Antminer S5 - Underclock - Undervolt - Best J/GH

[RichBC]

Not what has been reported so far, but my first tests show that reducing the Supply Voltage to an S5 does work.....

Have just taken delivery of a 2nd hand S5 and after some quick initial checks that everything was ok had a quick go at dropping the volts.

First, very early impression, is that things work as expected...

Tested at 125MHz with 9.0V at the connectors.

(Not wanting to tell people how to suck eggs, but perhaps those that are bragging about how much current they can push through a Molex connector, you need to measure the voltage that is actually getting to the board)

Only measured over a 30 Minute period the Hash was 440GH/s (a bit lower than expected, should be 470GH/s . Over the short time the rate was fluctuating quite a lot and needs further investigation)

Current from the supply (not the wall) was 12.1 Amps so 109 Watts

So that gives 109/440 = J/GH 0.248  Smiley

No xxx, HW error a bit higher than I would like to see @ 0.05%

Boots up and starts hashing at 9V, although to be fair it's 10V with no load because of the rubbish leads I am using on the adjustable PSU.

So a lot more tests to run but a better than expected start. Is it my S5, am I lucky, am I doing something wrong? Time and more measurements will tell.

Rich

[OgNasty]

Not what has been reported so far, but my first tests show that reducing the Supply Voltage to an S5 does work.....

Have just taken delivery of a 2nd hand S5 and after some quick initial checks that everything was ok had a quick go at dropping the volts.

First, very early impression, is that things work as expected...

Tested at 125MHz with 9.0V at the connectors.

(Not wanting to tell people how to suck eggs, but perhaps those that are bragging about how much current they can push through a Molex connector, you need to measure the voltage that is actually getting to the board)

Only measured over a 30 Minute period the Hash was 440GH/s (a bit lower than expected, should be 470GH/s . Over the short time the rate was fluctuating quite a lot and needs further investigation)

Current from the supply (not the wall) was 12.1 Amps so 109 Watts

So that gives 109/440 = J/GH 0.248  Smiley

No xxx, HW error a bit higher than I would like to see @ 0.05%

Boots up and starts hashing at 9V, although to be fair it's 10V with no load because of the rubbish leads I am using on the adjustable PSU.

So a lot more tests to run but a better than expected start. Is it my S5, am I lucky, am I doing something wrong? Time and more measurements will tell.

Rich

How did you go about reducing the voltage?

[RichBC]

I think MrTeal reported the 11V reliability floor, and I trust him to know what he's doing. What board version is that? I only have access to some partly-working V1.3 PCBs with six level-shifter diodes, but I've seen some others (not sure what PCB rev) with only five. I haven't checked what the difference is.

Also, I'm really surprised you got 0.25W/GH out of it with the extra fan and controller current.

Yes in the last couple of Weeks while getting the best I could from an S3 I read everything I could find on dropping the voltage to an S5 and it did not look too hopeful. So yes I was very surprised at the result. Having let it run a bit longer I think the Hash rate is a bit lower than I reported, closer to 410GH, but still a good result and still running after 3 Hours.

I have a 40 Amp 3-15V variable PSU, not efficient enough for mining, but fine for testing.

Rich

[RichBC]

I have tried  a couple of Software reboots and a complete power down and restart at 9V and everything was fine, have not tried starting with the miner completely cold.

As I was not going to have to listen to the noise I adjusted the Supply voltage and clock up a bit for an overnight run. Ran for 8 Hours with 7 HW Errors. Here were the settings.

Frequency 225MHz
Supply Voltage 10.6V
Current from Supply 24.5A
Power 260W
Average Hash 735GH/s
0.353 J/GH

So again about right looking at the BM1384 data sheet.

Will make some more measurements Today, but need also to set to on seeing if I can adjust a Server PSU to a lower voltage?

Rich

[RichBC]

Have done a lot more measurements at different frequencies & voltages and have got to say that testing an S5 on the bench at the higher frequencies is horrendous and has being doing my head in...

However at the lower frequencies it's quite bearable. 150MHz, 9.5V, Fan 1920rpm & 32 Deg. 

