Advanced search

Message boards : Graphics cards (GPUs) : GPU Needed?

Author Message
ashes999
Avatar
Send message
Joined: 28 May 10
Posts: 19
Credit: 3,135,753
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwat
Message 17374 - Posted: 28 May 2010 | 9:25:02 UTC
Last modified: 28 May 2010 | 9:27:27 UTC

This is going to sound kinda crazy, but, can I run this project if I don't currently have a GPU card in my box?

Edit: apparently not ^_^

Profile liveonc
Avatar
Send message
Joined: 1 Jan 10
Posts: 292
Credit: 41,567,650
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwat
Message 17381 - Posted: 28 May 2010 | 10:39:51 UTC - in response to Message 17374.
Last modified: 28 May 2010 | 10:40:33 UTC

[slight punn on] Only if UR Wonder Woman! ;-) [slight punn off]
The Multicore CPU thread has been dead for 122 days http://www.gpugrid.net/forum_forum.php?id=3
____________

Snow Crash
Send message
Joined: 4 Apr 09
Posts: 450
Credit: 539,316,349
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 17382 - Posted: 28 May 2010 | 10:41:52 UTC - in response to Message 17374.
Last modified: 28 May 2010 | 10:42:49 UTC

Unfortunately no.
The tests with CPU work here on GPUGrid proved that the science is so complex that a CPU does not produce usable results in a reasonable amount of time so they have stopped developing in that direction.
If you are not already familiar with World Community Grid, I beieve they are perhaps the best CPU crunching project not only for stability but also for the work the are doing (multiple different disease related subprojects)
http://www.worldcommunitygrid.org
____________
Thanks - Steve

ashes999
Avatar
Send message
Joined: 28 May 10
Posts: 19
Credit: 3,135,753
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwat
Message 17393 - Posted: 28 May 2010 | 14:29:07 UTC - in response to Message 17382.

Yep, already figured that out. Next question: if I have two processers (not that GPU Grid uses even one), and I put this and another CPU-heavy project on 100% resource share each (equal resource share), both will run at the same time right? Or will BOINC timeshare? :(

There's really no point to timeshare, since they use different crunching resources. I want to figure this out before I buy my GPU card.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 17394 - Posted: 28 May 2010 | 14:55:03 UTC - in response to Message 17393.

Next question: if I have two processers (not that GPU Grid uses even one), and I put this and another CPU-heavy project on 100% resource share each (equal resource share), both will run at the same time right? Or will BOINC timeshare? :(

There's really no point to timeshare, since they use different crunching resources. I want to figure this out before I buy my GPU card.

Each core of your CPU(s) can run a project (or WU) while each NVidia GPU runs an instance of GPUGRID. An exception is AQUA for the CPU which will use all CPU cores on the current WU. GPUGRID will still run on the GPU.

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 17404 - Posted: 28 May 2010 | 17:51:04 UTC - in response to Message 17393.

Yep, already figured that out. Next question: if I have two processers (not that GPU Grid uses even one), and I put this and another CPU-heavy project on 100% resource share each (equal resource share), both will run at the same time right? Or will BOINC timeshare? :(

There's really no point to timeshare, since they use different crunching resources. I want to figure this out before I buy my GPU card.

GPU projects for the most part have minimal load on the CPU side (Einstein is one exception) so that running GPU Grid, Collatz, MW, etc, have minimal to no impact on running projects like WCG. If you run WCG at RS 100 and GPU Grid at RS 100 both should run full time (work being available of course), if you run Rosetta, WCG, and GPU Grid at the same RS then you will spend about half the CPU time doing RaH tasks and half doing WCG tasks ... OVER TIME ... that means that you can and **WILL** have instances where two RaH tasks run or two WCG tasks run at the same time ... add in more projects and it gets more interesting ...

I am at the far end of the spectrum in that I attach to pretty much every project and donate time to them ... sometimes I have used unbalanced RS to allocate the time based on how much I like project A over B and at other times have done other things ...

Last point, the actual RS numbers themselves have no meaning ... it is the ratio of the numbers and the number of projects that have that RS number that is meaningful ... so, if I am allocating based on interest I would set the most interesting projects to 100, less interesting to 50, 25, or 10... but, even 10 RS projects on fast and wide (8 core or more) systems have a chance to be run once a day... Though not properly shown in BOINC Manager, the RS is actually split along resource types (ATI, CUDA, CPU, etc.) so ... in my system with an ATI card and a CUDA card it can get interesting ... :)

ashes999
Avatar
Send message
Joined: 28 May 10
Posts: 19
Credit: 3,135,753
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwat
Message 17411 - Posted: 28 May 2010 | 21:49:56 UTC
Last modified: 28 May 2010 | 22:02:03 UTC

So if I understand correctly, then:

1) four CPU cores + 1 GPU core = five tasks being executed in parallel (assuming 100% CPU availability, and connection to WCG and CPUGRID)
2) I should put RS to 100 for both to achieve #1

Is that right?

