Antminer S5 - Underclock - Undervolt - Best J/GH

[philipma1957]

Thank you


  my reasoning is this .  You may have stumbled on an improvement in :

A) firmware
B) pcb boards

I pushed hard because I wanted you to be correct.

Now that I see photo proof.  I don't mind buying gear to test out your findings.


I am like you in that my power is costly  so for now running s-5's  that do .51 watts a gh does not work well for me.

But if you have discover something that can be duplicated on other s-5's

Even if only a certain pcb with a certain firmware.  You have found a great secret and shared it to the rest of us. For that you did very well by all of us.

Would not it be nice if 20 to 30 percent of s-5's can clock as you claim?



your gui and meters are showing about 180 watts dc  to  get 578 gh   this is   180/578 = 0.311 watts a gh.

I would set up an s-5  right away based on this.

[MarkAz]

philipma1957, while it's cool you're making the offer, personally I don't think it matters at this point... He's still in the process of testing things out, and has been posting more updates than soliciting anything from anyone.  I don't think there's some immediate rush that would necessitate running out and verifying his results quite yet - and once he's nailed down things, I'll be more than happy to buy the converters myself and test on one of my rigs.  I'd rather see you buy them for him at that point, so the person spending the time figuring things out doesn't have to eat all the costs, instead of just some random guy who verifies his result.

I also think RichBC is spot on with his comment about the obsession primarily being hash rate, and while people get excited about efficiency, when you tell them their 1.1TH would be running doing 400GH, their interest wains.  

And to RichBC, love the work - and welcome to the community, but I think to philipma1957's point, this space is unfortunately filled with scammers and most people have gotten their hopes and dreams smashed against the rocks of reality on more than one occasion on here.  The nice thing is that once you're a known quantity, the community is very supportive and active - look at Sidehack or some of the other hardware threads out there and you'll see what I mean.

Either way, keep up the good work!

[philipma1957]

I am going to  rescind my offer to ship those converters out.  I am going to order my own s-5 and try to duplicate the op's results.

@ op  what was your firmware and board codes?

[MarkAz]

I would set up an s-5  right away based on this.

I'm totally with you - I would love to run a bunch of my S5's at greater efficiency vs greater speed.   It also means that you'd have a longer lifespan for the device, as difficulty raises you can slow it down and boost up efficiency.  I'm not a huge fan of the step-downs you found, but for testing things out they'd be fine... I think the ultimate solution would be to pick some decent power supply, and reverse engineer it enough the change the 12V output to some alternative fixed amount.

But the big question is probably less about whether or not the device can run for some period of time at a lower voltage - but how stably it does.  In the screenshot you can see he's got some hardware errors - and maybe his S5 has a bunch running normally, but to me that's reminiscent of overclocking - ie; the closer to the edge you get, the more errors you encounter.

The only other thing to consider also would be whether or not to pull the BB/Controller off of this voltage adjusted input - I can't think there's a reason you couldn't run them at 12V and just undervolt the blades.  The BB has a voltage regulator already in it, so it should be pretty tolerant in either case.

Ah, one other thought - are you running the fan off of this?  I notice the RPM is 2040, so one thing that could skew your numbers would be variations in what the fan pulls.  Personally when I do power calcs, I usually put the fans on a separate PSU just so I can focus on the device, and then run the fans at full so I know the worse case fan draw.

[RichBC]

But the big question is probably less about whether or not the device can run for some period of time at a lower voltage - but how stably it does.  In the screenshot you can see he's got some hardware errors - and maybe his S5 has a bunch running normally, but to me that's reminiscent of overclocking - ie; the closer to the edge you get, the more errors you encounter.

The only other thing to consider also would be whether or not to pull the BB/Controller off of this voltage adjusted input - I can't think there's a reason you couldn't run them at 12V and just undervolt the blades.  The BB has a voltage regulator already in it, so it should be pretty tolerant in either case.

Ah, one other thought - are you running the fan off of this?  I notice the RPM is 2040, so one thing that could skew your numbers would be variations in what the fan pulls.  Personally when I do power calcs, I usually put the fans on a separate PSU just so I can focus on the device, and then run the fans at full so I know the worse case fan draw.

Just a comment on these points. The HW errors are sort of deliberate. The S5 seem to very smoothly go from no errors to a few to a lot as you wind the volts down at any given frequency. So for each of the tests I adjusted it to be just into errors as this was an easy way of knowing I was on the edge. In practice you might want to add 0.1V to each of the levels, and they may well be differences between machines and on mine there is definitely a difference between boards of a bout 0.2V.

