| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Help Conquer Cancer
Thread: HCC with GPU |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 486
|
|
| Author |
|
|
TimAndHedy
Senior Cruncher Joined: Jan 27, 2009 Post Count: 267 Status: Offline Project Badges: 
|
There can be a difference in applications and their structure. That is the key. Not every class of problem is solved better by a GPU. It's not my area of expertise but, HCC looks to have to potential to be perfect for GPU. It may end up that nearly instant analysis of images is possible and we will be out of a job, but I'm OK with that. It all depends on the problem and the coder. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
But look at Folding@Home. They made a HUGE leap forward with Nvidia cards and I believe that folding/unfolding proteins (in simple words) is as complicated as what we are doing here. They invested plenty of time and money in developing GPU clients and now they see the effects.
I would say: GPU FTW! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
it has to do a lot with whether single or double precision maths are used. double precision currently is verry slow on gpus iirc
|
||
|
|
fablefox
Senior Cruncher Joined: May 31, 2010 Post Count: 168 Status: Offline Project Badges: 
|
Yeah, only Tesla capable of doing double precision point at the moment (AFAIK). Knowing that BOINC can check your hardware, maybe WCG team are looking at the percentage of people using Tesla based card to see if it's worth the effort. But if there is science that using single precision, maybe they should look into it.
----------------------------------------But hey, they are the scientest, not me, so they know things better :-) |
||
|
|
TimAndHedy
Senior Cruncher Joined: Jan 27, 2009 Post Count: 267 Status: Offline Project Badges:
|
I would expect them to push the boundaries on the capabilities and be thinking of the long term goal of no longer needing WCG for this type of analysis. Just a guess.
More important than single or double is thinking about the problem correctly to maximize the parallel nature of their solution. Proper solutions can increase the calculation speed by orders of magnitudes. We will have to wait to see what they come up with. |
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges:
|
Yeah, only Tesla capable of doing double precision point at the moment (AFAIK). I think the Fermi boards series 4XX and new 5XX are double precision capable. They cost much less than the Tesla. I am not so sure regarding the 2XX series. They work very well on GPUGrid but I do not know if double precision is needed. Anyway if HCC comes in GPU format I will be ready   [Edit 1 times, last edit by Hypernova at Nov 20, 2010 9:56:12 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I do not know any real details about the HCC programs, but I do know that standard video input can be handled by single-precision floating point. I do not know of any camera that provides even 16,384 by 16,384 frames so single-precision can easily map out the X,Y coordinates. The only point that requires a bit of caution is very small differences that might end up in the denominator, exploding tiny errors into large ones. Just a little thought usually prevents that.
Lawrence |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
Yeah, only Tesla capable of doing double precision point at the moment (AFAIK). FYI, Milkyway uses double-precision, and even some GPU-cards therefore can't run Milkyway, they're still the #1 GPU-project (by #active GPU's).  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
BladeD
Ace Cruncher USA Joined: Nov 17, 2004 Post Count: 28976 Status: Offline Project Badges:
|
it has to do a lot with whether single or double precision maths are used. double precision currently is verry slow on gpus iirc Oh. So why does a MW WU takes 12 hours on my CPU core and about 90 seconds on my GPU? |
||
|
|
fablefox
Senior Cruncher Joined: May 31, 2010 Post Count: 168 Status: Offline Project Badges:
|
BladeD, The moment you mention that I quickly went to join MW so that my CPU can work on WCG while my GPU can work on MW.
----------------------------------------And I got this message: 11/21/2010 5:35:01 PM Milkyway@home Message from server: Your NVIDIA GPU lacks the needed compute capability (1.3, required for double precision math I use GT 240. Oh well, stick with WCG for now then. But it shows that a lot (maybe most?) GPU project uses double precision, and most people doesn't have card that use it (me included :-( ) Oh, the other problem is the political side of thing. OpenCL not yet matured, and business politics is causing development he**. SETI@Home, AFAIK, is strictly CUDA, so that means it's a no-no for ATI user. See, knowing that only double persicion card can be supported is one thing (few user, IMHO), and with OpenCL still in development, only one implementation ca be used (NVIDIA or ATI). Selecting one will alienate the other. Does the work worth the hassle? I don't know as I don't have the staticstic. Maybe when WCG noticed that a significant number of user has double precision card capabilities, they will more towards that way. *edited to appropriate forum language - ErikaT ---------------------------------------- [Edit 1 times, last edit by ErikaT at Nov 22, 2010 12:53:13 PM] |
||
|
|
|