Bitcoin Forum

Hi guys, so I know it's not very profitable obviously but is it possible to underclock the antminer s2? From what I understand, you can change the frequency lower but you need to change the voltage too or not? I need help on the voltage part. Thanks!

Look through this thread  ;)

https://bitcointalk.org/index.php?topic=515448.0

Another with good info:

https://bitcointalk.org/index.php?topic=1023047.0

You can underclock it so its less heat/noise but the power consumption will always be fixed since you can't change the voltage

how the heat is reduced if there is the same wattage(you said the same consumption)? it does not make sense

I have to agree with Amph here. If underclocking reduces the heat output, then the power draw has been reduced. Watts are Watts and it all ends up as heat one way or another.  It's also typical for a reduced clock frequency to end up with a reduced power draw. This is all standard ASIC "physics", nothing magic.

It is also true that in many/most ASIC designs a reduced frequency would operate correctly with a reduced voltage. If the manufacturer hasn't the provided a means to adjust voltage, or their design makes that impractical (e.g. a string design), then you don't get as much power savings by simply reducing the clock frequency.

Perhaps what he means to say is the efficiency(W/GH) will always be fixed, not the power consumption.

Yes that's what I meant. The W/GH will always be fixed but the total watts will be lower along with the hashrate when underclocked.

I don't think Antminers downclock like that, at least S1's didn't. The lower you downclocked an S1, the more efficient it became. (higher GH/W ratio) Inversely, the higher you clock, the more inefficient it becomes. Downclocking made a huge difference with S1's as well - it would run at 180GH/s @ 360W (2W/GH) by default, but by downclocking to 140GH/s it would only take 170W. (efficiency of 1.21W/GH)

So as long as the S2 downclocks the same as the S1, (which I fully expect) the S2's efficiency will increase as well.

My understanding is that the S2 is fairly close to the maximum efficiency of the BM1380 ASIC.  Most likely there is not much more to be gained by underclocking.

1 watt per GH is about the best this chip can do.  Maybe .9 watt with some intensive tweaking of the voltages.  Would be very tedious for and S2.

The voltage's can't be altered however. THere is no option in the settings.

You have to "manually" tweak via pencil modding.  You add resistance to a feedback circuit in the regulator, hence lowering the voltage.  Would have to do this

40 times in the case of a S2.  With the S1, only 8 times.  This is why it would be tedious. ;)

Actually, you reduce resistance. I haven't tested on an S2, but based on my tests on the S1 it should be possible to get 0.8W/GH out of them, board-level.

If I have time this weekend I will give it a shot.  They're just gathering dust right now anyways.

It would be nice to see how it performs below default clock... Please post settings and firmware that you are running...

I've put up a new master cgminer AntS2 firmware in my cgminer-binaries git
of current git cgminer 4.9.1+ modified from the last Bitmain firmware 29-Oct-2014 and including all the changes I did last year
https://github.com/kanoi/cgminer-binaries/tree/master/AntS2

It includes a half speed half power setting in the Miner Configuration->Advanced Settings tab ... which isn't really all that useful, but shows that you probably wont get a better W/GHs lowering it, without hardware changes.

Just wondering if anyone has tried here to undervolt/underclock their S2 yet?

Not sure if the S2 has been done, but a lot of people have done the "Pencil Mod" on the S1, as is said above you have a 40 resistors to modify so a big job. If you search on S1 & pencil Mod you will find plenty of information.

Rich

I did but the layout is slightly different and there is no schematic online.

Right so I have found the correct resistor and managed to get it downvolted to 0.70-0.71Volts from the original 0.75 V.

However its very hard to get it to run stable. At 196 Mhz half of the chips were getting X's.

So I downclocked the frequency to 163Mhz, it hashes but getting lots of rejections and 2% HW errors.

Well done on finding the resistors. In some ways it's quite simple... If the voltage is too low for the frequency you will get too Many HW errors. So you either turn the frequency down some more until they become acceptable or you have to use an eraser on the resistors and redo them with less pencil and a higher voltage.

Have you just done a single board as an experiment or all of them? Also worth rechecking all the voltages if just one of them is too low than that one can cause the HW errors.

Rich

Mind taking a pic?  I have a couple S2s here that I might tinker with...I can use a soldering iron but i'm not sure what to touch with it :)

Yes I did the pencil mod however this is a pain in the ass because the resistors you are looking for are on the opposite end of the hashboard, meaning you have to pull the heatsink off each and every hash board. There are 4 resistors per board you need to mod, so 40 resistors in total. But the grunt work is taking all those screws off the heatsinks. It helps if you have an electric drill with the right hex socket.


If you find a high res photo of the rear end of the S2 blade, I can show you which resistor to pencil. I already put my heatsinks back on.

