Message boards :
Number crunching :
BOINC 7.6.9 - Scheduler behaves strange.
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 · Next
Author | Message |
---|---|
Darth Beaver Send message Joined: 20 Aug 99 Posts: 6728 Credit: 21,443,075 RAC: 3 |
Ulrich I have try'd to advise you on getting a better R.A.C for 2 gtx 640 GPU's I would expect you to be scoring at least 14,000 I was running my main rig with 2 GTX 650's Rac was well over 20,000 Good luck My Core 2 Qadd score is 90% MB's I'm not getting AP'S for the GPU at the moment on that 1 as there aren't enough to go around . |
Darth Beaver Send message Joined: 20 Aug 99 Posts: 6728 Credit: 21,443,075 RAC: 3 |
The reason your system has gone into EDF . The servers will give you only enough work to fill the cache but if your doing the units slower than what the servers think they should be done then they will run out of time and put Bionic into EDF . So lowering your cache may not help unless it now knows that the times are slower so it gives you less units . I have not had the Bionic client go into EDF but then I don't ask my machine to do the impossible and over work them . good luck |
Darth Beaver Send message Joined: 20 Aug 99 Posts: 6728 Credit: 21,443,075 RAC: 3 |
looking at the times that machine is doing 1 single unit per GPU I can see why it's gone into EDF 7900 sec run times for a single unit and 23,000 sec's The times should be closer to 3,000 and no more than 9,000 but no way should they get to 23,000 on that class of GPU My main rig has 1 650 in it (the severs are reporting I have 2) 680 but I only have 1 and the 650 is doing 2 AP's at a time and the run times do not exceed 9,000 sec per unit |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
|
rob smith Send message Joined: 7 Mar 03 Posts: 22223 Credit: 416,307,556 RAC: 380 |
EDF only applies to tasks that are seen by BOINC as not going to be completed by their deadline. There are a number of causes for this, first and perhaps most obvious is that the task was sent with a very short deadline. Another common one is that another project has grabbed all the resources for a time, and so the remaining tasks have to run in EDF to complete by their deadlines; another is that you have changed your crunching habits, say you used to crunch 24/7 bu now only crunch for an hour a week - your cache is full of work on the based on your previous level of activity; you have changed you cache settings to a pair that will lead to BOINC thinking some tasks (typically CPU tasks) will expire before their deadlines. There is a lot of confusion around the cache setting on the website - these are mine: Store at least 3 days of work Store up to an additional 0.1 days of work The first figure is the size of the cache, the second is an indicator as to how "late" a cache filling may take place. My PCs run 24/7, and when running SETI@Home I very rarely see any tasks running in EDF, however, if I reverse the two figures then just about everything runs in EDF (even though I know that I will clear 200 "GPU" tasks in most Tuesday outages... Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
Darth Beaver Send message Joined: 20 Aug 99 Posts: 6728 Credit: 21,443,075 RAC: 3 |
ok Ulrick your right I'm wrong . mm my first GPU was a first gen 9800 gt with only 16 CUDA cores it took 6 hrs to do a unit so I guess the tech hasn't improved much then and to think there's been 4 series since that card to get to your card and to think your card has 96 CUDA cores . I take it back don't by a core 2 Qadd go and by a new system and spend 600-1000 bucks . Seeing as you know so much you should know why I told you to get one and why it would be a cheep upgrade guess I should just let you waste your money . Wonder weather if you stopped using the CPU your times would improve . But hey the CPU is only 5% of your Rac so I spose it's a must thing to do and the GPU's would not be able to make up the slack CPU core does 2 units in 24hrs but a GPU should be able to do say 18 on 640 if they where being done at the right run times and the 430 I spose should be able to do 8 if they where being done at the right run times but hey you know better 3-4 units on your 430 in 24 hrs is pretty good . mm A CPU does 2 units per core so I guess the GPU 's are money well spent WHY THE HELL YOU ASKING QUESTIONS IN THE NUMBER CRUNCHING WHEN YOU WON"T LISEN TO WHAT PEOPLE ARE TELLING YOU . Better you stop wasting peoples freaking time and we can all ignore you seeing you don't wish to learn anything or take good honest advise meant to save you money and help you improve you R.A.C so "____ ___" fill in the blanks |
