Message boards :
Number crunching :
Lunatics Windows Installer v0.40 release notes
Message board moderation
Previous · 1 . . . 6 · 7 · 8 · 9 · 10 · 11 · 12 . . . 14 · Next
Author | Message |
---|---|
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
(removed from memory) If you have a reasonably strong machine with a decent amount of memory and disk (swap file) space - i.e. almost anything built in the last ten years - you will find it more efficient to select the preference to Leave tasks in memory while suspended? |
Cliff Harding Send message Joined: 18 Aug 99 Posts: 1432 Credit: 110,967,840 RAC: 67 |
(removed from memory) Changing the preference is not a problem, which I will try. My question is, why the preemption in the first place? Never was never an issue until after installing the new Lunatics. I don't buy computers, I build them!! |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
(removed from memory) I would expect that your first guess was right - it's likely to be because you now have Astropulse v6 tasks in your cache. They will have two characteristics that you can see in BOINC Manager: 1) An estimated runtime which (falsely) thinks that the tasks will take 150-200 hours. 2) A deadline which is closer than all MB tasks except shorties. Once you've processed 50 AP v6 tasks or so, the estimates will - quite suddently - return to sanity, and normal service will be resumed. Remind me to reply to Eric's email thoughts on that subject. |
Cliff Harding Send message Joined: 18 Aug 99 Posts: 1432 Credit: 110,967,840 RAC: 67 |
(removed from memory) Currently I have 53 MB tasks with a deadline before 21 April & 67 after 22 April, some of which have an approx running time of 1-2 hrs. So what happens to the 64 CUDA tasks that have a deadline of 11 April? How much is this depended on BOINC 7.0.22 and how much on Lunatics 0.40? I don't buy computers, I build them!! |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
(removed from memory) CUDA tasks will run in their own queue, in their own order, while MB and AP fight over the CPUs. Most of this is down to SETI@home, a little is due to BOINC (any/every version), and the only contribution by Lunatics v0.40 is that it allows you to process Astropulse v6 in the first place. Arrrgh, no - BOINC v7.0.22 and BOINC v7.0.23 (only) have a specific bug which may stop CUDA processing under these circumstances. Better to use v7.0.20 (or .21 - though I haven't tested that one). v7.0.24 isn't available yet, but will - subject to testing - have the fix I've been working on with David. Did I hear anybody mention the words 'high priority'? |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
It's possible to extend the CUDA app with cmdline settings, so the members could in-/decrease the priority himself? It would be nice to have this possibility, so I wouldn't need to use a 3rd party tool for to increase the priority.. It's possible to make a bench-test tool, a very easy for noobs like me, one click and the program say which app (CPU extension usage) is the best/fastest for the machine? Yes, but it didn't worked. I tested it with two tools. Now I know it through a few tests, r555 need a .DLL file - which I thought is only with a CUDA app needed. I just wanted to make a bench-test for to see which AP 6.01 app (r555 vs. r557) is faster on my machine, but I failed.. Thanks! I used the short AP WU: http://lunatics.kwsn.net/index.php?module=Downloads;sa=dlview;id=232. The result: AP6_win_x86_SSE_CPU_r555.exe Elapsed 290.766 secs CPU 288.547 secs ap_6.01r557_SSE2_331_AVX.exe Elapsed 276.469 secs CPU 274.297 secs I did everything fine, so it means the r557 app run faster on my machine*? [* Intel Core2 Duo E7600 @ 3.06 GHz, DDR2 800/5-5-5-18 (all stock, not OCed), WinXP 32bit] - Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC. - SETI@home needs your help. - |
Cliff Harding Send message Joined: 18 Aug 99 Posts: 1432 Credit: 110,967,840 RAC: 67 |
(removed from memory) High priority was stated in my first post here. It seems the 7.0.22 is the culprit and I will immediatedly revert back to 7.20, and will let you know what happens. I don't buy computers, I build them!! |
Cliff Harding Send message Joined: 18 Aug 99 Posts: 1432 Credit: 110,967,840 RAC: 67 |
BINGO!! Problem solved. Many thanks to everyone and especially Richard Haselgrove who nailed the main problem! Reverted to 7.0.20 and GPU activity immediately started. Some went into High Priority to catch up as was expected. Back to my B-SYS rebuild which is almost completed. I don't buy computers, I build them!! |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
