| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: GPU Support Forum
Thread: What is stopping IBM's World Community Grid from doing CUDA/GPU enabled builds? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 31
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Just think about all the projects you have, and the resources that could be put to use both on IBM's and other partner's side, as well as the contributed processing power!
I just asked the "Nutritious Rice for the World", when they will do CUDA/GPU builds: http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=26274#238935 - When will WCG do CUDA/GPU infrastructure/builds? ...It does seems that WCG are the ones we are waiting for... "And WCG has said that they want to start the feasibility/approval steps in the second half of 2009, i.e. in the coming months now." Source: http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=26175 No matter what the case is, how do we proceed from here? That's the issue. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
We're still working on it although it's not super high-priority for us. We were close to having a plan in place for going only with nVidia but now we are also looking at OpenCL. The only project that is feasible to run CUDA/GPU is Help Conquer Cancer.
|
||
|
|
nasher
Veteran Cruncher USA Joined: Dec 2, 2005 Post Count: 1423 Status: Offline Project Badges: 
|
thank you for your response...
----------------------------------------personaly i would be happy with any project here running CUDA... I understand that it will not happen till its ready and i am glad for that again thank you for your response  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
CUDA thread # 453,321.
Same answer: Not high priority. Not easy to do. Minimal demand by scientists. Not soon, maybe later, maybe never. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The only project that is feasible to run CUDA/GPU is Help Conquer Cancer. Why? I thought Autodock could run on CUDA. |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
An appetizer: http://www.vpac.org/files/OptimizingAutodockwithCUDA.pdf
----------------------------------------added: last paragraph of the conclusion: The restrictions currently made by CUDA impact upon the ease of porting general-purpose graphics processing unit (GPGPU4) computations to CUDA. The term GPGPU implies that CUDA is able to perform more than specific graphic computations. However, this does not mean CUDA can perform all general computations. For example, AutoDock contains computations that rely on dynamic memory allocation. While it is possible to program these in CUDA, it is very difficult and timeconsuming for the programmer. It is important to consider the extra time and resources that may be required before proceeding with such a task.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Jul 22, 2009 1:39:35 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
and WCG does not have the resources to do 'difficult and timeconsuming' unless it is the TOP prioirty and as long as CUDA is 'difficult and timeconsuming' it will never be TOP unless we run out of projects (which always come first).
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
An appetizer: http://www.vpac.org/files/OptimizingAutodockwithCUDA.pdf added: last paragraph of the conclusion: The restrictions currently made by CUDA impact upon the ease of porting general-purpose graphics processing unit (GPGPU4) computations to CUDA. The term GPGPU implies that CUDA is able to perform more than specific graphic computations. However, this does not mean CUDA can perform all general computations. For example, AutoDock contains computations that rely on dynamic memory allocation. While it is possible to program these in CUDA, it is very difficult and timeconsuming for the programmer. It is important to consider the extra time and resources that may be required before proceeding with such a task. Thanks Sek!! |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges: 
|
and WCG does not have the resources to do 'difficult and timeconsuming' unless it is the TOP prioirty and as long as CUDA is 'difficult and timeconsuming' it will never be TOP unless we run out of projects (which always come first). Once more, application programming is not done by WCG but on the Research side (either the lab writes its own application or it chooses an existing one which suits its needs). So, if there is no CUDA enabled application to run WCG cannot do anything. What WCG is involved with is making or asking for changes necessary for being able to manage the distribution of the workload. Also it is or would be involved in making the changes to Boinc for dealing with the variety of configurations on the member side, machines without GPU capability, machines with GPU capability but not wanting to use it, machines able and wanting to use GPU, obviously with all the various flavors of "not when I do this", "not at this time of the day", "just a little because it slows my display", etc... WCG is currently considering going on this not so easy new direction, but it is only at the very early stage of this process, i.e. building the necessary business case for getting approval from its sponsor. In any case, if there is nothing operational on the Research side there will not be anything in WCG, whether WCG is ready or not. Cheers. Jean. |
||
|
|
Rickjb
Veteran Cruncher Australia Joined: Sep 17, 2006 Post Count: 666 Status: Offline Project Badges: 
|
I'm not against any idea that increases the amount of computing power available to WCG, but ...
----------------------------------------1. What proportion of WCG's crunchers would be able or want to buy and run the up-market video cards that are GPGPU-capable? Looking on Sekerob's fabulous tapeworms , the averages of credits per hour for the fleet (12-14) suggest to me that the vast majority of computers are not high-end machines, and would not be available for GPGPU projects. Remember that only a small proportion of WCG contributors read these forums. I suspect that the proportion of very keen computer enthusiasts is higher at the grid computing projects that run GPGPU apps, than here at WCG. 2. Most science/engineering calculations are performed in 64-bit ("double-precision") floating point, rather than 32-bit; most multimedia apps only require 32-bit, which is the native high-speed mode of existing GPU cards. Sure, many of these GPUs can be made to perform double-precision calcs, but I read a few months ago that this slows them down so much that they lose most of their speed advantage, and end up only about 2 x faster than a CPU. This may of course be fixed by now or in the future. 3. If there is a boost in speed with GPGPU, it's not free. Current video cards use around 250 watts or more when crunching - see Client Folding GPU2 : les Radeon HD peu à l'honneur 4. The hardware and software involved with GPGPU is not mature. As an example, the first GPGPU project at folding@home has finished, and the GPU hardware used with it will not run the 2nd GPGPU project. How are the manufacturers of GPGPU cards travelling? Are they financially healthy? How's AMD/ATI's company balance sheet looking these days? Does someone want to commit programming resources to hardware that could be orphaned? 5. Following on from (4), there's a 3rd horse in this GPGPU technology race that no-one here seems to have mentioned, and I wouldn't be committing software resources to a CPGPU project until we can really see its capabilities. It looks like it will be more suitable for science computing than existing GPUs, and probably easier to program. It will be made by a little business that you may have heard of, that's struggling along in the GFC to not make record profits, even though it's spending $billions on R&D and new manufacturing facilities. The product is to be called Larrabee [Edit 2 times, last edit by Rickjb at Jul 23, 2009 5:16:05 AM] |
||
|
|
|