| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: GPU Support Forum
Thread: GPU vs CPU - trying to compare time and complexity of work |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 6
|
|
| Author |
|
|
micropro-be
Cruncher Joined: Dec 9, 2019 Post Count: 10 Status: Offline Project Badges: 
|
Hi all,
----------------------------------------I know I'm not the best when it comes to WCG research. However, I'm using a Nvidia Quadro P620 and I've tried some work units (still working as I'm writing). For a task with a Intel i5 9600K locked at 3,7Ghz, it takes about 1h30 at most to finish the job. With my Quadro, it takes about 11 minutes at most. So, question is simple, if the intel i5 9600K comes to analyse the same thing than the Quadro P620, how much time will it take roughly ? The Quadro P620 is not a cutting edge card. When it idles, it's at 32°C and when it's fully working for WCG it reaches 62°C easily with stock cooler (I've lost my belove RTX 2060 three days ago trying to fix the ventirad). I know the GPU can work on most complex tasks but... better work than the processor with such a "tiny" card ? Really ? Anyway, if somebody has some answer let me know. I'll adjust my way of working. Best regards, micropro  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
"I know I'm not the best when it comes to WCG research."
You know, rest assured, not the worst either. 🖖 /OT |
||
|
|
uplinger
Former World Community Grid Tech Joined: May 23, 2005 Post Count: 3952 Status: Offline Project Badges: 
|
The quick answer is, it is estimated to take 20x longer to run on CPU if it were to attempt an OPNG task.
Long answer: OPN1 (cpu work units) run on a single core, so please remember that if you have an 8 core machine, it could run 8 work units at a time. The benefit of the GPU is that it uses all the cores in a GPU card to run in parallel a single job within the work unit. This allows it to process things faster. Also, the GPU version of autodock uses another enhancement that allows it to stop when it gets to an answer. So, an 8 core machine may take 3 hours 45 minutes to complete a similar task your GPU took 11 minutes to run. Thanks, -Uplinger |
||
|
|
micropro-be
Cruncher Joined: Dec 9, 2019 Post Count: 10 Status: Offline Project Badges:
|
Thank you for your answer.
----------------------------------------The example of the 8 core taking much more time alongside a GPU taking 11 minutes convinces me to use the Quadro. Actually, I was worried about having a too old graphic card to get a good ration restult/time but it seems OK. Thanks again for the reply. Best regards, micropro PS : I might make some mistakes in english but it's not my native language, sorry  |
||
|
|
uplinger
Former World Community Grid Tech Joined: May 23, 2005 Post Count: 3952 Status: Offline Project Badges:
|
The one thing to note though is we are only able to send out 200 batches a day for OPNG (GPU). This means that only 81k work units are available which makes them harder to run. The CPU version is still sending out the same amount it was before. In the future, the GPU will encounter more difficult work units that should cause it to run longer than what is being seen right now. Your computer should be able to handle both at the same time, which means you'll be adding an equivalent of an extra machine computing results for OpenPandemics by running both.
Also, English is my native language, and I make many mistakes :) Also, your posts were good in my eyes :) Thanks for the questions and participation in World Community Grid. -Uplinger |
||
|
|
Ian-n-Steve C.
Senior Cruncher United States Joined: May 15, 2020 Post Count: 180 Status: Offline Project Badges: 
|
In the future, the GPU will encounter more difficult work units that should cause it to run longer than what is being seen right now. Two questions: 1. how difficult/long will these be? by more difficult do you mean more packaged sub-tasks, or longer/harder sub-tasks? just curious what kinds of plans you have for these. we talking like runtimes of hours or days for a single tasks? 2. any estimated time frame for when we might see these harder tasks?  EPYC 7V12 / [5] RTX A4000 EPYC 7B12 / [5] RTX 3080Ti + [2] RTX 2080Ti EPYC 7B12 / [6] RTX 3070Ti + [2] RTX 3060 [2] EPYC 7642 / [2] RTX 2080Ti |
||
|
|
|