Interesting to compare it to a typical overclocked S3 @ 0.5TH

Code:

	MHz	W	GH	J/GH
S3	237	394	500	0.788
S5	150	143	500	0.285

Rich

[MarkAz]

If you're interested, I wrote some programs to specifically do power monitoring and tests on the S5 - one program does realtime logging to a CSV file from the Watts Up Pro Meter (1s resolution), the other one runs the S5 through all the clock speeds, from the lowest all the way up to 450, at each clock speed it runs for 15 minutes to let it stabilize.  I have one more app that connects to CGMiner and logs all of it's data (5s resolution), so you can see all of the characteristics and what they do.  The full test takes about 12 hours to run on a machine, but my objective was to be able to build machine profiles, and see what kind of variation I had among my S5's.

Lately I've been focused on the A2's, so haven't given the S5's as much TLC as I normally do...

[RichBC]

Monitoring software sounds great, could save a lot of time listening to the S5... Just need an extension to cope with varying the supply voltage. 

Rich

[MarkAz]

Monitoring software sounds great, could save a lot of time listening to the S5... Just need an extension to cope with varying the supply voltage. 

How do you want to interact with it?  Right now it's part of a larger program I'm writing, but I could easily make a commandline one that you'd just specify the IP/port and it would just start logging - does that work for you?  It's Windows-based also, FWIW.

[RichBC]

Thanks for the offer but at the moment I do not have a PSU that is programmable for voltage. If I can come up with something it would be a nice idea. However I would like to try your logging program as would definitely help in the testing and measuring I am doing at the moment.

Rich

[MarkAz]

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...

[RichBC]

Hi

Thanks for the links. I have a 40A variable, 5 to 15V PSU which is fine for testing except that I have to twiddle a knob.  However this PSU is not efficient enough for Mining so I need to set to and see how much I can vary the voltage on one of the Server PSU's that I have. I think they are by far the best value for money, performance & relaibility you can get. If I can do that with a variable resistor somewhere than that can be replaced with a Digital Pot controlled by an Arduino and we then have programmable voltage.  Just need to find the time in amongst all the other things i am trying to do...

Rich

[MarkAz]

I've made a commandline program (Windows, 64bit) to do the logging for you - you can grab it here:

http://www.analogx.com/files/cglog.zip

The usage is pretty straight-forward:

cglog /ip 192.168.3.226 /device "Antminer S5" /file "log.csv" /verbose

For /device you can specify either "Antminer S5" or "A2 Mega", since those are the only two types of miners I have, they're what it supports.  While it's running, it will output this:

1:[PoolUsername][372]
Hash:1059.83  Errors:0  Temp1:54  Temp2:54  Freq:350  Shares:A145/0

The first line is the Pool 1 Username and how long the device has been running - the second line just gives some values that I'm usually interested in when running tests.  The meat and potatoes are in the "log.csv" file, where I basically save most everything:

Date,Time,elapsed,ghs 5s,ghs av,hardware percent,found blocks,total mh,local work,utility,work utility,frequency,fan1,fan2,temp1,temp2,chain_acs1,chain_acs2,p1 getworks,p1 accepted,p1 rejected,p1 discarded,p1 stale,p1 get failures,p1 remote failures,p1 last share time,p1 diff,p1 diff1 shares,p1 difficulty accepted,p1 difficulty rejected,p1 difficulty stale,p1 last share difficulty
2015-08-14,23:32:07,439,1238.65,1133.38,0,0,497528533.0,300347,22.96,15875.86,350,2880,0,54,55,oooooooo oooooooo oooooooo oooooo ,oooooooo oooooooo oooooooo oooooo ,11,168,0,172,0,0,0,0:00:00,512,116152,121854.140625,0.0,0.0,511.992188

It uses the CGMiner JSON interface, so it might work with other miners that run with it.  Once you run it, it will continue running until you hit a key, then it will shutdown and save everything.  It will not erase the file output, it will append to it - so you can run it multiple times and not have to worry about renaming or deleting the csv file.

Anyway, let me know if you run into any issues or have anything you'd like for me to add, or logging data I might have missed.

[RichBC]

Thanks very much. I have grabbed it and will give it a go when I get a chance, busy Weekend here as it's my Grandsons Christening. Don't use command lines much these days   but I guess it will all come back...

Do I need to set anything in the S5 cgminer for it to accept the command line?
Have you written anything to explore / display the info in the log file?


Rich

[MarkAz]

Thanks very much. I have grabbed it and will give it a go when I get a chance, busy Weekend here as it's my Grandsons Christening. Don't use command lines much these days   but I guess it will all come back...

heheh, I included a batch file that has examples of the commands, so if you just edit that and double click on it, it should work fine for you - the only thing you'll probably need to do is change the IP address of the machine, then you'll be good to go.

