Antminer S5 - Underclock - Undervolt - Best J/GH - page 6.

flikflak

sr. member

Activity: 408

Merit: 259

Right, at the bottom line its more or less the same result, using two psu's or bucks. As far as I know, those bucks are quite efficient (looks like they are using FETs or MOSFETs). The efficiency loweres by its delta 12V-10V = 2V delta is more efficient than a lower voltage like 9V.

Edit: But even with a platinum psu with 94% efficiency and lets assume those bucks are @ 96% (0.94 * 0.96 = 0.902) you will end up with 90% efficiency at all.

Both S5s are at 250 MHz:

100 watts doesnt seem a lot, but it will save you 72KW a month at current speed which is still 7.2 bucks/month with a $ 0.10/KWh plan. I think that money is quite a good investment, I'm gonna ROI those converters in 3.5 months - thanks again!

RichBC

hero member

Activity: 588

Merit: 500

I was initially sceptical about the Bucs when Phil mentioned them, however i have found the efficiency to be very good peaking at 98%. See here for the measurements I made.

https://bitcointalksearch.org/topic/m.12472243

Overall I get about the same efficiency out of a Platinum Server PSU + a Buck as I do from the pair of MeanWell 5V PSU,s which peak at 82%

Rich

Just seen flikflak post, will read it now.

flikflak

sr. member

Activity: 408

Merit: 259

Some more tests with those 2 buck converters @10.0V:

200.0MHz	661 GHs	0.371 J/GH	245W @ wall
212.5MHz	694 GHs	0.376 J/GH	261W @ wall
225.0MHz	738 GHs	0.374 J/GH	276W @ wall
237.5MHz	772 GHs	0.374 J/GH	289W @ wall	HW 0.0007%
250.0MHz	823 GHs	0.369 J/GH	304W @ wall	HW 0.04%	chain1 44°C chain2 41°C

Actually the S5 needs at least one hour to show an valid average speed.

I havn't tested any higher speeds yet, because the wiring of those buckconverters isn't thick enough to allow me more than a drawing of 15amps through it. Unfortunately, the seller is advertising those converters with a capability of more than 15 amps with a fan mounted -> please, dont do it! I measured a diameter of 1.3 mm = 1.5mm², so 15 amps is the absolute maximum!

Maybe the 40a buckconverter is arriving next week... I'll do some more tests with a proper wiring then.

Edit: Edited 237.5 MHz and 250Mhz speeds, running them right now.

1slickvdc

member

Activity: 67

Merit: 10

Hope the bucks are well made or you'll start losing a lot between the PSU and the converters by the time they get to the boards. MW has nicer supplies, but I'm too broke to buy any of those (88-90%+ eff.). That's about half the power consumption I'm getting when I run full bore on 2 hash boards at 425MHz, pulls about 830-850W at the wall. (I get $.075 power at night and weekends, so I switch speeds based on power cost)
Can't wait for this thing to show up though.

RichBC

hero member

Activity: 588

Merit: 500

Looks to be a nice supply, will be interested to hear how you get on? I have been running on my twin 5V setup for some time now with no problems. Am running with 3 Hash Boards connected, at 250MHz, 10V, 0.67V Core, 456W at the Wall, 1237GHz, 0.368J/GH

Reason I am running with 3 Boards, it that the 4th hash board does not perform as well, so I have put it aside to investigate, but have not got round to that yet... I have ordered a couple more of the Buck converters as I think my preferred solution will be a Server PSU feeding multiple Buck Converters, 1 per Hash Board so that the voltages can be individually set to get the best out of each of them.

Rich

1slickvdc

member

Activity: 67

Merit: 10

Ordered a Mean Well SE-600-12, should be adjustable down to 10V, so I can give my S5 a shot when that thing shows up and see what it does to power consumption. I'll measure up how efficient it is on something I can use that I know the load of and what not since I know this will be slightly more efficient than that server supply I'm using at the moment. I'll let y'all know later this week where I'm at once it shows up.

SLEI

member

Activity: 84

Merit: 10

I undervolted one more S5 with parallel DPS-600s and it's happy with 11.3V, 343,75Mhz, 1080Gh/s (avg) and 0,05errors in 20C garage.
I haven't tested used watts yet.
S5 outside in 5-17C with 11,3V don't like speeds over 306Mhz.