High priority was stated in my first post here. It seems the 7.0.22 is the culprit and I will immediatedly revert back to 7.20, and will let you know what happens. Ah, I see what happened. Yes, all the clues are there in your first post, but my tired old brain failed to line them up and join the dots, even though it's one of those problems which has a mental flag of "my bug" on it. Then, when the penny eventually dropped, I did a search of the thread for 'high priority', and failed to find your 'high-priority'. D****d literal computer searches - why can't they ignore hyphens? Anyway, we got there in the end. v7.0.20 should be better in the short term. |
Cliff Harding Send message Joined: 18 Aug 99 Posts: 1432 Credit: 110,967,840 RAC: 67 |
High priority was stated in my first post here. It seems the 7.0.22 is the culprit and I will immediatedly revert back to 7.20, and will let you know what happens. From an old man, you are probably much younger than I. That said, knowing how hard you have been working on everything and taking out time to solve my little problem; TAKE 5-10 MINUTES FOR YOURSELF, believe me it will be worth it. We can't afford to have you burned out before your time. I don't buy computers, I build them!! |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
From an old man, you are probably much younger than I. Aging by the minute. :-( That said, knowing how hard you have been working on everything and taking out time to solve my little problem; TAKE 5-10 MINUTES FOR YOURSELF, believe me it will be worth it. We can't afford to have you burned out before your time. Actually, I managed to get out for half-an-hour's stroll round the village in the spring sunshine, between a couple of the posts there. The benches outside the pub are looking particularly inviting for later this evening. |
shizaru Send message Joined: 14 Jun 04 Posts: 1130 Credit: 1,967,904 RAC: 0 |
Think you should take the opinion that "no news, is good news". People don't usually hesitate to complain. Words of wisdom. Installer seems idiot-proof to me, and I should know since I'm (still) Boinc-stupid:) Thanx to all who have worked on the package and it's insides! I would like to take this opportunity to to add a couple bullet-points to your never-ending wish-list. They will make you gurus cringe, I know that, but hopefully they will make enough sense that you won't hold what I'm about to say against me. So in a philosophical tone, not a demanding one, here goes: Dare a say "sticky"?:) Here's the thing... I need a way to know who needs my PC and when, for what and for how long. If I can volunteer my little laptop to run something for somebody (for Seti/Seti Beta or Lunatics) I'd be more than happy to. Please don't take this the wrong way but I've read this whole thread and still have no idea if there is some version of some app over at Lunatics that I could run for you guys and help you get just that little bit closer to eventually releasing it. And the other day over at Beta I had to ask Eric what he needed us to crunch. I'm sure most of you will agree that it's not a Project Scientist's job to answer my questions, and I'd rather not do it again. So if there was a place I could check in and see that Astro vX over at Beta needs crunching for a few weeks or which Lunatics app vX needs crunching and for how long, it would be great! In summary, I'd love to help but I have no way of knowing how to help. This next part is almost completely philosophical. If I had a "pretty" benchmark app I would go OCD on the thing. I'd check every WHQL nVidia driver from late 250's to current 290's. I'd turn Windows eye-candy, services and processes on and off. I'd play with nVidia settings. And whenever Jason-G would come out of left-field with, "Oh, you know, it could be your Wi-Fi that's interfering with crunching" I'd check that too:) I say "pretty" because I don't know how to work with black & white windows. I need installers and progress bars and buttons I can click on and things to hold my hand:) Of course I'm sure you guys have next-to NO time for such a thing which is why I started this paragraph with, "This next part is almost completely philosophical". OK, break's over! Back to the Lunatics Windows Installer v0.40:) |
Cosmic_Ocean Send message Joined: 23 Dec 00 Posts: 3027 Credit: 13,516,867 RAC: 13 |
My eight-banger AMD is running 4 cores only, but two nVidia cards two at a time with no interference. Like you, I'm looking forward to seeing what AVX does with AP v6 when I finally get to them. From what I've seen in my handy spreadsheet so far.. same hardware and everything from r409 (v505) to r557 with AVX (v6), v6 runs about 3,000 seconds faster. Down from ~43,000 seconds median to ~39,500 seconds median. Some quick math says that's ~9% faster. Now does SSE2 have anything to do with it, or is that all AVX? Either way, faster is faster. Linux laptop: record uptime: 1511d 20h 19m (ended due to the power brick giving-up) |
LadyL Send message Joined: 14 Sep 11 Posts: 1679 Credit: 5,230,097 RAC: 0 |
I'm pretty sure Jason will come up with something apropriate.
