| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Beta Testing
Forum: Beta Test Support Forum
Thread: Lead time for GPU beta |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 88
|
|
| Author |
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges: 
|
Don't forget to confirm you've got the OpenCL part installed. The GPU-Z tool will confirm, bottom left. Don't think that at this performance we'll ever going to see a CUDA specific GPU project at WCG. --//-- Thanks for the info Sek. I checked with GPU-Z. It tells me OpenCL is loaded version 1.1 and Cuda version 2.0. Is this OpenCL version ok?  |
||
|
|
KWSN - A Shrubbery
Master Cruncher Joined: Jan 8, 2006 Post Count: 1585 Status: Offline |
They're still debating that one. Most reports require OpenCL 1.2 but some are reporting success with 1.1. It may be a driver issue. Myself, I'm on Linux, so no work at all no matter the card.
---------------------------------------- Distributed computing volunteer since September 27, 2000 |
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges:
|
They're still debating that one. Most reports require OpenCL 1.2 but some are reporting success with 1.1. It may be a driver issue. Myself, I'm on Linux, so no work at all no matter the card. It is strange because the latest Nvidia Drivers under Windows 7 64bit have OpenCL 1.1. I understand this beta was Nvidia/Windows oriented. How to install OpenCL 1.2 on my GTX580 boards. No idea. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hypernova, the test was for both ATI & NVidia, limited for now to windows only [for the GPU part]. The CPU only tasks also went to other platforms. Had one on Linux, which practically got the same credit as the windows copy my client did, yet did it 41% faster... good parity.
----------------------------------------BETA_ X0000131280411201112141252_ 0-- 1854592 Valid 13-3-2012 20:34 13-3-2012 19:27 1,62 33,3 32,0 <win BETA_ X0000131280941201112141245_ 1-- 1767290 Valid 13-3-2012 20:34 13-3-2012 18:40 1,20 36,0 33,1 <lnx --//-- [Edit 1 times, last edit by Former Member at Mar 14, 2012 7:10:28 PM] |
||
|
|
sk..
Master Cruncher http://s17.rimg.info/ccb5d62bd3e856cc0d1df9b0ee2f7f6a.gif Joined: Mar 22, 2007 Post Count: 2324 Status: Offline Project Badges: 
|
All with the latest 296 series driver. These are all Cuda 2.0 or more, so they should be ok. There is a bug in the 295.x and 296.x drivers! Your NVidia cards have Compute Capability 2.0, not cuda2.0 ;)OpenCL 1.1 will work, assuming your dodgy drivers don’t mess things up, and they might! From my limited computer background, I get the impression that AMD/ATI is "better" as single FPU projects (basic numbers), while NVIDIA is more adept for double FPU projects, which are more complex. So for "basic" (not being used negatively) projects, yes AMD/ATI is better (PrimeGrid, Bitcoin are my two best examples.) NVidia is better for CUDA, saying as AMD/ATI doesn’t support cuda, and AMD/ATI is better at ATI Stream (APP), as it's not supported by NVidia. OpenCL is supported by both AMD and NVidia, on most newer cards. FP64 (double precision) is better on High end (only) AMD cards than high end desktop NVidia cards (GeForce cards). FP32 (single precision) performance varies from project to project. I expect the performance here will be comparable for comparable AMD/NVidia cards, but some cards will be sweeter than others (performance per Watt). |
||
|
|
Speedy51
Veteran Cruncher New Zealand Joined: Nov 4, 2005 Post Count: 1259 Status: Offline Project Badges: 
|
I have a NVIDIA GPU GeForce GTX 470 (driver version 26658, CUDA version 3020, compute capability 2.0, 1248MB, 1089 GFLOPS peak) with OpenCL 1.1 Hopefully I can use this card. Will Boinc version 6.10.xx/6.12.xx work or do we need beta 7.0.20 to get GPU tasks? (when available)
---------------------------------------- |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
WCG does not at this time require anyone to upgrade from 6.10.58 to support HCC GPU crunching. Key is:
At this time, the requirements to run the GPU version of the research application are as follows: 1) Has sufficient performance as determined by a small benchmark run at the start of a workunit. If the GPU fails the test, the application will exit and be reported as an error. You can check for this condition in the log file that is returned to the server by going to the Result Status page (https://secure.worldcommunitygrid.org/ms/viewBoincResults.do) and clicking on the link under 'Status' for the appropriate result. 2) It also must support the OpenCl extension cl_khr_local_int32_base_atomics. This is also checked at the beginning of the workunit and, if it fails, will be reported as an error and recorded in the log file. --//-- |
||
|
|
Speedy51
Veteran Cruncher New Zealand Joined: Nov 4, 2005 Post Count: 1259 Status: Offline Project Badges:
|
Thanks I'm all set to go.
---------------------------------------- |
||
|
|
|