Unadjusted DPS-600s give out 12,4V when connected to S1s.

RichBC

hero member

Activity: 588

Merit: 500

Quote from: RichBC on September 21, 2015, 08:22:41 AM

Quote from: VirosaGITS on September 17, 2015, 01:08:35 AM

Quote from: RichBC on September 17, 2015, 12:50:55 AM

As I said the the software is not at all my thing, but I assume there is a file that you can SSH into and edit on the controller board that will have the frequency divider & delay values? Same as there is for the S3?

Rich

Yes. I use WinSCP myself, pretty nice as you can edit everything from your windows computer more normally. Anyhow the json config file is /config/cgminer.conf. Open it and you will see the normal JSON style config.

Relevant line is;
"bitmain-freq" : ""

Example of proper value "bitmain-freq" : "3:393.75:1f06".

Am away from home but took a controller board with me and have just got round to SSH into it. Have found the line you refer to above and as far as i can see the first character "3" is the delay with "3" being around 10mS, this number increases as you drop the frequency and is 10 when you get to 100MHz. Second number "393.75" is the frequency however I think it might just be the text that appears in cgminer? The third Hex number "1f06" is to set up the divider and corresponds to the data in the BM1384 data sheet.

So I think this is where the current selected Frequency and associated setup information for Delay and Divider ratio are stored? So yes editing them will probably allow a different frequency to be set.

However there must also be a file somewhere which has the complete table of frequencies and with access to that an additional Frequency, Delay and Divider Ratio could be inserted which would then allow selection from within cgminer?

Question is does anyone know where this table is?

Rich

Back home so have tried editing the "bitmain-freq" line in cgminer.conf

The 3 fields were as I guessed above, Delay, Frequency is just the text that then appears in cgminer and then the Divider setting. 0583 is 75MHz & 0383 is 50MHz

So everything worked as expected with the Frequency text displayed on the Status page and the correct Hash rate was achieved. However there is no gain to be had as the Core Voltage that you need to get down to for these Frequencies to take advantage is lower than the Miner will hash at.

One of my Hash boards will start hashing at 9V (0.6V Core) however the other needs 9.3V (0.62V Core). You can however as described earlier then wind the voltage down. One thing that I have not tried but that is probably possible at these low voltages would be to run with no fan as the temperature is below 30 Deg

Rich

RichBC

hero member

Activity: 588

Merit: 500

Control board is fine at lower voltages, it's the hash Boards that drop out.

Rich

1slickvdc

member

Activity: 67

Merit: 10

Have we tried powering the control board separately from the hash boards? Maybe it has problems at lower voltages because the control board expects one voltage, whereas the hash board is more flexible?

RichBC

hero member

Activity: 588

Merit: 500

I have a broken S5 with v1.5 hash boards that i bought from ebay. Unfortunately it was more broken than I hoped with one board very trashed and not operational, the second board does recognise some of the chips so there is hope. however i have found extracting chips from the first board & planting them onto the second more difficult than I hoped. I have 26 chips operational at the moment but removing them from the first board is very difficult and always an uncertainty that the chip I am removing is actually good. however i will get there.

Rich

1slickvdc

member

Activity: 67

Merit: 10

Quote from: RichBC on September 26, 2015, 03:50:07 AM

At some point I am going to try and fully understand which of the differences enable the V1.91 to hash at lower voltages and perhaps come up with a mod for the earlier boards/

Rich

Do you have both? I'm willing to use my skills and available materials as part of your endeavor if you want another unit/pair of hands/head.

RichBC

hero member

Activity: 588

Merit: 500

Quote from: flikflak on September 25, 2015, 03:55:03 PM

I'm using two of those little fellas.

There are some 40A/400-500W buck converters in the shop -> that should be enough for VirosaGITS. I wouldnt connect 2 in parallel to one board, because that would be like 2 different PSUs to one hashingboard.

And yes, its a V1.91. I will do some more tests later.

Had not looked on AliEpress, that unit looks quite nice. As you say some 40A units available this one looks quite nice? I will probably buy a couple to test.

Agreed that parallel connecting carries too much risk as you have to set voltages separately and I have no idea if these will current share?

