| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Website Support
Thread: CPU/GPU |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 28
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I would KILL to be able to crunch with my GPU's. Let's get on with it!
|
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges: 
|
I would KILL to be able to crunch with my GPU's. Please don't do it.  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I agree that GPU support would help immensely. Running the same thing on my ATI 4350 card: 92 GFLOPs/22 watts = 4182.8 MFLOPs/watt Not quite. Your ATI-Card does have 80 Cores that means 80 processes running concurrently, which is roughly 6Megs/process - looking at the memory footprint of the applications this is not really a lot. It mostly depends on the localism of the data - i.e. when there is not a lot of memory involved - remember the memory and the need to frequently swap data between RAM and GRAM. So if you can e.g. use 10% of your gpu-cores the calculation is somewhat different - 9GLOPS/22W~400GFLOPS/Watt. Keep in mind that there are 2 versions needed for each architecture - Stream and CUDA and that this is somewhat dependant on the drivers used. And your CPU is somewhat dated - more than 3 times the age of your GPU. Compare with e.g. an Phenom X4 9950 - Quad-Core 2,2GHz, 95W, ~15GFLOPs = 157GFLOP/W |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
To once more inject, that the blind staring on what a GPU can do... the particular science has to fit and gain, not loose resolution, precision to the extend that a ported science can even run slower on a GPU. Till now only HCC is considered suitable [see post bbover3]... but wonder what happened in that decision process when the recent optimization made jobs go from 3.5 hours to 1.25 hours and less under Linux on a CPU. Getting the tasks in and out of WCG then is a potential bottleneck that needs being dealt with. We're heading already to 600,000 plus WU's per day.
----------------------------------------We'll see.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
AMD @ amdforum.se
Cruncher Joined: Oct 29, 2008 Post Count: 2 Status: Offline |
Give us GPU folding now! :)
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Give us GPU folding now! :) Agree, but reading the lot of others threads about this subject I think that WCG are more inclined in don't do it for now. I bet that we will wait sometime yet, maybe at least until a new project sponsor take the leadership to try use it here in a project (that will be a huge hit for sure !). |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I like it, but maybe need wait a little bit more.
I'm gammer user and like this suggestion ! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Without resorting to using GPUs, the obvious answer to flagging user support is to up the BOINC credits per WU on WCG. This has always been notoriously low, and a repeatedly contentious issue in user posts for years. I believe that you are incorrect. WCG follows the original standard for Boinc credits and many of the other projects desperately inflate their points. I have seen a single E8400 system with 2 cores (running Collatz) getting more than 125 times the credits as my E5200 with 2 cores. I'll give you double for the CPU efficiency and I'll give you another double for the CPU speed AND I'll give you another double for the (possible) overclocking. That's 8 not 125 times. Which project is wrong? Hard to say. [edit to change example] I crunch for collatz. I have an AMD Phenom II quad core processor, and a ATI Radeon 5770 video card. It takes the video card about 13 minutes to complete a Work Unit, while it takes one of my cores about 8 hours to complete another Work Unit. I suspect for certain types of calculations, a GPU is about 100 to 200 times faster than a CPU. I do not believe this type of efficiency will transfer across to all projects though. Collatz is special as it is only doing multiplication, division, and addition in integer format. Jimmy [Edit 1 times, last edit by Former Member at Apr 8, 2010 5:17:03 AM] |
||
|
|
|