| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: GPU Support Forum
Thread: Any New GPU projects planned guys? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 43
|
|
| Author |
|
|
Falconet
Master Cruncher Portugal Joined: Mar 9, 2009 Post Count: 3315 Status: Offline Project Badges: 
|
Those not wanting to bet on a one horse race, here a 2016 Phoronix article on NVidia Cuda v OpenCL. http://www.phoronix.com/scan.php?page=article...16-first-clcuda&num=1 The sum is simple... if a CUDA project would come to WCG, a very large portion of the volunteers gets cut out. If OpenCL, yes NVidia might not be up to par, but pretty much all modern cards could participate from both camps [And maybe even the Intel/AMD APUs]. +1 Nvidia OpenCl support has gotten better since 2011 but why they still lag behind AMD support is a mystery. I read somewhere a while ago that it was intentional to protect their CUDA app. Multi-platform is the future IMHO so it really makes no sense to me for Nvidia to continue to not fully support OpenCl. There will probably always be CUDA apps that can't be ported to OpenCl. JMHO well, SETi@home just launched OpenCL version of SaH & SoG science...compared to CUDA jobs, it's using a whole core instead of 5-20% of it!  so would like to see a CUDA app for nVidia cards...while OpenCL can be made for intel HD or ATi's... That is double the effort for scientists who don't have the manpower or resources.  - AMD Ryzen 5 1600AF 6C/12T 3.2 GHz - 85W - AMD Ryzen 5 2500U 4C/8T 2.0 GHz - 28W - AMD Ryzen 7 7730U 8C/16T 3.0 GHz [Edit 1 times, last edit by Falconet at Jun 3, 2016 11:04:43 AM] |
||
|
|
KLiK
Master Cruncher Croatia Joined: Nov 13, 2006 Post Count: 3108 Status: Offline Project Badges: 
|
Those not wanting to bet on a one horse race, here a 2016 Phoronix article on NVidia Cuda v OpenCL. http://www.phoronix.com/scan.php?page=article...16-first-clcuda&num=1 The sum is simple... if a CUDA project would come to WCG, a very large portion of the volunteers gets cut out. If OpenCL, yes NVidia might not be up to par, but pretty much all modern cards could participate from both camps [And maybe even the Intel/AMD APUs]. +1 Nvidia OpenCl support has gotten better since 2011 but why they still lag behind AMD support is a mystery. I read somewhere a while ago that it was intentional to protect their CUDA app. Multi-platform is the future IMHO so it really makes no sense to me for Nvidia to continue to not fully support OpenCl. There will probably always be CUDA apps that can't be ported to OpenCl. JMHO well, SETi@home just launched OpenCL version of SaH & SoG science...compared to CUDA jobs, it's using a whole core instead of 5-20% of it! so would like to see a CUDA app for nVidia cards...while OpenCL can be made for intel HD or ATi's... That is double the effort for scientists who don't have the manpower or resources. that's why I put "would like to see"...not sure if it will happen, but I'd like that! ;) |
||
|
|
loreson
Cruncher Joined: Apr 26, 2014 Post Count: 4 Status: Offline Project Badges: 
|
The official statement just made by jhindo is: No AD Vina porting to GPGPU. did he state what the problem is? No good performance on GPU or not enough manpower? |
||
|
|
loreson
Cruncher Joined: Apr 26, 2014 Post Count: 4 Status: Offline Project Badges:
|
Zinc, have you ever seen any results from them (F@h)? Have they ever published anything worth seeing? TIA, b. In fact F@h produces lots of papers, but it seems like many of them are not disease related but focus on simulation technology. Here is an example for a cancer related paper: Activation pathway of Src kinase reveals inte...s targets for drug design |
||
|
|
Falconet
Master Cruncher Portugal Joined: Mar 9, 2009 Post Count: 3315 Status: Offline Project Badges:
|
The official statement just made by jhindo is: No AD Vina porting to GPGPU. did he state what the problem is? No good performance on GPU or not enough manpower? Vina just isn't suitable for GPU's. But even if it were, I assume manpower would be an issue as well. - AMD Ryzen 5 1600AF 6C/12T 3.2 GHz - 85W - AMD Ryzen 5 2500U 4C/8T 2.0 GHz - 28W - AMD Ryzen 7 7730U 8C/16T 3.0 GHz |
||
|
|
KLiK
Master Cruncher Croatia Joined: Nov 13, 2006 Post Count: 3108 Status: Offline Project Badges:
|
The official statement just made by jhindo is: No AD Vina porting to GPGPU. did he state what the problem is? No good performance on GPU or not enough manpower? Vina just isn't suitable for GPU's. But even if it were, I assume manpower would be an issue as well. & also, VINA is only run on 2 projects right now...of which 1 will be over withing a 6m (more or less)!  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The amount of folding new GPU's can output is quite nutty, and from what Nvidia is planning on releasing here soon after the 1090s is going to be worth the power they use. We just need the software to support it.
|
||
|
|
KLiK
Master Cruncher Croatia Joined: Nov 13, 2006 Post Count: 3108 Status: Offline Project Badges:
|
The amount of folding new GPU's can output is quite nutty, and from what Nvidia is planning on releasing here soon after the 1090s is going to be worth the power they use. We just need the software to support it. software is there...check some project around on BOINC! just no scientist on WCG is using it...it's their choice! |
||
|
|
SekeRob
Master Cruncher Joined: Jan 7, 2013 Post Count: 2741 Status: Offline |
So funny, a spat between Intel and NVIDIA ongoing, this quote with my emphasis on why no GPGPU projects... http://betanews.com/2016/08/19/nvidia-accuses...lating-benchmark-results/
Machine learning is currently a really big deal. It’s a huge market with untapped potential in many industry verticals, which is why a lot of different companies are trying to get in on the action. It is widely taken as a fact that GPUs are a better solution than CPUs when it comes to deep learning, because neural networks require low precision computation, and not high-precision, which is what CPUs are usually made for. So still, it is between low and high precision computing... what do our project scientists require?(Sitting back and see the house experts 'opine'... I have none :-)))) |
||
|
|
Promilus
Cruncher Joined: Apr 11, 2008 Post Count: 9 Status: Offline Project Badges: 
|
That's right, NV can double throughput with 16bit mode (so 2xFP16 @ 1 cuda core) while both intel and amd can only achieve the same throughput as fp32 (+needs two converts). That makes wide cpu-gpu gap even wider. On the other hand if calculations were done fixed point/integer then SSE can work even with 8bits outpacing even GPU (with Phi of course)
|
||
|
|
|