi labeled task is at 71% and has ran for 1:14:00 with 0:31:00 till completion, while the standard task is at 63%, has ran for 7:57:20 and has 4:55:05 remaining. I think you're saying that those are both the same type of tasks, just using different apps to do them? When they are done, would they yield the same amount of credit? If so, it seems that the standard app isn't designed to do them nearly as efficiently as the AP app, and it would be better to leave them to that app, letting the CPU chew on tasks using the standard setiathome_enhanced 6.03 app? Thoughts on my reasoning?


Apples, pears and olives.

With CreditNew nobody know how much credit you'll get - you should get about the same whether it was done on the GPU or on the CPU (with at least 10% variace if not much more).

You don't have 'standard' tasks or apps. There are Multibeam MB tasks for with the installer hase something that calls itseld 6.03 for the CPU and 6.10 for the GPU. You have AP for which the installer has 6.01 for CPU. You have the manual install of the OpenCL NV AP app, which you seem to have labled 6.01 (cuda_fermi).

Effeciency is entirely relative.
If you want to estimate effeciency look at what an AP task takes on the GPU comapred to a MB task on the GPU. same for CPU. Then you can decide whch device is more efficent at which app.

On my system I see MB taking 3 hours on teh CPU and 1.5h on the GPU. I see AP taking 15 hours on CPU and 16 hours on GPU.

As I can't go higher on the AP parametres without getting driver restarts, I guess the MB app is FAR superior here.

But 'you milage may vary' you'll have to see on your own system.

edit: btw we are still chasing an error in the NV AP app. You seem to have a few invalids from it. I'm seriously considering disenfranching that app until the error has been fixed.
____________
I'm not the Pope. I don't speak Ex Cathedra!

Profile jason_geeProject donor
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 4963
Credit: 73,068,376
RAC: 14,767
Australia
Message 1223939 - Posted: 27 Apr 2012, 12:54:33 UTC - in response to Message 1223933.
Last modified: 27 Apr 2012, 13:21:54 UTC

ap_15jn11ac_B3_P0_00121_20120423_28241.wu_0 using the AP 6.01 (cuda_fermi)


No such thing. If you don't know what it is or how it got there, I would recommend aborting it. (as a start, prior to investigation as to why you'd be getting tasks assigned to non-existent applications)


it's the openCL app. What it identifies itself as in the () depends on what you have put in the <plan_class> field.

Since that app is beta and not installer supplied, that part of the app_info is self-written. The string could be anything. it's a bit tricky to get that relabled and not lose work.


Oh, that [application], figures, nuff said. I have no idea what that's doing in public release (beta or otherwise), or being labelled (misrepresented) as Cuda when it isn't. nevermind, nothing to do with me.
____________
"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change."
Charles Darwin

Profile perryjay
Volunteer tester
Avatar
Send message
Joined: 20 Aug 02
Posts: 3377
Credit: 15,300,701
RAC: 11,725
United States
Message 1223959 - Posted: 27 Apr 2012, 15:02:57 UTC - in response to Message 1223939.

That's just a quick way to tell it's being run on our GPUs I guess. (As if the speed difference wouldn't give it away! :) )

____________


PROUD MEMBER OF Team Starfire World BOINC

Profile arkaynProject donor
Volunteer tester
Avatar
Send message
Joined: 14 May 99
Posts: 3622
Credit: 48,536,020
RAC: 35,300
United States
Message 1223996 - Posted: 27 Apr 2012, 17:07:11 UTC

Plus you need BOINC 7.x.x to use a OpenCL class, calling it CUDA works with 6.x.x.
____________

AlProject donor
Send message
Joined: 3 Apr 99
Posts: 481
Credit: 52,040,624
RAC: 23,713
United States
Message 1229826 - Posted: 10 May 2012, 0:52:15 UTC