I think the BB/Controller is absolutely fine at 10V as it's ok at 9V, but definitely on the list of things to check.

I left the fan connected as normal as this seemed a fairer test, and if it was oscillating between 2 speeds took the current at the higher one. At this stage to be honest I was not after the last decimal point of accuracy, this was about does it work....

Rich

[GigaBit]

well your info is kind of too good to be true.

How about a few photos and screen shots to prove it?

Gotta say if there is one thing about my limited time mining I don't like, it's the lack of trust that there is in this community.... If I wanted to scam people I can just as easy edit some screen shots, but why would I, what's in it for me? I will try and do what you ask for but if you think about it there is nothing that I can post up that will actually prove what I am saying...

Here for what it's worth... is the complete table of measurements I have made.

Code:

MHz	V Core	V	A	W	GH	J/GH
350	0.77	11.5	41.5	477	1150	0.415
300	0.73	11	33.9	373	980	0.381
275	0.71	10.7	30.3	324	910	0.356
250	0.69	10.4	26.7	278	830	0.335
225	0.67	10.1	23.5	237	740	0.321
200	0.67	10	20.6	206	665	0.310
175	0.65	9.8	17.8	174	575	0.303
150	0.63	9.5	15	143	500	0.285
125	0.62	9.25	12.3	114	415	0.274
100	0.60	9	9.9	89	330	0.270

A couple of notes on the table.
1)They are measured in amps & volts output from the adjustable PSU, so are not At the Wall, you need to add 5-10% dependant on your PSU efficiency.
2) One of my hash boards is prone to dropping out below 9.25 volts so the data at 100Mz is not sustainable, the best efficiency is in practice at 125MHz,

I agree BTW with your comment on lack of interest. I put it down to my newness and most peoples obsession with overclocking....

Rich

Awesome!! 
I needed this, was thinking of a way to put 3 S5's on 1 breaker... just happens I'll need to lower 1 out of 3 to make it happening.
This kind of detailed listing, in this type of business should be given by Bitmain themselves... but some people have equipment, so amen.
Now I can buy more miners and increase my hashing power.... price of Bitcoin right now, the only thing worth spending on in hardware.
With this I'll be able to add an extra 3TH.s for sure, if not 4!!
Now with the S5+ that will push back the release of any new processes, not a bad idea to further invest in used S5's...

Might not be 100% accurate but it's a good rule of thumb to start with.

[QuintLeo]

Have you seen this:

http://www.electronics-tutorials.ws/blog/variable-voltage-power-supply.html

Might be worth hacking together to give you some more flexibility...

 78xx and LM317T are both fairly low current limit - 1 AMP for the 78xx series, I think 5 Amps but might be a lot less for the LM317, can't remember for sure since I've not worked with those in a while.

 Both are also VERY inefficient.

 It's possible to use an external pass transistor with them to up the current limit, but you're still dealing with a LINEAR regulator, and therefore VERY INEFFICIENT.


 Finally found my guestimates - in the RBox upgrade thread, not the thread I thought I posted them in.

 S5+ turned out to be a bit more efficient than my 16-per-string guess, but looking back at that guess I think I misfigured a bit - it should have been better efficiency that a stock S5.


 The figures posted for .6 are higher on GH per chip, but not by a lot, and pretty close to dead on for power-per-chip between my guestimate and the "9v" figure posted.
 I don't see ANY reason to doubt the posted "as measured on a S5" numbers, they're right in the ballpark of where they SHOULD be based on the published specs for both the S5 itself AND the BM1384.



 The real trick now is to either figure out how to turn the +12V on a low-cost EFFICIENT power supply down to 9-10V range, or find a higer amperage capability converter that can do 12V to 9-10V range, or find a high amperage PS that is high efficiency in the 9-10V output range.

[QuintLeo]

I needed this, was thinking of a way to put 3 S5's on 1 breaker... just happens I'll need to lower 1 out of 3 to make it happening.

 Why not split them out a bit, should be able to run 5 easily split between 2 circuits. 2 one one circuit, 2 on the other, 1 split between the 2, without pushing a 15A 117VAC circuit particularly close to it's limits.

[philipma1957]

Have you seen this:

http://www.electronics-tutorials.ws/blog/variable-voltage-power-supply.html

Might be worth hacking together to give you some more flexibility...

 78xx and LM317T are both fairly low current limit - 1 AMP for the 78xx series, I think 5 Amps but might be a lot less for the LM317, can't remember for sure since I've not worked with those in a while.

