| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: GPU Support Forum
Thread: Graphics Card Performance |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 254
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I don't understand why GPU points are so different from GpuGrid points with Boinc or WCG. My 570's use almost no CPU with GpuGrid WU's yet return much greater resulting points. Granted it's Cuda, but CPU actual use is less.
----------------------------------------Edit: points return with GpuGrid is more than 10Xish WCGs [Edit 1 times, last edit by Former Member at Nov 11, 2012 6:55:13 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi shdbcamping,
I am puzzled myself. Do you mean GPU where you write CPU?  Lawrence |
||
|
|
mmstick
Senior Cruncher Joined: Aug 19, 2010 Post Count: 151 Status: Offline Project Badges: 
|
WCG HCC app is about 10% efficient from what I have seen. My GPU has 5 TFLOPS of peak throughput, other projects may better utilize this 5 Tflops and give me higher points, but HCC on average for me returns about 500 GFLOPS worth of points. I would mark this up to either better software with GPUGrid, or simply our GPU architecture is not as efficient in this project.
|
||
|
|
nanoprobe
Master Cruncher Classified Joined: Aug 29, 2008 Post Count: 2998 Status: Offline Project Badges: 
|
WCG HCC app is about 10% efficient from what I have seen. My GPU has 5 TFLOPS of peak throughput, other projects may better utilize this 5 Tflops and give me higher points, but HCC on average for me returns about 500 GFLOPS worth of points. I would mark this up to either better software with GPUGrid, or simply our GPU architecture is not as efficient in this project. You're comparing apples and oranges. GPUGrid uses CUDA which is specifically for the Nvidia architecture. That's why GPUGrid is an Nvidia only project. WCG uses OpenCl which can be run on either ATI or Nvidia.
In 1969 I took an oath to defend and protect the U S Constitution against all enemies, both foreign and Domestic. There was no expiration date.
  |
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges: 
|
I think the use of various modified configuration file for graphics cards is something like cheating .. Or then wcg cell cheating cheating us who are not using something like that ..? .... Cooperation between the creators and developers in wcg graphics card should be more. It is not an online computer game here, where you could consider modifications or additions of some files are cheating. There is one objective here that is to make the best use of the hardware we have to contribute to scientific research. If some clever users find a way to maximise the crunching performance and that you can crunch more work units over the same period of time than many thanks to them, their contribution brings real value as it speeds up results and projects for the scientists. If you see my previous posts you see that I do not talk about points but finished and validated WU / day. I have the feeling that you are interested just in points (but if I am wrong then forgive me and receive my apologies) and that you feel "cheated" because some users can get a massive number of points using clever GPU crunching. But it is only you that has this negative feeling. Forget it and think about the positive side of things. And if you are here to race for points only then you are also most welcome to "cheat" as much as you can to get the maximum number of points, as long as it is correlated with an increased number of terminated and validated WUs   |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi shdbcamping, I am puzzled myself. Do you mean GPU where you write CPU? Lawrence It seems that all HCC1 results recieve points based on cpu use. |
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges:
|
Hi shdbcamping, I am puzzled myself. Do you mean GPU where you write CPU? Lawrence It seems that all HCC1 results recieve points based on cpu use. I see the same thing. I get the same number of points if the WU has been crunched by the CPU or the GPU. In a way there is some logic behind this. It means that the point value is related to the WU. If your machine is more powerful and needs less time per WU then you can crunch more WUs in the same unit of time and you get more points. The objective is to crunch WUs, points are a consequence. I would find strange that the same WU would have more value in points if crunched by a GPU and less if crunched by a CPU. |
||
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges:
|
Ty Hyper, very informative breakdown. I have heard that with NV upper end 5XX and 6XX 2 GPU per is the sweeter spot and almost doubles the 1 per output. Have you tried that? It would also avail more CPU cores to CPU only projects.... If you can get that to work. As it seems you have. Just food for thought. You are perfectly right. With 2 GPU on the GTX 680 and 10 CPU threads the daily WU creep up to 616 instead of 550. Over 10% improvement. The GTX 580 behaves similarely except that the num bers are higher, this due to the different architecture between the 680 and 580. I must say that for a company like Nvidia that targets supercomputers, parallel computing etc. it is an extreme disappointment for their performance in WCG. We are not talking small change here. The top ATI board is 6 times (600%) more powerful than the top Nvidia board. Cm'on Nvidia wake up!!!!! I tested the GTX 580 and the behavior is surprising. As a reminder CPU is 980X running at 4.0GHZ with Win 7 64bit and ram 6GB running at 1600Mhz. Boinc is version 7.0.28 64bit With 2 GPU + 10 CPU I get 824 WU / day With 4 GPU + 8 CPU I get 972 WU/day With 6 GPU + 6 CPU I get 955 WU/day reduces a bit With 10 GPU + 2CPU I get 1090 WU/day maximum performance yet and a little above to the HD 7970 with 1 GPU+ 11 CPU. The multiple thread GTX 580 can beat the stock monothread HD 7970. Wow my GTX boards are not fully dead yet.  Interesting scaling effect. I will have to try 12 GPU and no CPU. Stay tuned. Running 12 CPU on the GTX 580 is worse and starts generating compute errors. Really the sweet spot is between 8 and 10 and will depend upon the boards. There is a variability of 10% between boards of different manufacturers even if the clocks are identical. I have decided to put all my 580 boards at 10GPU + 2 CPU. |
||
|
|
RaymondFO
Veteran Cruncher USA Joined: Nov 30, 2004 Post Count: 561 Status: Offline Project Badges: 
|
I think the use of various modified configuration file for graphics cards is something like cheating .. Or then wcg cell cheating cheating us who are not using something like that ..? .... Cooperation between the creators and developers in wcg graphics card should be more. Cheating? Really?! I would like to think what others are doing here as simply maximizing their hardware utilization to return the greatest number of valid WU’s for the research scientists benefit. The “carrot” for the crunchers is superior results and performance while this and hopefully other projects will benefit from these undocumented enhancements. This is what happened when individuals are motivated to explore new vistas and find betters ways of accomplishing more efficiency. One last point: This information is being SHARED!! [Edit 1 times, last edit by RaymondFO at Nov 12, 2012 12:13:19 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
OMG.... Enough with ripping at each other.
Just let it go on both sides and just crunch till cancer is a thing of the past, |
||
|
|
|