| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Suggestions / Feedback
Thread: GPU programming in volunteers hands? Technical specifications in a new forum here? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 30
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hello everyone,
I have seen that World Community Grid and GPU (usage of grafical cards processsors) has been the topic of many discussions. For reasons unknown to me, the topic seems to be stuck somehow. I would like to make the following proposition: Why not give a try and give the gpu programming to volunteers hands? The technical specifications could be in a seperate forum here, If the application is ready, it could be publishted as beta testing, just as some projects are beeing beta testing here at the beginning. Just to say it already at the beginnig: I personally will certainly be not able to help in this direction, I am not a programmer. And I regret this anyway. Missed the train somehow.....Well, different story. Think about it.please. Good luck. Crunching all the time. Martin Schnellinger |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Quite a few obstacles, one being use of proprietary ** science engines maybe with the exception of AutoDock and Rosetta i.e. the code could not be released and second, the volunteer programmer probably not capable of judging if the output is valid (An AutoDock-GPU paper posted on the forums springs to mind). Aside this, I think any programmer skilled to do this is on someone's payroll... it's a rather sizable commitment and then, it's just the tip of the iceberg of hurdles to get on the WCG grid. Great idea though, surely others having different views on this.
----------------------------------------** CHARMM, Q-CHEM, MAXDO, Chrystal Vision, Rosetta, AutoDock are the ones presently used on WCG Researches.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
sk..
Master Cruncher http://s17.rimg.info/ccb5d62bd3e856cc0d1df9b0ee2f7f6a.gif Joined: Mar 22, 2007 Post Count: 2324 Status: Offline Project Badges: 
|
An open alpha project perhaps?
After finding would be programmers you would need a project, then the code - in that order. You can't just write some general streaming, CUDA or OpenCL code and expect it to work generically for a variety of projects; GPUGrid, MW, Collatz, folding... all use different GPU applications. I would suggest any programmers that might tackle this (for here or anywhere else) look to a very basic project, with simple computation they can write themselves, rather than having to try to patch round CUDA bugs. So the question is what type of projects could use GPUs to tackle lots of simple, parallel, repeating calculations? Then the second question, is it already being done? There is also the inherent problem of which cards to support/exclude (include ATI or NVidia, or both; streaming, CUDA or OpenCL). For such a project to actually go anywhere, it would eventually need to be handed over, so would WCG ever embrace it? If it was similar enough in nature to the other projects (health related) possibly, else probably not. So you are probably limited to a basic genetics, DNA, Protein or similarity project that would aid medicine but is not already being done, and is not likely to be done elsewhere any time soon. GPUGrid deals with complex molecular structures and interactions, folding deals with protein folding and structure. GPUGrid uses NVidia cards and Folding uses both NVidia and ATI. The other GPU projects do non medicine related research. Anyway if WCG does not move towards GPUs, perhaps GPU's will eventually come to WCG; in the form of Knights Corner |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
In my opinion, there are 2 points to make about GPU projects.
#1) Most of our projects use humongous software packages a decade or so old maintained by a university group or a spin-off corporation. If the maintaining group rewrites it for a GPU, then great! However, we cannot expect a research group to spend man-years rewriting a program for the GPU. One of the programs used by CEP had a software module rewritten for GPU code. Unfortunately it was not a module being used by CEP. NRW is writing GPU code to analyze the results we have returned to them. But this GPU code will run on their own computers. They do not need to jump through all the hoops needed to run on WCG. HCC is still the only current project that might develop some GPU code. My personal opinion is that if they are not busily coding away for the GPU right now, then they would be better served working on the results rather than trying to get the results faster. My personal opinion is that it is likely to be some time before we see a GPU project, even though WCG is committed to moving to a version of BOINC that allows GPU usage. We would probably have BOINC 6.10 up and running on WCG already if there were a GPU project in sight. #2) All distributed computer projects have to worry about jokers returning invalid results for credit. One of the most powerful techniques for preventing this is secret code in the software package. Even if there is an open software version of a package that we need, WCG is likely to use a proprietary version instead. Unless you take a highly ideological view of open software, proprietary software allows WCG to embed some encrypted validation routines that make the results much easier to validate. This greatly reduces the scope for open software development. There may be cases where open software is the best way to go, but this is a serious consideration. Lawrence |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hello,
I do wish this can get rolling, there is the example GPU template project available at: http://boinc.berkeley.edu/trac/wiki/GPUApp There is template for OpenCL, CUDA and ATI Stream. Autodock and Autodock Vina are open source, so possibly at least open source BOINC docking would be possible. We have optimized Autodock and Autodock Vina for native 64 bit computing which is the first step of optimization, have a look at: http://autodock.scripps.edu/faqs-help/how-to/...4-bit-windows-application http://autodock.scripps.edu/faqs-help/how-to/...ed-up-autodock-vina-1-1.1 Lets start looking into the subject! CARE - Cancer and Aids Research Enabler |
||
|
|
TimAndHedy
Senior Cruncher Joined: Jan 27, 2009 Post Count: 267 Status: Offline Project Badges: 
|
All distributed computer projects have to worry about jokers returning invalid results for credit. One of the most powerful techniques for preventing this is secret code in the software package. I don't get it. All projects have to worry about unreliable systems returning bad results also. Why would the cross checks for that not detect cheating also. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I don't get it. All projects have to worry about unreliable systems returning bad results also. Why would the cross checks for that not detect cheating also. Because if you start the project on one computer, you know who started and finished it. In a DC environment, you have no idea who is running the work |
||
|
|
TimAndHedy
Senior Cruncher Joined: Jan 27, 2009 Post Count: 267 Status: Offline Project Badges:
|
Because if you start the project on one computer, you know who started and finished it. In a DC environment, you have no idea who is running the work Sorry, I don't understand the reply. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hello TimAndHedy,
Back in 2005 I read a lot of web posts about the history of distributed computing. I came across a site with some very disturbing stories about just how tricky some people were in developing code that would give them unearned credit. It really can be quite difficult to spot them. Even if they boast about their exploits, it can be difficult to figure out who is quietly running copies of the hacked programs, returning invalid results. This is not an absolute show stopper, but it has to be considered when you develop your validation routine and decide whether or not you can get away with only validating a percentage thoroughly and checking everything else with a quick but cursory check versus running extensive validation on everything. By 2006 I wished that I had recorded the URL where I found this historical information, but I do not remember it. Oh well . . . Lawrence |
||
|
|
TimAndHedy
Senior Cruncher Joined: Jan 27, 2009 Post Count: 267 Status: Offline Project Badges:
|
I would not worry about hacking much. It would not take much to detect cheating. Unreliable systems are potentially a much bigger problem. The checks for that should detect both.
It's certainly not a good argument for closed systems. Myself, I have wanted to look at the code for HCC to see why it is so much slower under Windows. |
||
|
|
|