rob smith Send message Joined: 7 Mar 03 Posts: 22223 Credit: 416,307,556 RAC: 380 |
OK, so now we have some figures to look at. First observation, the two GPUs are running in the sort of region that it is reasonable to expect, thus can be assumed to be doing their job. The run times for an AstroPulse are about right for low to mid-range GPUs, so little to worry about on that front either. Now to the CPU - that is "strange". The grassy appearance is fairly typical of a processor that is doing a lot of task swapping. Let's see what is running (back to the data you posted a while back). Now it gets interesting - you have a two core processor, it is running two CPU tasks, while supporting two GPU OpenCL tasks, result is that the cores are struggling to keep up with all the task swapping going on. This is the source of your problem, and very much like the sort of problem that is observed on some older AMD processors - the time it takes to swap between a pair of CPU tasks (Note, this is tasks in the "Windows" meaning of the word, not necessarily BOINC tasks) is such that neither task really gets enough time to do anything constructive. So the solution? Well it is down to you, you could either let things run as they are, and not worry about not getting as high a RAC as one might expect from your system, or you could do something about it. It is quite simple to limit the amount of cores that BOINC uses directly - In BOINC Manager, advanced view, select Tools, Computing Preferences - a new window pops up, near the bottom of this window is a section called "other options", and in here are three options, the middle one determines how many of the available CPU cores are used directly by BOINC - set this to 50% (which means one core is left for BOINC processing, while the other is available to everything else, INCLUDING supporting OpenCL for the GPUs. Leave the other options as they are for now, we may need to come back and change them at another time. After this change I would expect the GPUs to continue as they are (or maybe a little faster), but the CPU processes to drop to One running, with possibly one "waiting to run". Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
Gordon Lowe Send message Joined: 5 Nov 00 Posts: 12094 Credit: 6,317,865 RAC: 0 |
WHY THE HELL YOU ASKING QUESTIONS IN THE NUMBER CRUNCHING WHEN YOU WON"T LISEN Please don't yell at people. The mind is a weird and mysterious place |
Darth Beaver Send message Joined: 20 Aug 99 Posts: 6728 Credit: 21,443,075 RAC: 3 |
mm wouldn't be this one http://www.nvidia.com/page/gpu_mobo_tech_specs.html onboard GPU if so it's PCI-Express not PCI and 9800 GT I had was AGP not PCI-E but i'll keep looking for a gtx 430 in a pci and not pci-e or agp would explain how you have 2 GPu's on a Core 2 MoBo |
Darth Beaver Send message Joined: 20 Aug 99 Posts: 6728 Credit: 21,443,075 RAC: 3 |
Na sorry I can't find a 430 pci there all PCI-E ALL OF THEM or do I have to list every one for you mate |
Darth Beaver Send message Joined: 20 Aug 99 Posts: 6728 Credit: 21,443,075 RAC: 3 |
Funny how they stopped making Graphics card with PCI about the time that AGP came out and now you can't buy a GPU as they are now called in a AGP anymore so how can you have a GPU that still work on the old 33mhz front bus PCI when they don't make any and have not for over 15 years ... Not good to lie about theses things when your talking to someone that's 52 and had his first computer when he was 17 which means I've seen and had pretty much every gen of MoBo and CPU and GPU that's come out . |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
With each crime and every kindness we birth our future. |
Darth Beaver Send message Joined: 20 Aug 99 Posts: 6728 Credit: 21,443,075 RAC: 3 |
So Mike you saying it's a agp or pci-e could be both can't really tell from that picture but certainly not a PCI and I did look at that one if it the one at the Nvida site which says PCI-e2 http://www.geforce.com/hardware/desktop-gpus/geforce-gt-430/specifications this one ?? |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14653 Credit: 200,643,578 RAC: 874 |
Glenn, when you're outside your personal experience, please stop digging. GT 430 cards were made in PCI (not express) configuration. Here's one: (source: http://www.newegg.com/Product/Product.aspx?Item=N82E16814500221) They may not be easily available at the moment, but if Ulrich has one (and many people have bought them for use on BOINC projects over the years), he is perfectly at liberty to go on using it. Edit - I didn't even notice this on the Newegg page! Top Favorable Review |