[Edit] read those posts again and it looks right. Cheers dudes.

Now, can anyone recommend a low-end (on the cheap, $50 or less) GPU card to wet my feet? :D

Profile liveonc
Avatar
Send message
Joined: 1 Jan 10
Posts: 292
Credit: 41,567,650
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwat
Message 17414 - Posted: 28 May 2010 | 22:40:32 UTC - in response to Message 17411.

Mostly I've read that the GT240 with DDR5 is both cheap, low power, & popular. Read many writing that many Online stores sell it with rebates.
____________

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 17431 - Posted: 29 May 2010 | 13:49:30 UTC - in response to Message 17411.

So if I understand correctly, then:

1) four CPU cores + 1 GPU core = five tasks being executed in parallel (assuming 100% CPU availability, and connection to WCG and CPUGRID)
2) I should put RS to 100 for both to achieve #1

Is that right?

[Edit] read those posts again and it looks right. Cheers dudes.

Now, can anyone recommend a low-end (on the cheap, $50 or less) GPU card to wet my feet? :D

That would be correct... 5 tasks, one on the GPU and 4 on the CPU ... With only one CPU project and one GPU project the machine would, in theory, always be working on the tasks from the appropriate projects. You *MAY* want to consider setting up a back-up project for each "side" so that the system will not go idle ... but you may also want to leave that for another day ... when the feet are a little more wet. WCG is a good project in that they do have a fantastic up-time and almost always have work of one sort or another on a variety of sub-projects ... the only way they suck from a user's perspective is that they are one of the lower paying projects ... if that becomes of interest ...

As to RS, with just WCG and GPU Grid (or the other projects as GPU only selections (Note: for Collatz you would have to change the on-lline settings to not send you any CPU side work as I have done)) the RS can be left to 100 as it will not matter. If you add additional projects on either "side" you will have to decide on what you want ... if you connect to WCG, GPU Grid and Rosetta at RS 100 the GPU will be full time on GPU Grid and the CPUs will run about 50/50 WCG and Rosetta tasks ...

I think they may still be a bit pricy for you but the 9800GT is a fair card though in my opinion barely adequate for GPU Grid (as would be the GT240)... GPU Grid is not that "friendly" to the lower end cards because of the heft of the computation task... my personal recommendation would be to consider Collatz to start with a low end card (or DNETC)... if you are too low end of a card you will not be able to consider MW due to the need for double precision ...

ashes999
Avatar
Send message
Joined: 28 May 10
Posts: 19
Credit: 3,135,753
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwatwat
Message 17437 - Posted: 29 May 2010 | 18:18:04 UTC - in response to Message 17431.

Awesome! Thanks for all the help. I placed my order last night (!) for the GT240. It rang in at just a couple bucks over $100 after rebate.

To be honest, WCG has been amazing for me, and I'll stick to GPUGRID -- because these are the projects that interest me :)

Looking forward to getting it in two weeks!!

darwincollins
Send message
Joined: 6 Jan 10
Posts: 22
Credit: 105,944,936
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 17448 - Posted: 30 May 2010 | 3:34:46 UTC - in response to Message 17437.
Last modified: 30 May 2010 | 3:35:48 UTC

I think that the earlier posts are excellent and explained it quite well.

I thought that I should share since I have a machine (win7 64bit, gtx480, i7 930) for running gpugrid and wcg for about a week or 2 now. (well, its running gpugrid/wcg when my kid is not playing games)

Boinc - Preferences settings: (as suggested by other users on this forum)
checkmark - While computer is in use
checkmark - Use GPU while computer is in use
On multiprocessor systems, use at most 99% of the processors
Use at most 100% cpu time

So, basically, it ends up that 7 of the threads are running wcg, and 1 thread is running gpugrid. Task manager is flatlined at 100% (for all threads).

imho... I would not suggest adding more than 2 projects to a box, since it gets interesting on how boinc schedules projects/etc.

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 17456 - Posted: 30 May 2010 | 10:39:41 UTC - in response to Message 17448.

imho... I would not suggest adding more than 2 projects to a box, since it gets interesting on how boinc schedules projects/etc.


