| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: GPU World Community Grid Projects and Support |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 10
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Could you advise if it is possible to currently use your GPU for existing World Community Grid Projects, if so which projects support this, what version of Boinc Manager is required and what cards are supported?
I want to get an ATI Radeon HD 4770, which operates at upto 960gflops, would you see this type of performance? It has the same amount of transistors as my Q9550 but this only operates at 45gflops. In addition to answering these questions, links to useful sources would be appreciated. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
There are no current projects here at WCG that support GPU crunching.
There are MANY threads here on this very subject. |
||
|
|
twilyth
Master Cruncher US Joined: Mar 30, 2007 Post Count: 2130 Status: Offline Project Badges: 
|
WCG is working on a GPU client for one of the cancer projects (HCC I think) but there is no ETA on the app. From past experience, I would assume that it probably won't be in beta for at least several more months. After that, it will depend on how things go.
----------------------------------------As was pointed out in a related recent thread, video cards don't handle being stressed 24/7 as well as CPU's. They tend to run closer to their thermal envelope and produce a significant amount of heat. So even if they get the app working flawlessly, it will have to be tested extensively before unleashing on our members. Even then, I think you shoud have to explicitly opt-in to run the app since hardware damage is possible. It's not likely but it's the sort of thing where I would only want to see people with a good understanding of what is under the hood running it.   |
||
|
|
damir1978
Senior Cruncher Joined: Apr 16, 2007 Post Count: 397 Status: Offline Project Badges: 
|
Comparing your GPU's 960 GFlops with your CPU 45GFlops it is like comparing apples with oranges. The Projects for apples work with different software then Projects for oranges.
----------------------------------------And all that depends on the scientists that produce that Project. ---------------------------------------- [Edit 1 times, last edit by damir1978 at Oct 14, 2009 1:14:43 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Can you advise where these posts are?
Presuming then that the software takes advantage of the capabilites of a GPU wouldnt it perform approx 20 times better? You have mentioned heat but what happens with playstaion 3 GPUs on the folding at home project that differentiates them from a PC GPU? This highlights the increase in performance over a CPU. http://boinc.berkeley.edu/gpu.php |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I am all for GPU computing but there is no project on WCG yet that uses GPUs.
Can you advise where these posts are? There is a search box in the top right of this page.Presuming then that the software takes advantage of the capabilites of a GPU wouldnt it perform approx 20 times better? likely yes, but your basic premis is that the software takes advantage of the GPU. This does not happen by magic, is not easy, and there are some problems that simply are not suited to be solved by the limited types of instructions available to GPUs. It's like comparing a horse and a human being. The horse can run much faster but it can't climb a tree. One of the big misconceptions in general is that WCG is somehow responsible for the projects we are running on our PCs but they are not. While they may review everything for security, optimizations etc. but they are not creating these projects (check out the research pages to see who created each of the projects). WCG, as an oversimplification, only hosting the services that allow all of us to work together as a grid. but what happens with playstaion 3 GPUs on the folding at home project that differentiates them from a PC GPU? Again ... comparing two dissimilar things ... the software for Ps3 is different, maybe ask folding at home ?I think the Milkyway project allows you to use ATI cards (most GPU projects are only CUDA/Nvidia) maybe check them out and get your system up to speed and temperatures down so you will be ready when WCG does host a GPU project.  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Can you advise where these posts are? Not always true. Autodock (used by FAAH, HFCC and formerly IADS and DDDT and who knows what else in the future) performance on GPU:Presuming then that the software takes advantage of the capabilites of a GPU wouldnt it perform approx 20 times better? You have mentioned heat but what happens with playstaion 3 GPUs on the folding at home project that differentiates them from a PC GPU? This highlights the increase in performance over a CPU. http://boinc.berkeley.edu/gpu.php (credit to Sekerob for finding this) http://www.vpac.org/files/OptimizingAutodockwithCUDA.pdf The overhead of all CUDA IEC and TI memory transfers when running with 2037 atoms or less is greater than the processing time for executing both the IEC and TI kernels. This is due to the number of instructions executed by each thread, which is too few for the cost of the memory transfer overhead. Therefore, even with the maximum number of threads executing in AutoDock, the current CUDA IEC and TI functions do not accelerate the LGA and therefore AutoDock. And a NRW scientist commenting on GPU: https://secure.worldcommunitygrid.org/forums/wcg/printpost_post,239244 |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Thanks for the discussion here is the answer from WCG yesterday
Thank you for taking the time to contact us. We are continuing to review our research projects for a good candidate. As you may imagine, our primary goal is to launch new research projects, maintain existing research projects, which includes ensuring the results from our members are returned to the research organizations in a timely manner, and to ensure our members receive error free workunits to process. Although utilizing this technology will not happen in 2009, we will continue to research implementation of this technology in 2010. Thank you very much for donating your unused computer cycles to help humanity and your wonderful support of World Community Grid! World Community Grid Support Team --------------------------------------------------------- World Community Grid |
||
|
|
twilyth
Master Cruncher US Joined: Mar 30, 2007 Post Count: 2130 Status: Offline Project Badges:
|
Here's the quote from esoteric's link - I think it's worth posting since it highlights the fact that GPU's really are much faster, but the problem is the radically different architecture and issues relating to the most efficient method of coding for it.
----------------------------------------Re: What is stopping Nutritious Rice for the World form doing CUDA/GPU enabled builds ? I can't speak for what's happening on the wcg side of things, only from the point of view of the protein folding algorithm. I have looked into gpu acceleration of the process. It is not quite a simple as you might think. The calculations can indeed be sped up many orders of magnitude. However, with the current algorithm, the limiting factor appears to be the rate of communication between the cpu and gpu, which is not uncommon or unexpected. The process has already been optimized for cpus to minimize expensive calculations while not worrying so much about memory bandwidth. While it certainly is possible to get very significant speed increases with the current algorithm, the gpu firepower is better exploited by allowing the removal of the compromises in the methodology that needed to be made so that it runs as fast as it does on cpus. I can only imagine the headaches on the wcg side of it in making sure that any gpu-enabled project runs properly on any combination of gpus and cpus... BTW, Folding@Home supports ATI cards - my 4850 runs fine on it. An old nVidia 8400GS, not so much. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
|
||
|
|
|