Bernie Vine Send message Joined: 26 May 99 Posts: 9954 Credit: 103,452,613 RAC: 328 |
Not good to lie about theses things when your talking to someone that's 52 and had his first computer when he was 17 which means I've seen and had pretty much every gen of MoBo and CPU and GPU that's come out . Please do not accuse others of lying. It took me two minutes to find there is one similar to the one Richard posted currently on sale on ebay. |
Darth Beaver Send message Joined: 20 Aug 99 Posts: 6728 Credit: 21,443,075 RAC: 3 |
Well thanks Richard I stand corrected there is 1 obscure 430 pci card . one question why does the servers say he has 2 640's and they have 2 gig or ram I would have thought it would say 1 gig seeing as that has only 512meg of ram . He certainly can't do any more than 1 unit with that card then . I didn't say he should stop using it . The spec's on that card are the same as the PCI-e and you know that seti is not that dependant on the speed of data flow , not so dependant on how fast the pci-e is 16x , 8 x 4x , 1x so I would not expect it to be as slow as it is . stil I'm not the only one that has told him he has two cores and doing more than 2 AP units will slow it down But thanks for giving him a out Might give an older desktop with no AGP mm he has a pci-e to fit the 640 so why buy it or is that also PCI |
rob smith Send message Joined: 7 Mar 03 Posts: 22223 Credit: 416,307,556 RAC: 380 |
Gents, I fear we may have headed off in a slightly wrong direction - Ulrich states in his post that it is the CPU that is doing strange things and not the GPUs which appear to be behaving themselves pretty much as one would expect for a heterogeneous pairing of cards. The CPUs on the hand are the one's producing the very extended run times, and are running in EDF mode. Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
Darth Beaver Send message Joined: 20 Aug 99 Posts: 6728 Credit: 21,443,075 RAC: 3 |
which brings us back to the fact I and others have told him he has 2 cores and that is the problem . The siv graph he put up , there is something he should note . The red part of the graph is the Kernels if he stops using the CPU's then that graph will show the Kernel in red as a constant unbroken line not what you see that it's jumping up and down with gaps in it he will also notice by not using the CPU's that there will be no green part of the graph . I use siv myself do either the GPU units 1 at a time per GPU or do units on the dam CPU but don't do both Ap's are done differently than MB's he can get away with using all the CPU cores and do units on the GPU without any problems |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14653 Credit: 200,643,578 RAC: 874 |
Gents, I fear we may have headed off in a slightly wrong direction - Ulrich states in his post that it is the CPU that is doing strange things and not the GPUs which appear to be behaving themselves pretty much as one would expect for a heterogeneous pairing of cards. The CPUs on the hand are the one's producing the very extended run times, and are running in EDF mode. Indeed. Please excuse - I had a very tiring 36-hours excursion to Edinburgh, four hours minimum by train each way and some intensive meetings while I was there. At eleven years older than Glenn, that takes it out of you. The log Ulrich posted yielded interesting clues - that task result 14/11/2015 13:37:21 | SETI@home | [cpu_sched_debug] scheduling ap_20oc11aa_B3_P0_00232_20151031_30892.wu_0 (high priority) - was still visible when I first looked this morning, but has been purged while we've been distracted by irrelevencies of GPU bus design. The most important feature I noticed was that CPU time recorded was only around 60% of the elapsed time. That's not good for a CPU task, and is evidence of the CPU thrashing that has been discussed. Now, if you'll excuse me, I'll go and have a lie down, and then some food. When I come back, we can perhaps discuss the triggers for EDF in a version of the English language (less idiomatic) which is perhaps more accessible for a questioner whose native language is almost certainly not even grammatically-correct English. |
rob smith Send message Joined: 7 Mar 03 Posts: 22223 Credit: 416,307,556 RAC: 380 |
Enjoy your well deserved kip. Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
©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.