Do I need to set anything in the S5 cgminer for it to accept the command line?
Have you written anything to explore / display the info in the log file?

I haven't written anything yet to specifically do that, but I have been working on an app to do updates via SCP of the image I made for the A2 Terminator's, so I could make something that gets it via that.  Is the log that useful for what you're doing?  TBH I haven't even looked through it, although most of the OS hacking I've done so far is just on the A2...

[philipma1957]

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

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

If your info is true no one needs any new gear other then the s-5.


If you can do this a guy with 100 s-5's would be able to drop his power bill more then 50%

I am stunned your thread has not attracted more interest.

Not what has been reported so far, but my first tests show that reducing the Supply Voltage to an S5 does work.....

Have just taken delivery of a 2nd hand S5 and after some quick initial checks that everything was ok had a quick go at dropping the volts.

First, very early impression, is that things work as expected...

Tested at 125MHz with 9.0V at the connectors.

(Not wanting to tell people how to suck eggs, but perhaps those that are bragging about how much current they can push through a Molex connector, you need to measure the voltage that is actually getting to the board)

Only measured over a 30 Minute period the Hash was 440GH/s (a bit lower than expected, should be 470GH/s . Over the short time the rate was fluctuating quite a lot and needs further investigation)

Current from the supply (not the wall) was 12.1 Amps so 109 Watts

So that gives 109/440 = J/GH 0.248  Smiley

No xxx, HW error a bit higher than I would like to see @ 0.05%

Boots up and starts hashing at 9V, although to be fair it's 10V with no load because of the rubbish leads I am using on the adjustable PSU.

So a lot more tests to run but a better than expected start. Is it my S5, am I lucky, am I doing something wrong? Time and more measurements will tell.

Rich

[philipma1957]

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

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

If your info is true no one needs any new gear other then the s-5.


If you can do this a guy with 100 s-5's would be able to drop his power bill more then 50%

I am stunned your thread has not attracted more interest.


now this is a link to a convertor.

http://www.amazon.com/DROK-Converter-Transformer-Synchronous-Adjustable/dp/B00C9UUFHC/ref=sr_1_2?s=electronics&ie=UTF8&qid=1439734487&sr=1-2&keywords=drok++converter+step+down

1 of these would cost you 15 dollars and provide about  15 amps x 10 volts = 150 watts 

2 of these would cost you 30 dollars and provide about 30 amps x 10 volts = 300 watts

if you do freq 125

2 of these would  really make your s-5 power efficient.

A grand project would be buy an s-5+ and 18 of these

If successful   you could run the 7.7th beast at 4th and about 1200 watts

Not what has been reported so far, but my first tests show that reducing the Supply Voltage to an S5 does work.....

Have just taken delivery of a 2nd hand S5 and after some quick initial checks that everything was ok had a quick go at dropping the volts.

First, very early impression, is that things work as expected...

Tested at 125MHz with 9.0V at the connectors.

(Not wanting to tell people how to suck eggs, but perhaps those that are bragging about how much current they can push through a Molex connector, you need to measure the voltage that is actually getting to the board)

Only measured over a 30 Minute period the Hash was 440GH/s (a bit lower than expected, should be 470GH/s . Over the short time the rate was fluctuating quite a lot and needs further investigation)

Current from the supply (not the wall) was 12.1 Amps so 109 Watts

So that gives 109/440 = J/GH 0.248  Smiley

No xxx, HW error a bit higher than I would like to see @ 0.05%

Boots up and starts hashing at 9V, although to be fair it's 10V with no load because of the rubbish leads I am using on the adjustable PSU.

So a lot more tests to run but a better than expected start. Is it my S5, am I lucky, am I doing something wrong? Time and more measurements will tell.

Rich

[RichBC]

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

[philipma1957]

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

Well look at it from my viewpoint.  My power cost is high.


 I can get those .95% efficient buck transformers I linked to you.  they would take my 93% plat and drop it to 88% which is solid gold rating.

If I get 200 watts and 610 gh   for an s-5 it is great.

I could run 2 s-5's at 1210gh  use 400 watts.

I now need to decide do I want to trust you and spend  4 x 15 = 60 dollars for the dc to dc  and 700 for 2 s-5's.

 I am 760 out of pocket if you are wrong.