 Both are also VERY inefficient.

 It's possible to use an external pass transistor with them to up the current limit, but you're still dealing with a LINEAR regulator, and therefore VERY INEFFICIENT.


 Finally found my guestimates - in the RBox upgrade thread, not the thread I thought I posted them in.

 S5+ turned out to be a bit more efficient than my 16-per-string guess, but looking back at that guess I think I misfigured a bit - it should have been better efficiency that a stock S5.


 The figures posted for .6 are higher on GH per chip, but not by a lot, and pretty close to dead on for power-per-chip between my guestimate and the "9v" figure posted.
 I don't see ANY reason to doubt the posted "as measured on a S5" numbers, they're right in the ballpark of where they SHOULD be based on the published specs for both the S5 itself AND the BM1384.



 The real trick now is to either figure out how to turn the +12V on a low-cost EFFICIENT power supply down to 9-10V range, or find a higer amperage capability converter that can do 12V to 9-10V range, or find a high amperage PS that is high efficiency in the 9-10V output range.

the buck step down I linked is 95% and can handle 15ampa at 10 volts.  so 2 for each s-5 is about 30 dollars

doing better then 95% is not going to be easy.  and 95 percent of a 93 percent platinum psu is a 88 percent psu.

my method will work but costs too much its would be 30 bucks an s-5 and add a few more for cables .  so close to 40 an s-5

[RichBC]

I am slightly annoyed with these comments, but I guess I am new so have to put up with it...

well your info is kind of too good to be true.

How about a few photos and screen shots to prove it?

No screen shots of a gui really look bad on your part.

And that...

You may have stumbled on an improvement in :

A) firmware
B) pcb boards

I did not stumble on it. I read the data sheets, looked at what sidehack has done and decided to find a way of making an S5 perform at a lower voltage. 

All I have tried to do in this thread is to share the information that I had found, albeit in an incomplete state, not to be taken to task over it's validity.

MarkAz thanks for your support, now let's move on.... 

I was wishing in some ways that I had just kept this to myself until I had done all the work to understand what the problem was and how to fix it, but thinking on it overnight this project is going to go ahead much quicker with some help.

My best theory at the moment is a simple one, that S5 working at a lower voltage is Hash Board specific, with changes being made in later versions which then make it possible. I am assuming that tests done by others were on earlier versions of the Hash board? They did not add the version to their posts, so I cannot prove that.

I can't afford to go out and buy a version of each of the revisions of the S5 in order to test and prove which versions work at a lower voltage and which do not. So I am collecting all the paper research I have done and my guess at what the problem / key change was, and will share as soon as I have put it all together.

Rich

[QuintLeo]

I'd prefer something that would go AC line -> 9-10V range with 90%+ efficiency, not ignoring your bucks but they are extra bucks.
I'm also not convinced they're good for 15A in a mining setup.

Output current: 0-15A(10A for long-term work)

 A couple of the other converters they list mention higher power output with a heatsink upgrade, though. Would have to look into the details.


 Might be worth looking at next summer after the halfing as a way to keep my S5s hashing at a profit for a while longer (don't need them on my SP20E of course), but I'd prefer to figure out how to adjust the voltage down on my existing power supplies.

[RichBC]

I have collected together a set of pictures of all the S5 hash board revisions I can find, may or not be complete, but it's enough to be getting on with. Here are the pictures along with my initial comments on the changes. There will at this stage almost certainly be errors & omissions, this has all been done in a quite short period of time.

V1.1   2014-12-03




First release. Easily identified by the large cleared copper areas.

VDD decoupling components on the front of the board

Only 5 of the 6 possible level shifters placed

16 Pin connector to controller board

PCI-e 6 pin power connectors close together in S3 compatible position

Single oscillator module Y1 on the ground level chip.

Credit to dogie for a decent high res pictures, wish the others were as good.


V1.3   2014-12-15




Move to 18 pin connector to controller board. (Have not investigated what the additional signals are for?)

PCI-e 6 pin Power connector spacing widened, not S3 case compatible any more.

All 6 level shifters fitted along with additional SMT chip marked G07F. I think this is a SN74LVCG07 Buffer / Driver chip. Possibly / probably associated with the additional data connections from the Controller board?

Single oscillator module Y1 on the ground level chip.



V1.5   2014-12-20



No immediately obvious difference to V1.3 but have not examined that closely.



V1.7   2014-12-22



VDD decoupling components moved to reverse of the board. I have assumed they have been moved to the reverse as I do not have a picture of that side of the board.

