My new rig configuration:
4x5970
3.2 GH/s @ 1080W
3.02 MH/W
That is on 240V circuit, 80-Plus Gold PSU, usb drive running linux, optimized BIOS, underclocked & undervolted sempron, and >3 GH/s per rig.
I am pretty sure it isn't possible to get higher efficiency from a GPU rig without undervolting the GPUs. A 4x7990 (or possibly 3x7990) rig will likely beat that though.
D&T, what clocks are you running to get 3.2G out of 4x5970. ?
High.
They are water cooled. Each card is clocked differently because once you remove the heat each GPUs individual ASIC quality starts becoming the performance bottleneck. They average ~860 but no two GPU is clocked the same. Most are in the range of 830 to 875 range. I have found you can't get past 900Mhz and keep them stable.
I think the 3 phase VRM (5870 uses 4 phase) just can't handle it. I do have some lagards though. I have some which can't get above 800 even cooled to <40C. I have two cards which I think have a bad VRM because 1 of the 2 GPUs can't go above 700.
Of the 48 GPUs I had one simply die when I watercooled it. Explain that? Maybe I pushed it too hard and without excessive heat to throttle it back it killed itself. It had been flaky prior to watercooling it though. I think I am going to remove the waterblock and just use the remaining core for gaming.
Due to the luck of the draw not all the rigs hit 3.2 GH/s. Of the 5 rigs they are 3.22 GH/s, 3.20 GH/s, 3.15 GH/s, 3.08 GH/s, 3.01 GH/s. I got one more rig to convert. I am hoping it is a 3.3 GH/s+. The 3.08 and 3.01 likely could be pushed a little harder but building the watercooled server rack was a lot of work so I am happy they are just stable right now.
Ironically I am now kinda "over" mining. I am moving on to other things hopefully they just mine away for a long time.