OK, I can't tell what exactly is going on, but something has happened to my machine with the update I did last month. I've been letting it run pretty much unattended and crunching like I did before doing this update, and have been watching my RAC drop by 20%+. I don't understand, as nothing has changed other than my client, the optimized app and the video driver, but I did the update on the 9th of April, and it peaked on the 22nd with a RAC of about 58,000. Looking at my Host Average in BOINC Mgr, it's now hovering between 44-45,000. That is a huge hit for doing nothing more than getting the current "best" setup to crunch. Not that I am a credit hog, but it is an indication that my machine doesn't appear to any longer be living up to it's potential. Can anyone give me any ideas as to what is happening? It's kind of frustrating, taking a known good working system, 'upgrading' it, and apparently getting significantly worse performance for all the effort. Very discouraging, I have to say, and of course I didn't Ghost the machine so I could restore it to the way it was before I began. I am currently not taking new tasks so I can burn thru what I have and be able to make any changes you guys may suggest, to get it somewhere close to where it used to be, getting it back to over 53-55k is my goal.
____________

Profile Gundolf Jahn
Send message
Joined: 19 Sep 00
Posts: 3184
Credit: 357,953
RAC: 37
Germany
Message 1229936 - Posted: 10 May 2012, 6:41:29 UTC - in response to Message 1229826.

Which of your machines?

None of your hosts shows as running BOINC 7.0.x.

Did you switch the values for "Maintain enough tasks..." and "... and up to an additional" in your Computing preferences?

Gruß,
Gundolf

ClaggyProject donor
Volunteer tester
Send message
Joined: 5 Jul 99
Posts: 4067
Credit: 32,877,796
RAC: 6,787
United Kingdom
Message 1229948 - Posted: 10 May 2012, 9:23:46 UTC - in response to Message 1229826.
Last modified: 10 May 2012, 9:29:04 UTC

OK, I can't tell what exactly is going on, but something has happened to my machine with the update I did last month. I've been letting it run pretty much unattended and crunching like I did before doing this update, and have been watching my RAC drop by 20%+. I don't understand, as nothing has changed other than my client, the optimized app and the video driver, but I did the update on the 9th of April, and it peaked on the 22nd with a RAC of about 58,000. Looking at my Host Average in BOINC Mgr, it's now hovering between 44-45,000. That is a huge hit for doing nothing more than getting the current "best" setup to crunch. Not that I am a credit hog, but it is an indication that my machine doesn't appear to any longer be living up to it's potential. Can anyone give me any ideas as to what is happening? It's kind of frustrating, taking a known good working system, 'upgrading' it, and apparently getting significantly worse performance for all the effort. Very discouraging, I have to say, and of course I didn't Ghost the machine so I could restore it to the way it was before I began. I am currently not taking new tasks so I can burn thru what I have and be able to make any changes you guys may suggest, to get it somewhere close to where it used to be, getting it back to over 53-55k is my goal.

Your i7 950 with it's four GTX260 is throwing huge amounts of inconclusives because of the Buggy Nvidia OpenCL Astropulse app, most of those inconclusives will turn out as being invalid,
the clue being that repetitive pulses is overflowing while single pulses isn't (single pulses should overflow first), please discontinue running the Nvidia Astropulse app until the Bug fixed version is released,

Pending AstroPulse v6 tasks for computer 5873188

see this thread for further info: Inconclusive Work Units Running AP Ver 6

Claggy

AlProject donor
Send message
Joined: 3 Apr 99
Posts: 481
Credit: 52,040,624
RAC: 23,713
United States
Message 1229987 - Posted: 10 May 2012, 12:51:35 UTC - in response to Message 1229948.

Thanks for the link, since obviously I was effected by this bug, should I consider joining the ranks of the beta testers? I think my cache will be dry by this afternoon when I get home from work, so I would be in a good position to help if needed.
____________

Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · Next

Message boards : Number crunching : Just wondering if there's any good reason to upgrade anything?

Copyright © 2014 University of California