The app needs a fftw lib (as does MB). r557 has that statically linked, r555 needs the dll. It doesn't say it needs the dll in the app_info because we found on windows, win will go looking for the dll in the project dir.
We've found that in most cases r557 is faster IF benching is done correctly. If you want to do more benching please PM - there are additional difficulties that I'd rather explain in German. I'm not the Pope. I don't speak Ex Cathedra! |
JLConawayII Send message Joined: 2 Apr 02 Posts: 188 Credit: 2,840,460 RAC: 0 |
Will those work to fix this on the ATI client as well? I don't understand why half of my WUs run normally and half of them crap out. Yeah I know, those guys were around last night when all this was going on. Hopefully everyone has gotten enough information now to deal with the issues appropriately. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
A couple of comments on freeing one core, as a refinement but not as a recommendation. It is 1, but there's enough compexity in how the fractions get combined that I agree with your idea of a safety factor. In fact in the last row of that table even the <count> is questionable. So here's a revised version: tasks/ 1 GPU 2 GPUs 3 GPUs 4 GPUs GPU <count> <avg_ncpus> <avg_ncpus> <avg_ncpus> <avg_ncpus> 1 1.0 1.001 0.501 0.334 0.251 2 0.5 0.501 0.251 0.167 0.126 3 0.33 0.334 0.167 0.112 0.084 4 0.25 0.251 0.126 0.084 0.063 5 0.199 0.201 0.101 0.067 0.051 Joe |
Michael W.F. Miles Send message Joined: 24 Mar 07 Posts: 268 Credit: 34,410,870 RAC: 0 |
Seeing the thread exploded over night, if I don't get to your question, please don't hesitate to repeat it. @Micheal Miles - The NVidia OpenCL apps run on 270+ drivers but hog a CPU core. You'll either have overcommit or you can do a workaround and free one core - I think Joe posted how to do that. Thank you for the answer I will stay with the present drivers as it is working very well As to the command line for nv r555 I have increased this to <cmdline>-instances_per_device 1 -unroll 6 -ffa_block 4096 -ffa_block_fetch 2048 -sbs 256</cmdline> From <cmdline>-instances_per_device 1 -unroll 4 -ffa_block 2048 -ffa_block_fetch 1024 -sbs 128</cmdline> sbs 256 ???? I have increased this to 256 from 128 Is this the ram the 460 is using and will this increase cause a problem? Michael Miles |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Some additional clarification about -sbs param: This param (single block size) will govern size of variable size buffer for upcoming MultiBeam update. It was also added to AP but for now it does nothing for AP. Any variation in memory allocation comes via unroll param as before. In next builds this param can be used for AP too. So, if one see memory-related issues with AP one should decrease unroll param. But don't forget, that minimal possible value is 2, not 1. Corresponding buffer shared inside program and can't be made too little. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Think you should take the opinion that "no news, is good news". People don't usually hesitate to complain. Lol, "black-and-white boxes" is very educational in our click and see era ;D If one master command line... well, maybe one will learn something new ;) And more serious answer: sorry we have no manpower currently to implement GUI-based benchmark tools. Current KWSN-bench based tools became much more advanced than were before thanks to Joe's work and can be used to test single core /multicore busy, different app params and so on apps behavior. Please take some time to read excellent readme for it Joe wrote and use it for tuning your PC. There is client_state.xml parcing script I wrote to get real-life data about apps performance, but it's Perl script and so "black-and-white" one that I'm scared even to mention it in this thread ;D. Regarding what apps need test. You need to monitor SETI beta area for that. We (I in particular) do posts about beta testing of some new builds there. And yes, if there is some new build to test, then it needs to be tested ;). Sometimes new build to test posted on SETI main too, announce about that goes in separate thread. Such info never goes in installer release notes and corresponding threads. Cause installer includes already tested apps, not apps to test. |
Michael W.F. Miles Send message Joined: 24 Mar 07 Posts: 268 Credit: 34,410,870 RAC: 0 |
I have been getting long run times with mu cmdline set to <cmdline>-instances_per_device 1 -unroll 6 -ffa_block 4096 -ffa_block_fetch 2048</cmdline> Took 3 hours / wu I have set to this <cmdline>-instances_per_device 1 -unroll 10 -ffa_block 8144 -ffa_block_fetch 4096</cmdline> Now run times are under 2 hours Question There are two new files in my project folder AstroPulse_Kernels_r555.cl_GeForce GTX 460.bin_V6 r555_AMDPhenomtmIIX61100TProcessor.wisdom What are these??? Michael Miles |
©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.