Now it could be some units do what you claim So I have to hope I get the right version.

So lets say I do.

  the upside is  really quiet s-5's running at .33 or .35 watts a gh.

I have lots of  power supplies. I do not need to buy one.

I just don't feel like doing all this testing  based on your word alone.

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

Now I have those buck converters listed. 


 Here is what I will do anyone with good long rep here  I will buy 2 converters and drop ship them to you so you can test your s-5 and see if you can get  these good numbers.

2 convertors would give you 150 watts each so your s-5 can run at 10volts at freq 200 and do around 600gh  which would be 200/600 = .333 watts a gh

[philipma1957]

So here is a screen shot of the two convertors each can go down to 10 volts at 15 amps and supply 150 watts.

so the s-5 will be able to get 300 watts  which can run freq 200   using 206 watts and hashing at 665 according to your graph.

[RichBC]

Phil I think you are being a little unfair, I am not asking you to trust me or to go and buy anything, and even if you were to experiment why not just buy one converter?  You are trying to push this along too quickly. So while I work on the proof, a couple of points to ponder on.

1) The data in the BM1380 and BM1382 data sheets, Core Voltage & GH/s and J/GH has been shown to be correct so why doubt what is possible with the BM1384?

2) sidehack has shown similar efficiency is possible.

3) This is not actually that relevant to most people at the moment. It is to me because of my electricity cost 0.943 Pence / KWH (14.7 Cents) This is the pivot point for an S5. At current difficulty and 15 Cents you are best profit wise to run an S5 flat out. If your electricity cost is lower than this then Overclock. If your electricity cost is higher then Underclock.

4) I had expected this to be a long haul and had prepared the following post as an introduction to the project. I then expected to have a string of posts over a period of time explaining my progress, successes and failures.

The Project

My target for a complete system is total hash of about 600GH/S with 200 Watts at the wall this would be 0.34J/GH. Is this achievable? Almost certainly not but it's always good to have a target in mind, and will be disappointed if something under 0.4J/GH cannot be achieved.

My hope is that on the release of the S6/7 that the S5 might come down in price on the 2nd hand market, as it seems to be holding up quite well at the moment . Then I could have 4 x S5 running from a Platinum Efficiency Intel Server PSU DPS-1200TB.

 http://www.plugloadsolutions.com/psu_reports/INTEL%20CORPORATION_DPS-1200TB%20A_1200W_SO-379_Report.pdf

Which I happen to have a couple of.


So I have decided to stick my neck out here, this project could succeed or fall flat on it's face, at the moment I do not even own an S5, one is on the way from ebay.

Please feel free to add anything that you think might be useful, comments, questions ideas, other information that you have found or read. I will post up progress good or bad as it develops.

This was then going to followed with.

Possibilities for how to improve the S5 J/GH

1) Find the problem that is preventing the S5 Voltage to below 10V. I am going to look at this one first, with the first line of attack to be following up on Sidehack's theory that the problem is in the interstage level shifters.

If this is successful then the next step will be to find efficient 9V PSU's. My current thought on this is to find a way of adjusting a server PSU.

2) Do some cutting and pasting of the BM1384 string adding additional device pairs, probably between 3 & 5 to the chain to reduce the voltage. If you could add 5 Device pairs, which is theoretically possible, then with 2 S5's and 4 Boards you would take 3 Boards & cut and paste 5 pairs from the 4th board to each of the other 3 giving 3 lots of 20 chip pair chains.

3) Convert the S5 back to a conventional not string system adding adjustable VRM's.

However I posted none of this because I switched on the S5 turned the volts down, made some measurements and posted them.


I am now working on why it works however this is hampered by a few things.

1) Most of my effort, in the time I have for mining, has gone it to making a set of measurements to validate what can be achieved

2) I only have a single example of an S5.

3) Working with the string design is not very nice. You do not have a nice common ground and the ability to easily make measurements and to hook the Logic Analyser up to multiple points. You have to be very careful what you are earthed to at any point in time otherwise there will be a big bang....

Finally for what they are worth a couple of pictures.



Here is cgminer showing the run at 175MHz



And here is a shot of the PSU on the same run. 10.0V on the PSU but then 9.8V on the meter measured on the board and 17.8 Amps being taken.

Best I can offer at the moment....


BTW testing with those converters has no value at the moment. They will almost certainly be fine with my S5. The key issue is understanding why this works which is what I am going to work on next.

Rich