FINALLY SUCCESS: 10,000 coblestones/day @34-35 watts

David Rapalyea
David Rapalyea
Joined: 3 Jan 13
Posts: 79
Credit: 63,886,821
RAC: 0
Topic 196916

FREE DC reports production at 45,000 per day on my new computer built. At an observed cord power draw of 150-160 watts. PC build is described below.


First, I have a lot of posts to consider for the next build. I thank everyone for their comments and recommendations, they will be helpful. On this, my first ever dedicated computer build, this is what I learned.

1) Power draw went down and production went up when I took a sugestion to turn off the CPU for non-cuda tasks. This dedicated the CPU for driving the cuda cards. Thanks guys! It can be done on individual computers [tools/computing preferences then checkmark 'Use GPU' then enter 0.0 the box 'For multiprocessor sustems, use at most .... of the processors']. The same effect can be achieved for all your computers from the BOINC preferences, probably in a couple of ways. For instance, right now I am only accepting BRP tasks. BUT GPU MUST BE TURNED ON AT THE INDIVIDUAL COMPUTER LEVEL - at least in all my machines.

2) The build described below could not support my plan to run four GTX 650 cards. These cards draw power from the pcie slots and my builder reported the slots could only power two. Further, even when I mounted a third card with six pin independent power [GTX 660] BOINC only found it and one of the GTX 650 cards. More experienced members described how to fix this, and I may attempt that later, but this settup has met my goal anyway so is not a priority right now.

3)The GTX 660 produced the same production in both x16 and x8 slots [two BRP in 43.5 minutes], but pulled less power in the x8 slots. However, it has not lived up to expectations in this build. Although the 660 benchmarks at more then twice the 650s, it does not produce twice the work. Each 650 produces two standard BRP tasks in about 69 minutes or less, but the 660 requires 43.5 minutes. And always at disproportionate increased in power draw in comparison to the GTX 650s.

I doubt a more capable CPU would make much difference. I performed these same tests earlier using the same rig but with an intel G630T dual core 'Sandy Bridge' processor @ 2.3Ghz. The results were roughly the same. However, the Ivy Bridge i3 E3220 @3.3 GHZ draws 10 watts less and is not even at half capacity. The G630T was at max capacity.

5) The GTX 650s absolutely required x16 slots. In fact, they required the two in cluster number one; slots one and three. My build includes 32 pcie 3.0 lanes distributed in two clusters. Any combination that include the second cluster [slots four through seven] decreased production. Specifically, one of the cards always increased task time from about 70 minutes to more then 75 minutes.

6) I will never acquire another fly-wheel hard drive. Without the SSD the testing described above would have been impractical.

It will have the most minimal motherboard that will support the same CPU; and it must have two properly spaced x16 pcie 3.0 slots for my extra GTX 650s. Power will be fanless 80+ platinum. I suspect it will break 35w per/10k stones. The question is [drum rollllll] is it even possible to hit 29 watts!?

-- ASRock Full Gaming Motherboard
-- 32 pcie 3.0 lanes branched for two clusters [x16 x8 x16] [x8 x16 x1 x16]
-- Intel Ivy Bridge i3 E3220 @3.3 GHZ [two cores, two threads each] under full GPU load [two GTX 650 cards] total percentage load on both cores total never reached 40%
-- Seasonic ATX 360 watt 80+Gold
-- Solid State Drive, 30 Gig
-- 4 gig ddr
-- Windows Home 7-64
-- Two nvidia GTX 650 GPU mounted in slots 1 and 3
-- Full Gaming case with bays for seven hard drives, four optical drives, cooled by five independent case fans [all dissabled as superflous to this low power build].

Arecibo 19 Oct 2012
Just Because The Space Alien Is Green
Does Not Mean You Should Go

Joined: 13 Aug 10
Posts: 332
Credit: 1,714,373,961
RAC: 0

FINALLY SUCCESS: 10,000 coblestones/day @34-35 watts

Interesting read and a worthy goal to try to reduce the electricity bill. Good for the environment AND the wallet.

Power draw went down and production went up when I took a sugestion to turn off the CPU for non-cuda tasks.

I myself have considered doing this but since a big part (major part?) of this project is gravitational wave detection I wouldn't feel good about not crunching those CPU only workunits. Hopefully the clever people at E@H can figure out a way to run those workunints on a GPU as well. I know this has been suggested and good reasons have been given why it won't work, but I have faith in cleverness :)

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.