The cleared copper area on the reverse of the chips is now a clear rectangle. I could guess clearing the way to easily fit a mini heatsink that comes later.



V1.9   2015-01-08



Board layout changed to make provision for additional oscillator modules. Y1 to Y5. See V1.91 for more info.

Also provision for Oscillator modules on all stages Y1 to Y16.




V1.91   2015-03-28



Additional oscillator modules Y1 to Y5 fitted. The level shifting chain is broken at each of the points that an additional oscillators are fitted. Each oscillator drives 4 pairs of chips, and the last on in the chain 3 pairs.

Addition. The LDO regulators on all but the last stage are removed. Voltage for the IO & PLL is now taken directly from the Core Voltage of the stage above.
This is a significant change & means that the IO & PLL voltage will track up and down in proportion to the supply voltage. Also it might have been possible (Have not measured) in the earlier boards at lower supply voltages for the LDO, which took it's supply from the core voltage 3 stages up, to drop out of regulation.


V1.91 2015-03-28 (With Mini Heatsinks)




Another version of V1.91 built with mini heatsinks fitted. Look the same as the ones fitted to S5+, so perhaps a trial? Do not know why they, on all the ones I have seen, did not fit heatsink to U20 & U29?



So there are 2 "big" changes that have taken place. First was on the transition from V1.1 to V1.3.The 6th level shifter was brought into play along with it's Driver Chip. The change from 16 to 18 connections to the Controller Board and the other Controller board changes was also made at this time and may be associated with this?

This could be the key change, but have no way, at the moment, of confirming that. I suspect it is not, but it would still be useful to understand exactly what this change does?

The second change is the one I am putting my money on being the key enabler of lower voltage working. V1.9 of the PCB made provision for additional oscillators, at V1.91 they were fitted.

The Oscillator count goes from 1 to 5 and the oscillator level shifting chain is broken at each of the points that an additional oscillators are fitted. Each oscillator then drives 4 pairs of chips, and the last on in the chain 3 pairs.

This change is consistent with one of sidehack's theory's that some of the signals are not making their way through the chain if the voltage is too low.

There are also pads on the layout for an oscillator for every pair of chips Y1 to Y16. These additional Oscillators have a different pad layout to the fitted ones Y1 to Y5

My S5 is has the V1.91 with heatsinks and can definitely be undervolted as shown in my previous posts. Most of the testing was with the 20150415 Firmware, but over the last few days have been using  20150715 with no noticeable difference.

I want also to take a look at the LDO regulator(s) for the PLL Analog & Digital power. Silly thing is I can only find one. It's either staring me in the face, on the reverse of the board or not there? I am sure that sidehack could put me right here.

If this all works my favoured solution for powering would be a modified Server PSU. They mostly have provision for a small adjustment 11.7V - 12.7V. I think some can be adjusted more than this, perhaps you could jack the sense input up and reduce the voltage? If anyone already has a solution, or wants to take a look that would be great? I am not that keen on the additional complexity, cost & reduction in efficiency of an additional VRM.

More to come but I have now shared everything I know and if anyone else wants to pitch in trying versions or understanding the changes please do so and let us know what you find?

Rich

[philipma1957]

 so the facts are the early versions tested don't do well at 9 or 10 volts


the newest version did great at 10 volts  even 9.5 volts.

v1.1   2014-12-03
v1.3   2014-12-15
v1.5   2014-12-20
v1.7   2014-12-22
v1.9   2015-01-08



one from below was able to clock low

v1.91 2015-03-28
v1.91 2015-03-28 -With mini heatsinks

so we need the people that tested and got bad results to get back to this thread and let us know what boards they tested.


Like you said it may be multiple units of s-5 can do this.

not just version v1.91

this would let the 'good s-5' run at 650gh 200 watts and with quiet fans.

it may allow  for the s5+ to drop to 6th and use only 2400 watts.

lots of possibilities.

I think I may get an s-5+ and test this out. rather then get an s-5.

P.S. don't be sensitive to my style of writing I have been through a lot on this site and have been pretty blunt.

  I  would think this thread should get interest and it does not get a lot.

 I am making an effort to acquire  some btc and get  the s-5+
I would like to think it can be down clocked to a sane sound /power use via your methods.

[RichBC]

If anyone wants to run a test on their S5. Remember you do not need a high current supply to do it. Only connect one of the hash boards to the supply, the one with the 4 pin connector to the control board, this will halve the current needed for the test.

