Antminer efficiency in the real world doesn't match specs

Hey folks, I'm in the middle of testing several Antminer S21 XP Hyd units to see if they can actually work for a full container deployment. I kicked things off by ordering 36 units to fit a small hydro cooling setup. I've been running tests on these machines with different firmware options: Stock firmware, VNish, and Braiins OS. What’s surprising is that in nearly every case, the real-world efficiency falls short of what the manufacturer claims. Out of all the units, only about 1 or 2 on the stock firmware are hitting those efficiency numbers. Plus, even for those few, reaching that spec is only doable if the room temperature stays below 30°C. So, I wanted to check in with you all: What’s your experience with the S21 XP Hyd? Are other Antminer users actually seeing performance and efficiency that line up with the official specs in your setups? Can’t wait to see your thoughts and experiences.

What's not working for you? Are your results deviating from the manufacturer's hashrate and energy efficiency specifications? (1-1) The hashrate value, and power efficiency on wall are all typical values. The actual hashrate value fluctuates by ±3%, and the actual power efficiency on wall fluctuate by ±5%. https://shop.bitmain.com/product/detail?pid=00020250806183122100IOerDTTH068C

what were your results then? Off by how much?

This ASIC has three operating modes with standard firmware: Normal Mode: 473 TH/s and 5676 W. High Performance / Plus Mode: 500–515 TH/s and 6100–6300 W. Energy Saving / Low Power Mode: 380–410 TH/s and 4300–4600 W. It also has an additional overheat protection mode called Dynamic Frequency Scaling. As the chip temperature increases, the operating frequency decreases, and the hashrate drops. S21 XP Hyd Specification

bitmain has temperature sensitivity with air-cooled gear. nice to see it is consistent with liquid cooled gear. the chips are what do it. once they warm beyond 75C the efficiency drops. my s21xps will do 3400 to 3700 watts depending on ambient air temps. they are hosted in Iowa and in the winter under 40f they run way better then in the summer over 80f. my guess on water cooled if the radiators can't eliminate heat as well when it is 30c or more. How bad is it for you 10% off is common when warm and -5% is common when cold with air cooled. average it out and you are near 5% off for the year. please post back as I would love to get some info on the liquid cooler gear. My host wants to begin with liquid cooled gear and I am on the fence about getting a few s23xp hydros.

I am sure that the lower the temperature of the ASIC chips, the better the efficiency. This does not depend on cooling if the chips have the same temperature. I would first see how the hosting would do with water cooling, and then buy ASICs there. My colleagues modified the liquid cooling system several times to get the desired result.