At some point I am going to try and fully understand which of the differences enable the V1.91 to hash at lower voltages and perhaps come up with a mod for the earlier boards/

Rich

RichBC

hero member

Activity: 588

Merit: 500

Quote from: philipma1957 on September 25, 2015, 11:03:56 PM

some not so good news on s-7 underclocking. using this psu

https://www.trcelectronics.com/ecomm/pdf/hrp600.pdf

okay I set freq to 300 which is half speed and I did 2400gh at 612 watts which is normal volts were at 12.15

I dropped volts to 11.85 2400 gh held watts went just under 600 to 594

I dropped volts to 11.50 2400gh held watts went down to 579

At 11.25 volts asics dropped off

so there is some type of governor/regulator on the early s-7 just like the early s-5.

and to be sure the controller was out of the power line it was run with a different psu steady at 12.15

Phil that is a pity, but perhaps not a big surprise given the mention of Bitmain wanting to ship some miners with a reduced Hash and the tight spec they have put on the 12V.

However 11.5V represents a Core Voltage of 0.64V so as the spec sheet only goes down to 0.6V that is not bad.

Rich

RichBC

hero member

Activity: 588

Merit: 500

Quote from: VirosaGITS on September 25, 2015, 10:51:42 PM

Quote from: 1slickvdc on September 25, 2015, 10:49:40 PM

Hash board V1.3 also.

From a low-level sense, there's no reason the chip voltage absolutely couldn't be controlled from the software except by intentional design. Most likely to prevent twits from cranking it up and blowing out their miners. What about a hardware modification? It's a switching supply of sorts to step it down to chip voltage, there's got to be a voltage reference point somewhere that can be modified, eh?

I really don't know the details. Apparently its because they switched to a "string design", the voltage must be controlled at the feed/PSU level. But again only the 1.91v seem to support it without dropping chips.

Maybe RichBC can elaborate on that, but i'm not the guy for that.

Yes unlike the S1 & S3 which have Buck Converters that reduce the 12V down to the 0.6-0.8V range needed by the chips, the S5, S5+ & S7 use a string supply for the core voltage. The chips are connected in series (Like Xmas Tree Lights) and so with 15 Chip pairs in the string in an S5 and 12V you end up with 12/15 = 0.8V / chip. Reducing the supply to 9V you then get 9/15 = 0.6V.

1slickvdc your efficiencies at 12V are very poor it could be because of the PSU you are using is low efficieny? You need to understand and sort that before worrying about undervolting. Also remember at the moment "reliable" undervolting has only proved possible with the V1.91 Hash Board.

Rich

1slickvdc

member

Activity: 67

Merit: 10

Gotta finish fixing the most recent batch of dead laptops before I'll have the money for anything more, so anything I can do with what I've got is fair game, but in terms of buying anything else, that's gonna be at least another couple weeks or so.

VirosaGITS

legendary

Activity: 1302

Merit: 1068

Quote from: 1slickvdc on September 26, 2015, 12:48:36 AM

Worst case scenario and it's got problem at lower voltages, I end up with a nice high current supply for testing car audio gear I repair. Have to definitely consider this unit.

Well as reported you might be able to go down to 11.5v. Unless something weird happens and you get results that we've only been able to repeat on v1.91. But if not, its not like it can't be used as a "Normal PSU" as well, since it work just fine at 12v. The efficiency isint too bad so, why not.

1slickvdc

member

Activity: 67

Merit: 10

Worst case scenario and it's got problem at lower voltages, I end up with a nice high current supply for testing car audio gear I repair. Have to definitely consider this unit.

VirosaGITS

legendary

Activity: 1302

Merit: 1068

Quote from: 1slickvdc on September 26, 2015, 12:38:04 AM

Had that problem on this specific PSU?

Turn out the problem with that one was the price. Its being discussed in page 4 and 5. RichBC eventually went with 2x 5v~ to get a range of 9v-11v. Since the one you are interested in work well, but down to 10V.

So its pretty good, if you get it cheap. Others were discussing prices per unit at 230$ and 125$. The link you have is quite cheaper. So look like a good PSU.

1slickvdc

member

Activity: 67

Merit: 10

Had that problem on this specific PSU?

Topic: Antminer S5 - Underclock - Undervolt - Best J/GH - page 6. (Read 31165 times)