| 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 wii be keeping my 560M for my laptop;BUT have purchased a xfx 7870 double d for my new home built desktop w/asus p8z68-v pro gen3 and 750 watt supply.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
For the numbers listed, it seems like the "AMD Radeon HD 7700 series (Cape Verde)" is a hit. They are relatively cheap ($100-$150 depending on the model) and "fourth-best" in knreed's list.
 |
||
|
|
Crystal Pellet
Veteran Cruncher Joined: May 21, 2008 Post Count: 1317 Status: Offline Project Badges: 
|
For the numbers listed, it seems like the "AMD Radeon HD 7700 series (Cape Verde)" is a hit. They are relatively cheap ($100-$150 depending on the model) and "fourth-best" in knreed's list. The 7770 has 1.280 GFlops and the 7750 has 819 GFlops. For those who has to pay the electricity bill their selves: 110 secs AMD Radeon HD 7970 (Tahiti) 250W Be aware: the Card Watt numbers is with maximum GPU-load. 1 single HCC GPU task is not running to the max load and half of the time the card is idling, because only the CPU is working at the beginning and at the end of a task. My machine's used wattage is going from 120W (GPU idle) up to 150W when the GPU is really working on a task.  |
||
|
|
SuperMecha
Cruncher Joined: Apr 25, 2011 Post Count: 12 Status: Offline Project Badges: 
|
Does anyone know why there is a discrepancy in the relative performance?
For example the Radeon 7970 should be about 3 times faster than a 7770 and the 5870 should be twice as fast as the 5770. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hello SuperMecha,
I do not have a GPU to test, but most of the time is spent on the CPU, not the GPU. Even when the work unit is on the GPU, there is no certainty that the work unit is programmed in a way that uses the full GPU. It may be running across a subset of the GPU. Therefore, a less capable GPU may come closer to matching a high-end board than you would expect. Lawrence |
||
|
|
Promilus
Cruncher Joined: Apr 11, 2008 Post Count: 9 Status: Offline Project Badges: 
|
Does anyone know why there is a discrepancy in the relative performance? For example the Radeon 7970 should be about 3 times faster than a 7770 and the 5870 should be twice as fast as the 5770. Because single HCC GPU WU can't saturate anything above 7750 so u have to run multiple work units on single GPU. But don't be hasty - 7750 with single WU makes computer noticeably less responsive, it doesn't the same with 7850 or 7950 - but once you use app_info to get some 2-3 WUs simultaneously those powerful cards will "lag" too. |
||
|
|
Jim1348
Veteran Cruncher USA Joined: Jul 13, 2009 Post Count: 1066 Status: Offline Project Badges: 
|
It seems that most distributing projects use Nvidia cards, probably for performance reasons related to CUDA. But that leaves a large number of people with AMD cards without much to do. Therefore, there may actually be more free computing power available for AMD cards, even if individual Nvidia cards might perform better. And I think that Nvidia is moving to OpenCL in their next generation of cards (Maxwell), so that might be a consideration to take into account.
But I doubt that IBM will want to dictate one or the other; their antitrust lawyers would probably take a dim view of that, even if no one else did. |
||
|
|
mmstick
Senior Cruncher Joined: Aug 19, 2010 Post Count: 151 Status: Offline Project Badges:
|
Does anyone know why there is a discrepancy in the relative performance? For example the Radeon 7970 should be about 3 times faster than a 7770 and the 5870 should be twice as fast as the 5770. His chart isn't accurate. Realistically it is hard to fully utilize a single 7970 with only one work unit, therefore you need to run 4+ work units on the high end cards to fully utilize them, which leads to longer times per single work unit completed but overall faster returns. Example, my 7950 running 8 work units completes one work unit every 30 seconds, someone with a 7970 could easily achieve higher results if they overclock as far as I did my 7950, yet I still have room to improve my scores by over 30% if I clock my processor and graphics card higher. Therefore, Radeon HD 7970 IS 4x more powerful than 7770 at this project, and therefore much more power efficient, supposing 7770 actually consumes 80 watts, while 7970 only consumes 250 watts but gets 4x higher results. [Edit 1 times, last edit by mmstick at Oct 17, 2012 2:33:05 PM] |
||
|
|
Jim1348
Veteran Cruncher USA Joined: Jul 13, 2009 Post Count: 1066 Status: Offline Project Badges:
|
My whole PC draws only 103 watts when running HCC on my HD 7770 GPU, and 86 watts when it is running on the CPU (an E8400 at 3.0 GHz). Therefore, the HD 7770 draws 17 watts (for a singel task at a time). Note that there is also some static power in that, so if you removed the card entirely it would probably drop another 5 watts, so the card draws about 22 watts maximum at the plug. But if you want to be stricly accurate, you have to account for the power supply efficiency; I have a high-efficiency Seasonic that is about 85% efficient in that power range, so the card is really drawing only about 19 watts during the GPU portion of HCC.
And as posted above, averaged over 10 tasks I get a kernel time of 68.5 seconds. So you can compare accordingly. |
||
|
|
mmstick
Senior Cruncher Joined: Aug 19, 2010 Post Count: 151 Status: Offline Project Badges:
|
That's an incredibly high kernel time, you should fix that. If your kernel time is 68 seconds, something is very wrong. My kernel time is only 2.36 seconds with my 7950 averaged across those work units.
----------------------------------------[Edit 1 times, last edit by mmstick at Oct 17, 2012 2:41:19 PM] |
||
|
|
|