I've been tuning my L3's recently and I have a few questions I was hoping someone could clear up?
1) Are HW errors, rejected/stale shares, etc factored into the reported average hashrate on the miner's status page? In other words, if I have my miner clocked really fast but it's getting way to many HW errors, will it still report a really high hash rate or will it display an "effective" hash rate that is basically all the work done per unit time excluding HW errors?
2) My understanding as far as determining if your HW errors are too high is that the best metric to use is the Nonce# HW error %. I seem to recall that this should be fairly low, but what is a good upper limit to aim for?
2.5) Am I correct in assuming that the DiffA# HW % field is representing the percentage of difficulty=1 equivalent shares (each share multiplied by it's difficulty) that was/would have been accepted but was instead wasted because of a HW error?
3) I'm keeping logs in a spreadsheet of certain metrics for different tuning parameters and calculating other things to use to evaluate how well those parameters perform. One of those is HW/min which is calculated by taking the number of HW errors per board and total and dividing by the total runtime. Is this a reasonable/meaningful way to measure/estimate how much farther I can push a given board/machine (Ie, increasing clock speed or lowering voltage) before it starts to produce an unacceptable amount of HW errors? Are there other things I should be recording/calculating?
4) What is the meaning of the Local Work, Utility, and WU fields?
5) I always assumed the "best share" field is showing what the maximum network difficulty the best share/hash computed by the miner could have been if the miner were to have found a block. In other words, if the best share field read "500,000" then that would be like saying "If the network difficulty were 500,000, the best hash found would have satisfied the requirements to solve/create a new block". But I frequently see the best share field displaying a number substantially greater than actual global network difficulty. Is this because I misunderstand what that field means?
here's a picture showing a sample of what I'm talking about in some of these questions.
https://imgbox.com/ix01KdiMhttps://ibb.co/X55mCFvHi, thanks for bringing this up. I’m currently tuning a few L3+ units as well, so your questions are really interesting to me.
From my own testing, I’ve noticed that the miner status page can look better than what the pool actually sees. That is why I personally don’t trust the local hashrate alone. I mostly compare it with the pool-side hashrate over a longer period, ideally 12–24 hours.
My current setup is 4x Antminer L3+ with Hiveon firmware. Three of them are running around 375–378 MHz at 9.32V and they are doing roughly 490 MH/s each at about 645–650W, with very low HW errors and chip temps mostly in the low-to-mid 40°C range.
One unit I tested overnight at 394 MHz and 9.44V. It ran for about 8h 40m at around 513 MH/s average, about 701W, with 49 HW errors and around 0.005% HW error rate. Temps were still fine, max chip temp around 53°C. So it was stable, but the efficiency was slightly worse than the lower-voltage profile.
For me, the important lesson so far is: higher frequency is not automatically better. If the pool-side hashrate does not increase enough, the extra heat, PSU load and HW errors are not really worth it.
I’m also trying to understand the best way to judge HW errors. My current approach is to look more at the percentage values and pool-side performance than raw HW/min, because HW/min alone can look worse just because the miner is doing more work overall.
So far, my best “safe” profile seems to be around 375–378 MHz at 9.32V. It gives nearly 500 MH/s per miner, low temps, low errors, and seems much easier on the power supplies (2x 1500w).
Would be very interested to hear what limits other people consider acceptable for HW error percentage and whether they trust Nonce# HW % or DiffA# HW % more for tuning decisions.