The stock resistor is around 2.1KOhm, I got it down to 1.5KOhm and the voltage is around 0.70V or so from the stock 0.75V.

However I need to find a stable clock speed because I am getting huge errors.

Yes I only did one board for now because its alot of work.

I checked the voltages and they were like 0.70,0.70,0.71,0.71 or so for each of the buck converter outputs so its close enough.

Alright I found a photo posted online somewhere and outlined. Either you want to overvolt or undervolt these units.

http://i61.tinypic.com/2hxcdxz.jpg

Just looking back at my S1 notes my best results were at 0.76V. Remember that the standard voltage is 0.85V and the BM1380 spec sheet only goes to 0.75V, so  it's just possible that going to 0.7V is pushing it too far?

Rich

At 0.75V what clockrate were you running at ?

At 0.76V 225MHz

Rich

With the S2, you can't run it at 225Mhz without it getting HW errors. The most you can run it at is around 212Mhz.

196Mhz is the stock rate for the S2.

So I am surprised the 163Mhz gives errors at 0.70Volts.

Alright I got it stable at 162 Frequency.

I need help figuring out the efficency since I only modded one blade.

AT THE WALL WITH AN UNMODIFIED S2 BLADE I GET 101WATTS hashing at 82Gh/s

AT THE WALL WITH THE UNDERVOLTED S2 BLADE I GET 88WATTS hashing at 82Gh/s


So taking into account the fans, PSU, controller, how would one calculate if its more efficient or not.

Since the unmodified blade uses 1W=1GHS it means that 82Watts goes to the blade and the other (101-82=19Watts) are the fans so subtrading 19 Watts from 88 Watts leads to ~69 Watts which means at 82Gh/s it consumes 69 Watts of power which is ~0.84W/GHS meaning an Antminer S3 efficiency.

Neat, thanks.  I may have to try this out :)

I think that sounds about right.  Refresh my memory, the controller has a separate power input from the backplane, right?  If so you could just unplug the fans and power the controller from another source, which would leave you powering nothing but the blade on the internal PSU, then you could see for sure what the difference looks like by swapping between an undervolted and non undervolted blade.

How would I power the controller from another board? I don't think I can.

I reran the test with all 5 fans disconnected.

AT THE WALL WITH AN UNMODIFIED S2 BLADE I GET 88 WATTS hashing at 82Gh/s

AT THE WALL WITH THE UNDERVOLTED S2 BLADE I GET 78 WATTS hashing at 82GH/s

So assuming 1W=1GHS it means that the PSU+Controller are using only 6 watts. So the undervolted blade is using 72 Watts.

72 Watts at 82Gh/s means 0.878W/GHS.

Hard to tell if this is correct until I get all 10 blades undervolted which can take hours.

Doesn't the controller get 5V separately from the backplane that it's plugged into?  If so, just feed it 5V from another source.  I doubt it's using much anyway, honestly.  I think those numbers make sense, the fan load can be variable, I bet the controller is pretty constant.

I think the controller gets +12V from the same area as the 10 other blades, basically from the PCIe looking female end.

Oh, never mind then.

Seems to also run at 175Mhz putting out 88Gh/s at the same efficiency.

SUCCESS!!!

Finally finished doing all 10 blades on one of the miners.

@ 162 Mhz frequency


STOCK VOLTAGE 820GHash/s = 867 WATTS = 1.07W/GHS

UNDERVOLT 0.70V 820Ghash/s = 729 WATTS = 0.889W/GHS


-----------------------------

Tomorrow I will try tweaking and see if I can run it at a higher clockrate. The miner is very quiet now. And runs exactly the same efficency as my Antminer S3.

Nice job, you just need to do the S3 now.  :)

Here's what I managed to get my S3 to.


The R Ohm is the value of the resistor soldered in parallel with the existing VFB resistor.

Rich

Yes I might as well mod it to make it quieter.

What is the stock voltage for the S3? 0.75V ?

UPDATE!

Had more time to play around with these miners. Basically the ASIC CHIP cannot run at all under 0.70 Volts. I tried 0.67 Volts and its impossible to make it run under any frequency without getting huge hardware errors. 0.70-.71 Volts is the limit on these chips.

I have managed to get it to run on 181Mhz frequency and its very stable, no "X", and gets like 0.004% hardware errors. It hashes around 915Gh/s and uses around 830Watts of power at the wall. So 0.90Ghs/W and the stock would of been 1.1Ghs/W using the same hashrate.

Runs cooler and will pro-long the life of the stock Enermax 1000W PSU.

That's not bad at all.

Sorry for writing to that almost three years later...
But what is the hex value for 181MHz? And what timeout did you use? ("bitmain-freq" : „40:175:0681“,)

thanks in advance

freeapp