| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: CPU and GPU usage |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 22
|
|
| Author |
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
To me, Dataman indicated that the client will tell how much F@H takes away from BOINC, and it will. For those observant they'll see more Elapsed in the 6.4+ BM and less progress per unit of time... which is why many started to run the BOINCTasks just to see both Elapsed and CPU time and an efficiency percent.
----------------------------------------
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
To me, Dataman indicated that the client will tell how much F@H takes away from BOINC, and it will. For those observant they'll see more Elapsed in the 6.4+ BM and less progress per unit of time... which is why many started to run the BOINCTasks just to see both Elapsed and CPU time and an efficiency percent. I read Dataman's message more along the lines of various BOINC-projects tasks shows-up needing "0.05 cpu, 1 gpu", or "0.1 cpu, 1 gpu", or for that matter as "8 cpu's" (for AQUA on an 8-way system). This info shows-up in the Manager atleast under executing. Besides, when it comes to FAH's Ati-client, a quick look on Task Manager will show the cpu-usage of the FahCore_11-process...  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
So we can see it in BOINC just as Dataman and you said you too.
----------------------------------------My favorite dance is the shake, rattle and roll .
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Dataman
Ace Cruncher Joined: Nov 16, 2004 Post Count: 4865 Status: Offline Project Badges: 
|
I don't know what F@H is as I don't crunch it. Easy to tell though, just crunch a wu and BOINC will tell you the usage. Folding@Home doesn't use BOINC, so BOINC won't tell you this... Ingleside is correct it does not use BOINC. As I said, I don't crunch it. Any BOINC GPU project will show it in the Status column of the Tasks tab. EDIT: I've never looked this up but "I think" this is % of a core NOT % of a thread. Not sure though. If you have a compatible card, Milkyway is the best beginner's GPU project. It is very benign using only about 80% GPU and 0.05 CPU. Also their wu's run on either a CPU or GPU so you can easily see the difference in speed. For example on a GTX285, a wu take ~11 minutes vs. ~6 hours on a i920 thread. I hope this helps.    [Edit 1 times, last edit by Dataman at Sep 1, 2010 2:57:26 PM] |
||
|
|
Dataman
Ace Cruncher Joined: Nov 16, 2004 Post Count: 4865 Status: Offline Project Badges:
|
or for that matter as "8 cpu's" (for AQUA on an 8-way system). This info shows-up in the Manager atleast under executing. IMO, AQUA is not for a beginner. As Ingleside said, I uses all threads to run 1 wu and is (was?) very greedy. When I crunched it, BOINC had difficulties managing LTD for AQUA and, if running multiple projects, AQUA got way more than it's share of time. There have been several BOINC releases since I last ran it so this may have been corrected. GPUGrid, in my opinion, has good science but is nVidia only. It drives your GPU's at 100% and takes 0.50 CPU/GPU so heat can be a problem if you are not well ventilated especially if you have 2 or 3 GPU's/machine. As I said, each GPU project has it's own peculiarities.  [Edit 1 times, last edit by Dataman at Sep 1, 2010 3:20:17 PM] |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
I don't know what F@H is as I don't crunch it. Easy to tell though, just crunch a wu and BOINC will tell you the usage. Folding@Home doesn't use BOINC, so BOINC won't tell you this... Ingleside is correct it does not use BOINC. As I said, I don't crunch it. Any BOINC GPU project will show it in the Status column of the Tasks tab. EDIT: I've never looked this up but "I think" this is % of a core NOT % of a thread. Not sure though. If you have a compatible card, Milkyway is the best beginner's GPU project. It is very benign using only about 80% GPU and 0.05 CPU. Also their wu's run on either a CPU or GPU so you can easily see the difference in speed. For example on a GTX285, a wu take ~11 minutes vs. ~6 hours on a i920 thread. I hope this helps. OK. So it's not obvious to the untrained eye and again underlining the "I'm BOINC, resistance is futile" regardless if F@H was there before in the offerings (FAICR). Does BOINC back off if F@H is running from trying to grab the card? From a discussion on Berkeley forums it does seem to cause a frictious condition, so it looks like the permission of specific cards as are allowed to be set in the cc_config.xml is a probable requirement to have these clients coexist. Holding my breath when WCG launches in. Probably very low weight and very opt-in only.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Dataman
Ace Cruncher Joined: Nov 16, 2004 Post Count: 4865 Status: Offline Project Badges:
|
OK. So it's not obvious to the untrained eye and again underlining the "I'm BOINC, resistance is futile" regardless if F@H was there before in the offerings (FAICR). Does BOINC back off if F@H is running from trying to grab the card? From a discussion on Berkeley forums it does seem to cause a frictious condition, so it looks like the permission of specific cards as are allowed to be set in the cc_config.xml is a probable requirement to have these clients coexist. Holding my breath when WCG launches in. Probably very low weight and very opt-in only. From what little I know they coexist but are mutually exclusive on the CPU side. I don't know if or how they would handle demands for a GPU. Interesting question though. Everything I know is from the point of view of a BOINC/Windows/nVidia/CUDA "bigot".  Yes WCG will have to careful entering this technology and it MUST be opt-in. Otherwise it will be a disaster. To crunch CPU at WCG, you don't have to know anything about a computer other than how to install an application. To crunch GPU, the BOINC release, the nVidia driver and the CUDA .dll's are always changing and not always compatible. However it is much, much better than when I started GPU crunching a few years ago. If you run multiple GPU's/machine it gets even more interesting with "dummy plugs", SLI, heat, air flow, power supplies, etc. Not to mention cost. I am trying to learn Fermi technology now and my forehead is bloody from beating my head on my workbench.  However, setting aside the cussing and fussing about credits, GPU's can scream though wu's at hundreds of times the speed of CPU's. |
||
|
|
sk..
