| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Help Conquer Cancer
Thread: GPU Optimisations |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 198
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Do you really get 8 with .125 on the GPU as using 0.05 does some round down to 19? Reported at devs mail list, who did not think it worth responding, but one commenter there speaking up as it just being a minor annoyance, if any. [Seemingly annoyed by the report of a minor bug :O . Makes one wonder.] Hey Rob, Yeah eight tasks running on each GPU without errors (as per (blurry) screen capture shot) What I suspect might be up with the error you mentioned is with uneven (odd 3,5,7 ect) amounts of WUs. There is definately an issue with the code (somewhere) as with a few settings I have tried I was getting extra cpu task/s running than should be with my app_confing settings. The extra cpu task/s (I assume) were the cause of the computation errors that were generated with them. |
||
|
|
mmstick
Senior Cruncher Joined: Aug 19, 2010 Post Count: 151 Status: Offline Project Badges: 
|
Three 7770s? Christ, you should have bought a 7950.
|
||
|
|
branjo
Master Cruncher Slovakia Joined: Jun 29, 2012 Post Count: 1892 Status: Offline Project Badges: 
|
Computing power of 7770 is 1.28 TFLOPS SP. X3 = 3.84 TFLOPS SP.
----------------------------------------Computing power of 7950 is 2.87 TFLOPS SP. Q: Why he should take 2.87 TFLOPS instead of 3.84 TFLOPS?  Crunching@Home since January 13 2000. Shrubbing@Home since January 5 2006  [Edit 1 times, last edit by branjo at Mar 9, 2013 8:58:21 PM] |
||
|
|
mmstick
Senior Cruncher Joined: Aug 19, 2010 Post Count: 151 Status: Offline Project Badges:
|
Computing power of 7770 is 1.28 TFLOPS SP. X3 = 3.84 TFLOPS SP. Computing power of 7950 is 2.87 TFLOPS SP. Q: Why he should take 2.87 TFLOPS instead of 3.84 TFLOPS? You are off by quite a significant margin... therefore I can tell you don't really know much about FLOP ratings of hardware you are speaking about. A 7950 operates easily at 4 TFLOPS with a decent OC to 1100Mhz; All 7950s operate at 1000MHz ever since the BOOST thing, so they don't operate anywhere near this fictional 2.9 TFLOP number. Total kernel time is 27.3 seconds for one of my work units, processing 16 work units on one 7950. Average time is 14 minutes to complete 16 work units, or 52.5 seconds per work unit. I'd like to see your measly 7770s overcome that. I will speak with red truth, a single 7770 is only one third the power of a 7950. It is impossible to make them operate faster than the whole. [Edit 2 times, last edit by mmstick at Mar 10, 2013 12:33:17 AM] |
||
|
|
Jim1348
Veteran Cruncher USA Joined: Jul 13, 2009 Post Count: 1066 Status: Offline Project Badges:
|
I've just run some tests on my HD 7770. Running 4 work units, and devoting all four cores of my i5-3550 (one core per work unit), it takes 11 minutes 25 seconds to complete the four work units, or an average of 171 seconds per work unit. However, that is starting all for work units at the same time, so that the maximum CPU usage overlaps, and hence the GPU is not always running at maximum (99%). So if you allow time for the maximum CPU usage to diverge, then the average run time would be somewhat less. Also, the card is in a PCIe 2.0 x4 slot, which will cause a small speed reduction (though I don't think much, since the CPU and GPU usages aren't overlapping) as compared to a PCIe 3.0 x8 or x16 slot.
----------------------------------------The bottom line is that if you ran three cards over a period of time, you should average somewhat less than 171/3 = 57 seconds per work unit. (This assumes that you have enough CPU power to support three cards, so that would make a difference if not.) [Edit 2 times, last edit by Jim1348 at Mar 12, 2013 2:50:36 AM] |
||
|
|
Jim1348
Veteran Cruncher USA Joined: Jul 13, 2009 Post Count: 1066 Status: Offline Project Badges:
|
I am now running four work units on a single CPU core, since I don't like tying up my i5-3550 only for HCC, and did the above only for a test. But by allowing the work units to run for several hours, the peak CPU times no longer overlap, and have spread out very nicely. The GPU is running at 99% utilization almost all the time, and the four work units complete in 10.5 minutes, or an average of 157.5 seconds per work unit.
----------------------------------------So if I had three HD 7770s and enough PCIe slots (I have only two), I could get a work unit done in 52.5 second using only three cores of an i5-3550. (But finally note that all such numbers should be averaged over a long period of time to do valid comparisons; I have just been at it with multiple work units for a few hours, so they should not be taken for more than they are worth.) [Edit 3 times, last edit by Jim1348 at Mar 12, 2013 8:30:58 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
my hd 7870 w/i5 2500 doing only hcc gpu
----------------------------------------11 day average wu completed in 24 hour's is: 1,296.9 runing 10 wu's at a time @ oc gpu 1140mhz [Edit 1 times, last edit by Former Member at Mar 12, 2013 12:01:34 PM] |
||
|
|
Lost Cavallero
Cruncher Joined: Jan 12, 2005 Post Count: 6 Status: Offline Project Badges: 
|
Good day to you all.
I have question about my ATI HD 6570 2GB. I installed latest drivers for it and I have latest BOINC 7.0.56. System: Win 7 32bit CPU: Core 2 Duo 3.16GHz RAM: 4GB I'm doing only 1 WU and GPU is 98% used. I finish 1 WU every 10min. I tried doing 2 WU and it works fine (no errors reported) but every 4h my system restarts itself with some Win system error reporting. Probably driver... Is this normal speed for this card or I can do something to work at least 2 WU? Thank you all for any suggestion. |
||
|
|
branjo
Master Cruncher Slovakia Joined: Jun 29, 2012 Post Count: 1892 Status: Offline Project Badges:
|
Time seems to be OK.
----------------------------------------So, you have installed "Catalyst Software Suite" v 13.1 from here, right? Have you tried "Latest Beta Drive" from the same link (requires .NET 4.0 support for Windows Vista and Windows 7 users)? I think you can try 2 concurrent, but since your GPU is utilized by 98% with 1 task, do not expect significant increase of production. Cheers and good luck  Crunching@Home since January 13 2000. Shrubbing@Home since January 5 2006 [Edit 1 times, last edit by branjo at Mar 14, 2013 10:49:23 AM] |
||
|
|
Lost Cavallero
Cruncher Joined: Jan 12, 2005 Post Count: 6 Status: Offline Project Badges:
|
Time seems to be OK. Thank you for your reply. That's what I expected. So, you have installed "Catalyst Software Suite" v 13.1 from here, right? Have you tried "Latest Beta Drive" from the same link (requires .NET 4.0 support for Windows Vista and Windows 7 users)? That's right. I tried old, new and beta driver. Old drivers utilized my CPU to 90% but new one doesn't have that problem now. New driver use 98% - 99% of GPU and 17% of CPU. I think you can try 2 concurrent, but since your GPU is utilized by 98% with 1 task, do not expect significant increase of production. I run 2 concurrent now after installing all Windows Updates ( I'll try that too  ). Time is 15min for two... that's 7.5min per unit. I'm just hoping that my Windows won't restart after few hours. |
||
|
|
|