I'm not sure how accurately the Bitmain driver handles the work difficulty from the pool.
I do it as follow:
https://github.com/ckolivas/cgminer/blob/master/driver-bitmain.c#L592

Thus the nonce finding variance of the Ant is close to the share finding variance - which reduces internal IO from the Ant chip also.
The Ants can do work difficulty at a power of 2 - 1, 2, 4, 8, 16, 32, 64 ,128, 256, 512, 1024, 2048, 4096 (I stop at 4096 so direct solo gives results)
So if the pool difficulty is e.g. 300 the ant will mine internally at 256 difficulty and since the hash rate is based on nonces found, the hash rate's variance will be the equivalent of mining on a pool at 256 difficulty so will be noticeable.
Bitmain's driver does this also, but I'm not sure how accurately - it may be the same or lower for them.

If it's consistent, then it is hashing faster.

Where it starts, depends on the luck in the early time after the start ... up to an hour or so ... then it will gradually average out to the expected rate.
If you are lucky at the start it can be quite high, I've had my S3 mine at around 500GHs for an hour - luck/variance - but eventually it settles down and heads towards 440.

Think of it like block finding.
A pool can get lucky and find lots of blocks fast and get unlucky and have long dry spells of no blocks.
Over time the expected average of a pool is 100% diff (less orphan rate)

WIth the S3, it's on a much smaller scale, but the same idea.
Over a days or two it will certainly approach the average expected hash rate, but when you first start it up the variance can seem quite high.
This is, again, due to the fact that it is returning a LOT fewer nonces.
If it is internally mining at 256 diff, then it will have only 1/256th of the amount of results - so it's sort of like saying it will take 256 times longer to average out - well it's not 256, but the idea is the same. That's variance.

It's not affecting the number of shares you are submitting to the pool.

I know this has been asked before but i never saw a answer  any place, I wasn't the one that asked it before but am now I upgraded a s1 to S3 how do i make fan 2 show the speed . it shows everything else but that even with the new update . again  i know its been asked but i have yet to see any good answer why or how to and this is the first time i have come out and  asked it .


Just peaced to gather a S3 from parts i bought i had laying around for fun it does the same thing . is it because of the controller to power board for the S1 to  S3 .

I found a Full S3 controller i have coming some one sold me if that's it , ( The controller from S1 to S3 ) .  i may buy one more full controller off the guy.

You'll probably have to swap the PWM and tachometer terminals on the fan if it is not showing speeds on the second fan.

Right, and generally what I see reported at 5s doesn't get me excited.  I only mentioned it in the recent posts because I'd never seen it report back as high (or low) as it has been since updating to 4.9.2.

For clarification, my other comments re:500GH and dying down to 480GH after 1-5 days is not looking at the 5s number, obviously, but at the average for the entire uptime.  20 hours since last restart and average hash rate is holding at 506GH/s, all systems go.

Downloaded and installed on my S3s.  Thanks again for your work, kano!

I am getting regularly
on latest cgminer-4.9.2-windows running on winusb on antminer U2
what does it mean? is it critical for solo mining scenario?

Yes, it is U2 (ANT USB 1.2). U2 is downclocked to 200 MHz. Seems the error is because of Logitech USB cable.
Do I loose shares or other data because of error? Are consequences recoverable?

Firstly I've worked a little different to other devs ... the below refers to me only

Well my first AntS1 I bought but Bitmain let me in the 1st batch.
My first AntS2 someone else bought for me (not part of Bitmain)
But bitmain have supplied AntS1, AntS3, AntS5 (currently dead), 2xU3 (that's why I know they are so crap), U1s and U2s

Nowhere along the U1/U2/U3 path have they taken any notice of:
https://bt.irlbtc.com/view/294499.0
Though they definitely knew about that post

I have bought some hardware over the years ... paid 128BTC for my first BFL fpga
At the time that was well priced and I got it in 1 week, not from BFL, unlike everyone else who spent months after paying waiting for them.

I've never been too concerned about the value of the hardware vs the work I've done, until recently.
Over the past year or so it's been all about manufacturers selling hardware to business and the bitcoin community has been pretty much left behind in the dust ... or buying hardware that they'll definitely lose BTC on - so yeah I've not been liking working at slave labour rates so been expecting them to cough up some BTC - also since the hardware is mostly worthless in anything but large quatities running on cheap power in a data center somewhere, it's sometimes a loss for me to get the hardware for dev purposes at home ...