Master Cruncher http://s17.rimg.info/ccb5d62bd3e856cc0d1df9b0ee2f7f6a.gif Joined: Mar 22, 2007 Post Count: 2324 Status: Offline Project Badges: 
|
F@H is not a Boinc Project.
There are separate CPU and GPU applications for F@H. The GPU application is not too CPU intensive, unless you have both apps running, or you have several big GPU’s (Fermi’s) and a weak CPU. It has very high GPU utilization, so you will need adequate GPU cooling. The easiest way to see how much CPU is being used is to suspend Boinc and look at Task Manager. If you want to look at the GPU usage use GPU-Z. I use to have an HD4850 hooked up, as they can use ATI cards too. Now I occasionally use it as a GPU backup project on some cards (Fermi down to GT240) - never had any issues when it came to crunching Boinc CPU tasks and F@H. In my mind it certainly beats counting pi on your GPU. |
||
|
|
Jim1348
Veteran Cruncher USA Joined: Jul 13, 2009 Post Count: 1066 Status: Offline Project Badges: 
|
Will using the GPU to run F@H slow down the crunch times for the CPU running BOINC or will it just take more power? I have a Radeon 4670 doing nada at the moment. My own experience with Nvidia GT240s (I run 4 of them dedicated to F@H) is that Task Manager shows less than 1 percent CPU utilization (typically 0 percent) for each card with the latest drivers. I use Win7 64-bit with either dual-core or quad-cores at 3.0 and 2.5 GHz respectively. But I concur with the others that the ATI cards are another story. Back when I was using a Radeon 4350, it took several percent, though I don't remember the numbers. Your 4670 will suck up some CPU power no doubt. |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Experimenting on my laptop with 6.10.57 client. It has a GS GO 7400 NVidia and this started showing *after* clearing the local prefs unticking the Use GPU while computer is in use box.
----------------------------------------518 11-09-2010 11:22:07 Running CPU benchmarks 519 11-09-2010 11:22:07 Suspending computation - running CPU benchmarks 520 World Community Grid 11-09-2010 11:22:07 [cpu_sched] Preempting c4cw_target02_001613480_0 (left in memory) 521 World Community Grid 11-09-2010 11:22:07 [cpu_sched] Preempting c4cw_target02_001613079_0 (left in memory) 522 11-09-2010 11:22:39 FP benchmark ran only 1.140625 sec; ignoring 523 11-09-2010 11:22:39 [dcf] scaling all duration correction factors by 1.000000 524 11-09-2010 11:22:40 Resuming computation 525 World Community Grid 11-09-2010 11:22:40 [cpu_sched] Resuming c4cw_target02_001613480_0 526 World Community Grid 11-09-2010 11:22:40 [cpu_sched] Resuming c4cw_target02_001613079_0 527 World Community Grid 11-09-2010 11:23:30 [checkpoint_debug] result c4cw_target02_001613480_0 checkpointed 528 11-09-2010 11:24:32 [cpu_sched] resuming GPU activity 529 11-09-2010 11:35:50 [cpu_sched] suspending GPU activity 530 World Community Grid 11-09-2010 11:36:42 task c4cw_target02_001619454_0 resumed by user 531 World Community Grid 11-09-2010 11:37:11 [checkpoint_debug] result c4cw_target02_001613079_0 checkpointed 532 World Community Grid 11-09-2010 11:37:30 task c4cw_target02_001619359_0 resumed by user 533 World Community Grid 11-09-2010 11:38:15 task c4cw_target02_001618082_0 resumed by user 534 11-09-2010 11:39:27 [cpu_sched] resuming GPU activity 535 11-09-2010 11:39:39 [cpu_sched] suspending GPU activity Shortly I'll be putting the <no_gpus>1</no_gpus> line into the <options> section of the cc_config.xml to disable this BOINC 6.4+ feature, but thought of sharing in case members start seeing this when upgrading from 6.2.28 to 6.10.xx. Normally it is not visible unless the <cpu_sched> log flag is set, but the client seems to be going in and out, seemingly wasting cycles on this GPU checking process. update: after putting the <no_gpus>1</no_gpus> in and reading in the cc_config.xml the log printed: 563 11-09-2010 12:26:25 Config: don't use coprocessors. May explain seeing spurious "out of video memory... close application" kind of pop-ups. Now I'm going to watch run times on clean water if they stay same, improve, deteriorate. PS, yes the benchmark failed so it seemed to have abstained from adjusting the DCF... yes indeed... still showing as 2.02. edit2: Well seems that only by setting the GPU option while computer is in use to *on*, will it stop the GPU on/off scheduling. 578 11-09-2010 12:54:01 [cpu_sched] resuming GPU activity 579 11-09-2010 12:54:36 [cpu_sched] suspending GPU activity 580 World Community Grid 11-09-2010 12:58:43 [checkpoint_debug] result c4cw_target02_001613480_0 checkpointed 581 11-09-2010 13:00:58 [cpu_sched] resuming GPU activity 582 11-09-2010 13:01:40 [cpu_sched] suspending GPU activity oink
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 2 times, last edit by Sekerob at Sep 11, 2010 11:09:37 AM] |
||
|
|
|