Yesterday I got the robot set up to run out 2Pacs, today I did a bit of dabbling with voltage ranges on the buck. I had noted that these guys don't particularly like to initialize below about 630mV per node even though at low frequencies (100MHz) they should be good to 600mV, but at the bottom threshold of voltage, chip power draw balance and node-level storage matters a whole heck of a lot. So I figured up buck voltage adjustment resistors that'll give you 630mV to about 830mV, which means it should be good for at least 400MHz.
I ran a quick efficiency test, only up to 300MHz because my USB power meter is only rated for 3A.
Freq | GH | Vcore | Power | J/GH |
100 | 11.00 | 1.26 | 3.67 | 0.33 |
125 | 13.75 | 1.26 | 4.48 | 0.33 |
150 | 16.50 | 1.26 | 5.27 | 0.32 |
175 | 19.25 | 1.26 | 6.12 | 0.32 |
200 | 22.00 | 1.32 | 7.43 | 0.34 |
225 | 24.75 | 1.36 | 8.73 | 0.35 |
250 | 27.50 | 1.40 | 10.11 | 0.37 |
275 | 30.25 | 1.46 | 12.25 | 0.40 |
300 | 33.00 | 1.50 | 13.98 | 0.42 |
As you can see, efficiency across the bottom end of frequencies is pretty flat. That's because they're all at the same voltage level. Efficiency actually improves slightly with an increase in frequency because the maintenance current (buck controller, LDOs, CP2102 etc) is pretty constant so a higher proportion of power is used for mining.
At 300MHz, the total device efficiency is only about 15% worse than the datasheet chip-level efficiency, which I think is pretty good for a stick miner. It also puts up over 30GH, which I think is pretty darn good for a stickminer. Note this is about the same power draw the original Compac expected to see at 22GH, so we're seeing a 50% improvement in hashrate. If the stick could be pushed to 400MHz you'd see it drawing about 23W for 44GH, but I don't want to test this until I have a bit stouter measurement tool.
This is some nice number!