| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: GPU Support Forum
Thread: Cuda 6 -> Easy GPU support for WCG ? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 15
|
|
| Author |
|
|
branjo
Master Cruncher Slovakia Joined: Jun 29, 2012 Post Count: 1892 Status: Offline Project Badges: 
|
Well Coleslaw, at least I posted the comparison of computation power of the two of the top GPGPU's. And I recalculated their FLOPS capabilities by TDP and by price. It shows that 7990 is by 100% better than Titan in Single Precision. For me, 10% is significant. 100% is interstellar gap. And I posted the comparison of the newest and most advanced NVIDIA vs. not the newest AMD GPGPU.
----------------------------------------What you did? Just wrote your own feelings. I would also caution making such statements because CUDA (according to everything I have read) is a lot easier to code for and is more adopted by academics. Says who? You and? See why, for example, Asteroids@Home started with CUDA: http://asteroidsathome.net/boinc/forum_thread.php?id=131&postid=2012#2012 Never mind. I feel with you.  Crunching@Home since January 13 2000. Shrubbing@Home since January 5 2006  [Edit 2 times, last edit by branjo at Apr 24, 2014 8:04:54 PM] |
||
|
|
RaymondFO
Veteran Cruncher USA Joined: Nov 30, 2004 Post Count: 561 Status: Offline Project Badges: 
|
Here is an interesting statement from Coursera (https://www.coursera.org/course/hetero ). This is found toward the bottom of the page under Q&A
Why not teach the whole course using OpenCL? While OpenCL is an industry standard and widely supported by many CPU and GPU vendors, it is much more complex and tedious to use than CUDA. The complexity and tedious details distract from the concepts and techniques that one should master. From our experience, it is much more productive to use CUDA to teach the concepts and techniques. We will then teach the additional complexities of OpenCL so that students can comfortably apply all the concepts to OpenCL. |
||
|
|
branjo
Master Cruncher Slovakia Joined: Jun 29, 2012 Post Count: 1892 Status: Offline Project Badges:
|
This one is also interesting:
----------------------------------------What is the coolest thing I'll learn if I take this class? You will learn how to unleash the massive computing power from mobile processors to supercomputers for your applications. Crunching@Home since January 13 2000. Shrubbing@Home since January 5 2006 |
||
|
|
KLiK
Master Cruncher Croatia Joined: Nov 13, 2006 Post Count: 3108 Status: Offline Project Badges: 
|
also, bear in mind that OpenCL is also supported for Intel HD GPUs on some desktops with built-in GPU on MBO...for example mine FUJITSU-SIEMENS ESPRIMO 710 has an Intel HD 2500, which crunches SETI@home-GPU units within 1 to 2h... ;)
---------------------------------------- |
||
|
|
Coleslaw
Veteran Cruncher USA Joined: Mar 29, 2007 Post Count: 1343 Status: Offline Project Badges: 
|
KLiK, I agree that OpenCL is very versatile. That wasn't the statement made. I don't argue its versatility. However, the question that followed was
----------------------------------------So, why would anyone (ETA:) sane port its application for CUDA? and the answer pretty much is obvious. Because you would have much better output from nVidia if you did. However, that second question does not fully justify the statement made that AMD on OpenCL is superior. That is subjective to opinion and too broad to claim as fact. I respect branjo's opinions on this, however he can't attack mine when he too is doing just that. branjo, how does the numbers you show up above reflect efficiency in the coding? An example of what I'm getting at is that it was recently stated that 32bit software runs more efficient and thus better on 64bit ARM processors. That is an example of how programming can make a big difference with different architecture. You can't compare AMD to nVidia using different programming like you are. You need hard numbers from real science. Not spec sheets or gaming comparisons. Edit: I think the biggest way to even begin to prove your point is to run the same work unit running on a fully optimized CUDA app and then run it on a fully optimized OpenCL app both on nVidia so that you can see the big difference. Then you could begin comparing nVidia to AMD and the use of OpenCL. Unfortunately, very few projects bother doing this. Another Edit: RaymondFO thank you for proving one of my points. Adoption to me is a big part of superiority. Not everyone will move on to the advanced courses to begin learning OpenCL and not all universities even go that far. So, more exposure out there to CUDA. That means the greater likelihood of someone actually coding for GPU's. That to me is a very strong selling point.     [Edit 3 times, last edit by Coleslaw at Apr 25, 2014 1:25:28 PM] |
||
|
|
|