| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Active Research
Forum: OpenPandemics - COVID-19 Project
Thread: Mandatory "Request for Project to be Ported to GPU" Thread |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 133
|
|
| Author |
|
|
BladeD
Ace Cruncher USA Joined: Nov 17, 2004 Post Count: 28976 Status: Offline Project Badges: 
|
" the Nvidia cards on CUDA always did the best." That's strange! Don't know about that, but it's not true. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
" the Nvidia cards on CUDA always did the best." That's strange! Only to people who don't have Nvidia cards. But by recollection OpenCl on NVidia does not as well as OpenCl of AMD/compatible cards. If the developers find the time to code both CUDA and OpenCl, great, but my eggs are in the OpenCl basket, one size fits most, no one feeling locked out, sort of. The list of blocked/incompatible cards for the HCC run was quite long. |
||
|
|
Speedy51
Veteran Cruncher New Zealand Joined: Nov 4, 2005 Post Count: 1326 Status: Offline Project Badges: 
|
" the Nvidia cards on CUDA always did the best." That's strange! Only to people who don't have Nvidia cards. But by recollection OpenCl on NVidia does not as well as OpenCl of AMD/compatible cards. If the developers find the time to code both CUDA and OpenCl, great, but my eggs are in the OpenCl basket, one size fits most, no one feeling locked out, sort of. The list of blocked/incompatible cards for the HCC run was quite long. I agree the list of blocked/incompatible cards was quite long. If I remember correctly on my GTX 470 tasks were complete in under 10 minutes. I am looking forward to the client when it arrives. I know this will be some time away  |
||
|
|
robertmiles
Senior Cruncher US Joined: Apr 16, 2008 Post Count: 445 Status: Offline Project Badges: 
|
" the Nvidia cards on CUDA always did the best." That's strange! The CUDA language is more closely linked to the GPU design for Nvidia-based graphics cards than languages such as OpenCL that are not linked to just one GPU company's GPU designs. Therefore, it is POSSIBLE to write programs in CUDA that will run faster on Nvidia GPUs than similar programs written in OpenCL. It's also possible to write such programs that will run slower, but very few BOINC projects want to do that. [Edit 1 times, last edit by robertmiles at Jun 26, 2020 3:14:51 AM] |
||
|
|
mgpointner
Advanced Cruncher Argentina Joined: Nov 16, 2009 Post Count: 55 Status: Offline Project Badges:
|
For information only…
----------------------------------------From https://www.biorxiv.org/content/10.1101/2020.06.27.175430v1.full.pdf To estimate the performance of Folding@home, we make the conservative assumption that each CPU core performs at 0.0127 TFLOPS and each GPU at 1.672 native TFLOPS (or 3.53 X86-equivalent TFLOPS), as explained in our long-standing performance estimate (https://stats.foldingathome.org/os). For reference, a GTX 980 (which was released in 2014) can achieve 5 native TFLOPS (or 10.56 X86-equivalent TFLOPS). An Intel Core i7 4770K (released in 2013) can achieve 0.046 TFLOPS/core. We report x86-equivalent FLOPS."  |
||
|
|
mdxi
Advanced Cruncher Joined: Dec 6, 2017 Post Count: 109 Status: Offline Project Badges: 
|
I'm not going anywhere near the debate on Nvidia vs. AMD. But this
----------------------------------------Most AMD cards have problems getting the drivers installed. hasn't been the case for a while now. AMDGPU drivers are part of the kernel and have been for 2+ years. As a concrete example, to get GPGPU compute working, on AMD, on Arch Linux, all you need to do is install two packages 'sudo pacman -S opencl-mesa ocl-icd' and reboot. AMD and Nvidia cards are equally easy to use on Linux these days.  |
||
|
|
TheCruelLogician
Cruncher Joined: Mar 5, 2019 Post Count: 10 Status: Offline Project Badges: 
|
I'm not going anywhere near the debate on Nvidia vs. AMD. But this Most AMD cards have problems getting the drivers installed. hasn't been the case for a while now. AMDGPU drivers are part of the kernel and have been for 2+ years. As a concrete example, to get GPGPU compute working, on AMD, on Arch Linux, all you need to do is install two packages 'sudo pacman -S opencl-mesa ocl-icd' and reboot. AMD and Nvidia cards are equally easy to use on Linux these days. well the problem is that a lot of the documentation is just plain wrong. I run Manjaro, and the methodology of sudo mwhd -a pci nonfree 0300 didn't do diddly. For anyone running Manjaro, if you are having problems, sudo mhwd -i pci video-nvidia-440xx, followed by sudo reboot will do the trick.  |
||
|
|
TheCruelLogician
Cruncher Joined: Mar 5, 2019 Post Count: 10 Status: Offline Project Badges:
|
For information only… From https://www.biorxiv.org/content/10.1101/2020.06.27.175430v1.full.pdf To estimate the performance of Folding@home, we make the conservative assumption that each CPU core performs at 0.0127 TFLOPS and each GPU at 1.672 native TFLOPS (or 3.53 X86-equivalent TFLOPS), as explained in our long-standing performance estimate (https://stats.foldingathome.org/os). For reference, a GTX 980 (which was released in 2014) can achieve 5 native TFLOPS (or 10.56 X86-equivalent TFLOPS). An Intel Core i7 4770K (released in 2013) can achieve 0.046 TFLOPS/core. We report x86-equivalent FLOPS." per physical or logical core? fwiw, my 9900KS LINPACKs at 275 GFLOP/s, (admittedly overclocked) and my 8700K at 145 (just benched the other day at stock clocks) |
||
|
|
[VENETO] boboviz
Senior Cruncher Joined: Aug 17, 2008 Post Count: 184 Status: Offline Project Badges: 
|
" the Nvidia cards on CUDA always did the best." That's strange! Only to people who don't have Nvidia cards. I forget emoticons for irony... It's OBVIOUS that a CLOSED sw born to work with SPECIFIC hw is faster than OPEN sw born to work with EVERYTHING. P.S. And it's not always true, sometimes OpenCl is better than Cuda. |
||
|
|
Jim1348
Veteran Cruncher USA Joined: Jul 13, 2009 Post Count: 1066 Status: Offline Project Badges:
|
It's OBVIOUS that a CLOSED sw born to work with SPECIFIC hw is faster than OPEN sw born to work with EVERYTHING. Only to people who do not have the experience. I have used both CUDA and OpenCl (with both Nvidia and AMD) on every project that has both. CUDA on Nvidia is faster. That is why Nvidia did it. If OpenCl were faster, they could have stayed with that and saved themselves the development money, for both the hardware and the software. I would have thought that was obvious, or should have been. |
||
|
|
|