Note that those $/MH are based on 266 MH/s for each X6500. I have every reason to believe that we will reach higher hashrates with improvements to the software and firmware in the future.
This is fascinating to see someone actually do an FPGA farm in scale. I could not stomach the upfront capital cost needed to replicate my 18,000 MH/s. My my rough math, it would cost $35,000+ to get to that speed!
Now clearly cost savings would start to offset the capital cost, but that is a loooong payback. Also, the salvage value is much higher on GPU's, which has a huge impact on the NPV calculation.
I have invested $12,500 into my operation, and get 18,000 MH/s. $.69/MH. Pretty damn good!
I think it's time to update those calculations with our newer performance figures. As of the last few weeks, I've been running all of these boards at 180 MHz, or 360 MH/s per X6500. In the last couple days, I've been experimenting with 200 MHz, which seems to not work reliably with every board (some handle it fine as is). On one board that was having trouble, I removed the heatsinks and replaced the thermal tape with thermal epoxy. After that it has been running perfectly at 200 MHz. Power usage measured at the wall with a Kill-a-watt (before the 80 Plus Bronze PSU) is roughly 20 W (still need to measure this carefully at the board).
Assuming that switching to thermal epoxy will work well for the rest of them, let's say each X6500 will do 400 MH/s. This gives, for the 25 unit system, 10 GH/s instead of the 6.6 GH/s I calculated before. The price is still the same, so the cost/performance is reduced to only 1.358 $/MH/s.
Not too shabby, right?