2 CPU projects is fine. It may not be optimal all the time (as Paul would argue) but it usually gets the job done without wierd stuff happening. And you get the benefit of not running dry if one project fails.

MrS
____________
Scanning for our furry friends since Jan 2002

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 17460 - Posted: 30 May 2010 | 14:38:49 UTC - in response to Message 17456.

imho... I would not suggest adding more than 2 projects to a box, since it gets interesting on how boinc schedules projects/etc.


2 CPU projects is fine. It may not be optimal all the time (as Paul would argue) but it usually gets the job done without wierd stuff happening. And you get the benefit of not running dry if one project fails.

Running only two project not optimal? Or BOINC not running optimally?

I have no complaint about the first thought ... in fact I recommend at least two projects... Those that are super-adicted to SaH for instance these last couple of weeks have been having a really tough go of it as the project is having one of its rough patches... that is why I suggested WCG with an alternative of Rosetta as they are both projects that concentrate on the disease area and both have decent up-times and reliable flows of work ...

The second way of reading your assertion is correct in that I do argue that after 5 years of development BOINC does not serve the user communities well, not the single project fanatics, nor the reasonable 5-10 project group, and least of all the oddballs that run them all (only about 3,000 of us)... what makes this so astounding is that Dr. Anderson wrote a paper on this aspect, but his actions are of those of a man that has never read the very paper he wrote ... very strange ...

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 17468 - Posted: 30 May 2010 | 18:14:47 UTC - in response to Message 17460.

Didn't have anything specific in mind, just didn't want to sound overly enthusiastic and leave the debate to others ;)

MrS
____________
Scanning for our furry friends since Jan 2002

darwincollins
Send message
Joined: 6 Jan 10
Posts: 22
Credit: 105,944,936
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 17484 - Posted: 1 Jun 2010 | 4:31:30 UTC - in response to Message 17460.

I have seen times where work was 'Ready to start' but cores were idle.

The greatest frequency of my observation was on a 8core (8thread) box with wcg, aqua, and rosetta. Normally, it would have 8 'running' (with only or a mix of wcg/rosetta), except when Aqua wanted all 8cores.

It was not rare for me to see a few cores idle when there was plenty of wcg or rosetta work 'ready to start'.

However, on another box, I have only the wcg project. I have not seen this 'idle core' behavior.

so, hence, why I suggested to not have more than 2 projects.

Alain Maes
Send message
Joined: 8 Sep 08
Posts: 63
Credit: 1,437,484,959
RAC: 85,170
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 17486 - Posted: 1 Jun 2010 | 5:33:06 UTC - in response to Message 17484.

The effect you describe here is in fact a side effect, a bug, typical for multi CPU projects like AQUA. In fact you should report it on the AQUA site since it was supposed to have been taking care of,

With single CPU projects this phenomenon of idle cores never occurs unless of course the number of active cores/threads has intentionally been limited for some reason.

Kind regards

Alain

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 17487 - Posted: 1 Jun 2010 | 7:53:46 UTC - in response to Message 17486.

Continuing that thought, this is also partly a design decision of UCB wherein the "desire" of the multi-CPU project for all available resources will override the rule of no idle resources. Until all the running applications have check-pointed more and more resources will be idled until the MP task can be started.

As Alain noted, this issue was supposed to be addressed to make the effect minimal so, yes, reporting it on Aqua and the development boards is indicated if it is still occurring with the later generations of the BOINC and Aqua applications ...

darwincollins
Send message
Joined: 6 Jan 10
Posts: 22
Credit: 105,944,936
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 17489 - Posted: 1 Jun 2010 | 13:41:59 UTC - in response to Message 17487.

Ok, I will post on Aqua. Thanks

I am currently looking at the boinc manager on a box not mentioned above. On this box, it has 4 cores/threads: running wcg, rosetta and aqua. (the video card is only a 8600gt so why no gpugrid).

rosetta: 8 ready to start, 1 running, 4 ready to report
wcg: 7 ready to start, 1 running, 1 waiting to run
aqua: 1 waiting to run
so, 2 cores idle.

I had already checked that memory/disk storage is sufficient.

darwincollins
Send message
Joined: 6 Jan 10
Posts: 22
Credit: 105,944,936
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 17549 - Posted: 9 Jun 2010 | 5:01:41 UTC - in response to Message 17489.

minor update. since, installing the latest boinc, I have not observered that issue with idle cores.

Post to thread

Message boards : Graphics cards (GPUs) : GPU Needed?

//