Message boards :
Number crunching :
More WU's?
Message board moderation
Author | Message |
---|---|
Send message Joined: 18 Jul 11 Posts: 22 Credit: 4,160,197 RAC: 0 |
Back with another question. Just bought an ATI 6950 so I could do MW, & it's working fine. Problem is, even with priority set @ 400, project will only give me 12 WU's at a time. I crunch them in a few minutes & it takes Boinc considerable time to update & replace them. Had to add Primegrid & Collatz just to keep card busy. Is there some way to get the project to send more MW WU's? Thanks! |
Send message Joined: 26 Feb 11 Posts: 170 Credit: 205,557,553 RAC: 0 |
There is nothing YOU can do to change this. Its a project limit. We hope all that it get raised some day ^^ DSKAG Austria Research Team: http://www.research.dskag.at |
Send message Joined: 18 Jul 11 Posts: 22 Credit: 4,160,197 RAC: 0 |
Thank you! |
Send message Joined: 21 Sep 11 Posts: 41 Credit: 185,458,736 RAC: 0 |
Mmmmm. 6950. yum. I have to ask, how long does take to do a work unit ? :-) |
Send message Joined: 18 Jul 11 Posts: 22 Credit: 4,160,197 RAC: 0 |
Avg total run time on my system is about 83 seconds. Range is 66 to 130 or so. Does that sound right for the ati14 task? I'm still new at this. |
Send message Joined: 21 Sep 11 Posts: 41 Credit: 185,458,736 RAC: 0 |
That's excellent! My creaky old 5850 runs about 20-25% slower than that, so it sounds about right. BTW, you can select a task and hit the "Properties" button to get the apps estimate of your GFLOPS/Second. My gpu tasks read 382.32 while the poor Phenom cores rate 2.6 each! How did we ever crunch without GPU's, eh? :-) |
Send message Joined: 13 Sep 11 Posts: 3 Credit: 9,342,428 RAC: 0 |
I am curious about the GFLOPs/sec rate of that HD6950 too. The VLIW4 architecture is supposed to be more efficient than the older VLIW5 (like HD5870) on double precision operations. AMD says that 6950 delivers a theoretical maximum of 562.5 GFLOPs/sec at 800 MHz. But you should first let the MW project run for some time (like a week or more) so that the BOINC client can adjust the flops factor accordingly. Before that the GFLOPs/sec reading will be smaller than the real value. My o.c.ed 5870s give 442.45 GFLOPS/sec each (and still slowly increasing). That's 77% of the theoretical peak performance of 576 GFLOPs/sec at 900 MHz and [in my opinion as a computational scientist] is quite a good utilisation of the full GPU potential. |
Send message Joined: 18 Jul 11 Posts: 22 Credit: 4,160,197 RAC: 0 |
The task I checked (I didn't know you could do that) is running @ 413.79. The card is running at the default 830 Mhz. It has been running MW for about 4 days. My problem is getting enough WU's to keep it busy. According to my figures, MW yields 150 credits/running minute, while Primegrid is at 87, Collatz at 111, and Moo at 199. I find if I only run one other ATI app, I get the most MW units for some reason, so I'm running Moo and MW. I chose the 6950 over Nvidia 'cause I wanted to run MW, and it didn't matter for the video editing I'm doing. It's disappointing that I can't get enough work to keep it busy. Oh well, I can't get SETI WU's either right now. |
Send message Joined: 2 Oct 10 Posts: 74 Credit: 18,362,557 RAC: 0 |
My 6970 with no OC (@830) reaches a minimum 500 GFLOPS/Second easily. |
©2024 Astroinformatics Group