Set the frequency to 125Mhz. You then just need a supply with a voltage between 9.25 & 10 Volts at 6 Amps and that will be good enough to give a strong indication that the tested S5 will or will not support undervolting.

For a quick test you can probably get away with putting two or three 10 Amp Silicon Diodes in series with your 12V supply. You will need to monitor the voltage getting to the hash board when it is hashing to find the exact voltage it is running at.


Rich


Just to add. The most important board to test at this stage is a V1.9. As far as I can see it's identical to V1.91 but without the additional Clock Modules. If this does not Undervolt almost certainly none of the earlier versions will.

[philipma1957]

If anyone wants to run a test on their S5. Remember you do not need a high current supply to do it. Only connect one of the hash boards to the supply, the one with the 4 pin connector to the control board, this will halve the current needed for the test.

Set the frequency to 125Mhz. You then just need a supply with a voltage between 9.25 & 10 Volts at 6 Amps and that will be good enough to give a strong indication that the tested S5 will or will not support undervolting.

For a quick test you can probably get away with putting two or three 10 Amp Silicon Diodes in series with your 12V supply. You will need to monitor the voltage getting to the hash board when it is hashing to find the exact voltage it is running at.


Rich


Just to add. The most important board to test at this stage is a V1.9. As far as I can see it's identical to V1.91 but without the additional Clock Modules. If this does not Undervolt almost certainly none of the earlier versions will.

your research on this is to be commended.  I really think it would allow some s-5's

v1.9, v1.91  an under clock

I also think it works for the s-5++   if so  the s-5++ becomes a better piece of gear.  For the home miner.

I ordered btc from coinbase it arrives this friday. I will order the s-5++  and test it.

[RichBC]

Phil do remember that the S5 is yet to be understood & the S5+ is completely uncharted. You would be hoping that the S5 solution is something like the multiple Clock Modules and that they carried that design through to the S5+

Rich

[philipma1957]

Phil do remember that the S5 is yet to be understood & the S5+ is completely uncharted. You would be hoping that the S5 solution is something like the multiple Clock Modules and that they carried that design through to the S5+

Rich

yeah but if true it makes the  s-5++ worth quite a bit more.

Start out with whatever your power/sound/heat will bear.  Then clock differently as your prices for power change.

In my case power is 16.9 cents in the summer  add more for ac.  Not that I use ac to cool it but it will warm my house a bit   so I am in the 18 cent a kwatt for power right now.

If it works on an s-5+  I can down clock until my winter power rates come in  13 cents minus a 3 cent benefit for the heat. 10 cent power.

It allows me the s-5+ in house.

 If it does not work  I have a place to host it.  I ordered the coins from coinbase  they come on friday.

[RichBC]

My gut feeling, For what that's worth,  is that it will probably be ok....   You will of course need a lot of amps at the reduced supply voltage, so we will need to crack modifying server PSU's to run at a lower voltage. The good news is that because they have extended the chain length to 16 all of the voltages in the table will be Plus about 0.7V for the same J/GH. So in the order 10V to 11V

Rich

[philipma1957]

My gut feeling, For what that's worth,  is that it will probably be ok....   You will of course need a lot of amps at the reduced supply voltage, so we will need to crack modifying server PSU's to run at a lower voltage. The good news is that because they have extended the chain length to 16 all of the voltages in the table will be Plus about 0.7V for the same J/GH. So in the order 10V to 11V

Rich

I did some quick math on it.  the step downs I found  cost 15 each    9 boards with 3 connectors = 27 stepdowns at 15 is 405 not acceptable to spend that much.

But to test this I can use  1 board run it at freq 200 get   power numbers.     

With 12 volts each board should use  close to 385 watts at freq 325

Once i see what the one board does at freq 200 (my guess is 237 watts as  200/325 x 385 = 237)


then run the one board with 2 step downs which are rated for 150 + 150 = 300 watts   if I go down to 10 volts  10/12 x 237 = 197.5 watts a board

so the full thing  would be 9 x 197.5= 1777 watts  and hash at   4.75 th      or .375 watts a gh.






If all of the above is true  the s-5+ just got better.

[RichBC]

Possible good news on the S5+ design. Have had a look at the low res images of the Hash board front & back and as far as I can see there are multiple clock modules and unless my eyes deceive me I can see the level shifting diode missing from every interstage clock coupling.

So now the IF's  IF I am right on this then there is a clock module for every triple, 16 in total. So IF this is what makes the S5 undervolt it may well carry through to the S5+

Rich