Bitmain Antminer S9 + Low Power Enhanced Mode = 85 J/TH
BLUF: Check both boxes, ensure your miner is stable, and enjoy the efficiency gain.
We happened to notice a new Bitmain firmware released early December 2018 that included the AsicBoost upgrade (Low Power Mode) plus a new Enhanced Low Power Mode which lowers the hashrate in support of a lower power usage per TH.
At first we installed across our S9s but since we could only measure the power usage across the MoneyPit mine and we run mixed miner types it was initially tough to confirm the magnitude of the gains so we decided to back out the firmware changes to and do a an isolated test prior to rolling out the changes.
We proceeded to hack together a a quick tester that could extract the stats of a single miner (~every minute) while also monitoring the power usage (~every second) and post stats to our Elasticsearch + Kibana monitoring cluster to get a clear picture of the efficiency differences between the various modes in the firmware
We then installed the firmware on a single S9 (14TH) w/ an APW3++ and ran the unit with each configuration on/off to measure the J/TH.
The screenshots below show a 60 minute slice of each test run once the miner connected to the pool and stabilized.
Low Power Mode [OFF] / Lower Power Enhanced Mode [OFF] = 114 J / TH @ 14TH
NOTE: I was a little shocked how high this J/TH was, so may rerun at some point.
Low Power Mode [ON] / Lower Power Enhanced Mode [OFF] = 99 J / TH @ 14TH
TEST: Low Power Mode [OFF] / Lower Power Enhanced Mode [ON] = 97 J / TH @ 9.6 TH
TEST: Low Power Mode [ON] / Lower Power Enhanced Mode [ON] = 85 J / TH @ 9.6 TH
NOTE: In case you are curious , the temp in the room changed approximately 10degrees over the course of the various test run which did result in the rack supply temp changing. So on the absolute J/TH there might be a small affect from the temp, but I am confident things trended in the right directly since the “most efficient” runs (ON/ON) was during during the time when there was the highest supply temp.
