Different workunit sizes added

\n studio-striking\n

Message boards : News : Different workunit sizes added
Message board moderation

To post messages, you must log in.

Previous · 1 · 2

AuthorMessage
Profile [AF>EDLS] Polynesia
Avatar

Send message
Joined: 1 May 11
Posts: 23
Credit: 1,574,433
RAC: 0
Message 467 - Posted: 24 May 2011, 19:05:25 UTC

My card is 1176 Gflops from boinc manager, I think ... So should I get by the big units?

24/05/2011 20:57:57 | | NVIDIA GPU 0: GeForce GTX 470 (driver version 27527, CUDA version 4000, compute capability 2.0, 1217MB, 1176 GFLOPS peak)


Config : i7 860 2.8ghz, 8g ram, boinc : 6.12.26, GPU : GTX 470 Zotac Amp Edition 1280 mo DDR5
ID: 467 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Teemu Mannermaa
Project administrator
Project developer
Project tester

Send message
Joined: 20 Apr 11
Posts: 388
Credit: 822,356,221
RAC: 0
Message 469 - Posted: 25 May 2011, 6:13:44 UTC - in response to Message 467.  

24/05/2011 20:57:57 | | NVIDIA GPU 0: GeForce GTX 470 (driver version 27527, CUDA version 4000, compute capability 2.0, 1217MB, 1176 GFLOPS peak)


That's peak performance computed using manufacturer provided algorithm. Real performance is measured by our scheduler and is something different, usually lower. Here's the value for your host:

(cuda31) setting projected flops based on host elapsed time avg: 351.61G


So unfortunately this means you are getting normal work units.

-w
ID: 469 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile [AF>EDLS] Polynesia
Avatar

Send message
Joined: 1 May 11
Posts: 23
Credit: 1,574,433
RAC: 0
Message 481 - Posted: 26 May 2011, 17:46:28 UTC - in response to Message 469.  

24/05/2011 20:57:57 | | NVIDIA GPU 0: GeForce GTX 470 (driver version 27527, CUDA version 4000, compute capability 2.0, 1217MB, 1176 GFLOPS peak)


That's peak performance computed using manufacturer provided algorithm. Real performance is measured by our scheduler and is something different, usually lower. Here's the value for your host:

(cuda31) setting projected flops based on host elapsed time avg: 351.61G


So unfortunately this means you are getting normal work units.

-w


a unit end 192 is it a normal unit?

Config : i7 860 2.8ghz, 8g ram, boinc : 6.12.26, GPU : GTX 470 Zotac Amp Edition 1280 mo DDR5
ID: 481 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Purple Rabbit
Avatar

Send message
Joined: 14 May 11
Posts: 8
Credit: 415,505,070
RAC: 42,416
Message 483 - Posted: 26 May 2011, 19:15:36 UTC
Last modified: 26 May 2011, 19:17:01 UTC

I think something broke. I was getting 30 minute tasks for my HD 4770 in Tomato (http://moowrap.net/show_host_detail.php?hostid=809). I look away for a couple of hours and starting at 1600UTC I get a bunch of tasks with an estimated time of 3:43 hours. They are still really only 30 minutes, but poor Tomato thinks it has 6 days of work.

Don't do that! :) Tomato had a GPU coronary and I fell out of my chair. I think both of us have recovered, but the scheduling carnage is horrendous :) I think I'll live, but Tomato is mad at me.
ID: 483 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Zydor
Avatar

Send message
Joined: 5 May 11
Posts: 233
Credit: 351,414,150
RAC: 0
Message 484 - Posted: 26 May 2011, 19:24:01 UTC - in response to Message 483.  
Last modified: 26 May 2011, 19:25:38 UTC

BOINC broke ..... its a long standing problem. Its vertually impossible to startup new types of WUs with correct timing estimates. There is nothing the Project can do about it, its deep insde the basic way BOINC functions. There are some parameters that aliviate the issue which Projects do use, but its cursory only, it will never resolve the basic issue.

To illustarate that - do a day or so short running WUs of a few mins duration each, then switch to a Project that runs WUs of several hours and watch what the estimated time says :)

I believe CPDN - in the days of WUs that lasted literally months, some over 12 months - hold the record ..... a switch to them after doing short run WUs had estimates in thousands of days rofl ...... Tomatoe would expire on the spot by the sound of his sensitive soul :)

Regards
Zy
ID: 484 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Purple Rabbit
Avatar

Send message
Joined: 14 May 11
Posts: 8
Credit: 415,505,070
RAC: 42,416
Message 485 - Posted: 26 May 2011, 19:38:50 UTC - in response to Message 484.  
Last modified: 26 May 2011, 20:36:56 UTC

I'm not so much reacting to "inaccuracy" as I am to "different". No big deal. I'm not adverse to thinking that Tomato is playing a joke on me though :) I started with BOINC 1.xx and CPDN in 2004 so I understand :)

My order of blame is always: (1) Me, (2) Me again-What I thought should happen, but didn't, (3) the computer, (4) BOINC and (5) the project. Something hiccuped somewhere.

I haven't eliminated reasons 1 through 4, but I thought I'd mention it in case the project tweaked something and something else happened.

I've been down in the basement consoling Tomato. He has been doing his work contentiously for 6 years and this is the thanks I give him. He's on his second CPU/Motherboard transplant, third operating system, and I finally gave him a GPU worth something more than spit. The only hope he has is knowing that he is better than Cabbage. Cabbage (1.3 GHz Celeron) is sitting next to him and is having trouble finding meaningful work. Cabbage started on SETI in 1999.
ID: 485 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Teemu Mannermaa
Project administrator
Project developer
Project tester

Send message
Joined: 20 Apr 11
Posts: 388
Credit: 822,356,221
RAC: 0
Message 517 - Posted: 28 May 2011, 9:57:18 UTC - in response to Message 481.  

a unit end 192 is it a normal unit?


Yes. Anything between 50 and 300 stat units is considered a normal.

-w
ID: 517 · Rating: 0 · rate: Rate + / Rate - Report as offensive
AMDave

Send message
Joined: 3 May 11
Posts: 4
Credit: 135,212,867
RAC: 0
Message 525 - Posted: 28 May 2011, 15:13:29 UTC
Last modified: 28 May 2011, 15:58:16 UTC

I got the new WUs and they run 4 times longer on HD5700.
But it's not good.
The credit is averaging waaaay lower than the small wu's.
My completed and 'valid' stats are showing 400% more work but only 95% of the credit that I was getting for the small WUs.
That's so disappointing that I have to vote with my GPUs.
I'm ditching until that is sorted out.
I'll try again later.
ID: 525 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Previous · 1 · 2

Message boards : News : Different workunit sizes added


 
Copyright © 2011-2024 Moo! Wrapper Project