Topic: Antminer S5, Spondoolies SP20 or 3 x Antminer S3? (Read 4654 times)

Thanks Rich,

I am using Kano's latest that has a voltage setting in the advanced tab of miner config. But, I'm not sure if I'm inputting the values correctly. See link below where I'm going to continue this thread in hopes of also getting input from Kano and some others on software questions. I think continuing in the topic below will give it better visibility.

https://bitcointalksearch.org/topic/m.12230103

I can't find the spec on the Digi+ VRM AGP1209, however the Digi+ VRM ASP1212 is a programmable Multiphase Buck Controller. So I think you have found the elusive programmable S3. In which case it should respond to voltage programming from cgminer so long as you have a firmware version that supports it.

Am not certain which version you need so some experimentation may be needed. versions to try are.

140826 141013 141024 141126 141219

Be careful with 141024 as this is the one where the reset button does not work

Rich

Here are some close up pics where you can read the markings.

I installed Kano's latest [CGMiner] => 4.9.2  [API] => 3.6, details of which are at https://github.com/kanoi/cgminer-binaries/tree/master/AntS3.

It has the voltage adjustment, which I tried setting to .65 V and restarting as instructed and nothing changed. Still hashing at 450 Gh/s and 350 W for the same 0.79 J/GH I was getting before. As you know, if voltage would have actually changed I would have expected X's in the ASIC status and having to reduce from the current 225 MHz to stabilize. Would have been great if it was that simple.

It is an early S3, but the newest one of the 5 that I bought. Below are front/back pics from one of my other 4 boards showing the resistor mods on the backside. They seem to be the same version that you're modding. I'll try to get a pic of of the top of the mystery board that shows the markings.




Thanks for your help.

Well that is definitely a version of S3 I have never seen before. Have you tried the firmware version that allows voltage adjustment from cgminer to see if it responds to that. It still had Buck converters but there are trio's of chips on each one. Can you take a picture just of the top of the board where the components are such that the component markings can be read?

I see the ident says V1.0 so am assuming it is an early S3?

Rich

Hey, Rich
Here are pics of front and back of board. Let me know if you have any suggestions on how to undervolt. Thanks.

I believe I stated the price is too much at the moment.
From reading the posts and work done to the S3 and S3+ I believe he has as much chance as anyone to make the Sfards chip sing. \
The simple fact us none of us know because there haven't been any / many if any seen in the wild. Jabber has done work with them on his dev chips, but very few people have done anything.

My only recommendation or suggestion was that he check out the chip, the technology for the lower efficiency. If you notice, that is exactly what RichBC seems to be skilled at doing.

How can you honestly suggest something cannot ROI when you haven't seen the person's skills?

There are many talented people in the bitcoin scene and I try to listen to all of them. I suggest you try your hand at more listening to some of those folks and understanding the context of other's posts.

RichBC thanks for what you've shown. Regardless of any repeat in the end result, it is refreshing to see a different application with the rheostats which I can get on the cheap. That is a fantastic way to dial in what an individual miner may do the best at.
Please don't stop sharing!
 

If there were no obvious Buck Converters it's possible you were looking at the S3+ which was a trial for the S5 with string power and no DC-DC converters? However I have seen pictures of an S3 with a different chip and associated circuitry but still 8 Converters?

I will post up my stuff in the Undervolt thread when I get a moment. Am also exploring S5 undervolt.

Rich

Thanks for the quick response.

I'll get some PNs and will post some pics. I tried looking for the hashing chip driver and an associated voltage divider circuit without luck. I was looking for 8 similar circuits like we're seeing on the boards we modded, but there are not 8. I did find 4 similar circuits on each board so it's possible that these 4 are doing what the 8 TPS53355 synchronous buck converter is doing on the older boards. But when I tried googling the PNs nothing showed up.

May be good to post to undervolt thread and continue there???

Yes I was pleased.   I actually ended up with 0.49J/GH as with more testing it started dropping the odd chip so I had to push the voltage up slightly. Yes the 270 Ohms was the additional resistor in parallel the VFB feed resistor, and ended up being  a 330 Ohms.

I only have the one S3+ & one S3 they both have the TPS533355 so that's all I have played with. Have seen pictures of what looks like a different chip, do you know the part number?

I will post up all my measurements on the S3 Undervolt thread if you think that would be of use?


Rich

0.48 J/GH is a great result! I soldered a 2.2k resistor across each of the 8 VFB resistors per board for 4 of my S3+ and tried various frequency settings to get the lowest J/GH. Looking at the OOXX results, measuring power, and GH/s, the results varied for each ranging from 250 - 390 GH/s with the effeciencies being all around 0.61 J/GH. It sounds like I need to lower the resistance more. A couple questions.

Were you adding 270 ohms across VFB or was that the total resistance across VFB after you applied your additional parallel resistance? If the latter, what was the additional resistance?

Some of the newer S3+ boards don't use the TPS533355 chips. Have you ever tried to undervolt them?

There is an S3 undervolt thread at https://bitcointalksearch.org/topic/undervolt-antminer-s3-771979
I'm sure they would be interested in your results.
Thanks for the info.

Nothing on the market currently will ROI at 15 cents/KWH electric cost. Even turning the SP20E down to it's lowest undervolt setting isn't profitable enough to ROI before the bitcoin reward halfing next summer kills it's profitability.

 Technically, you CAN mine profitably right now at 15 cents/KWH with the S5, the SP20E, or the SF3301 - but unless they're almost paid off already, forget getting back the money you'd pay to buy one right now (and don't forget cost of power supply adds to that). I'm not sure but I think the Avalon 4.1 MIGHT be able to mine profitably at 15 cent electric, but that would be VERY marginal.


 New hardware is still a major question mark.

 The SF3301 COULD be profitable and hit ROI if they dropped the price down to something reasonable, but at the current $1199 PLUS SHIPPING it's not likely to ROI unless you have VERY CHEAP electric - and looking very iffy even THEN, IF you can actually get one direct sometime this month as opposed to paying the crazy pricing from folks like Zoomhash.

 No clue about performance or cost on the announced S7 yet - but that should be soon, I'm certain Bitmain is MAKING them but replacing S5s in their own farm before they start selling S7s. I would guess late this month or sometime in September we'll finally get some hard information.

 No firm knowlage about anything based on Innosilicon's announced A3 (ditto the A4 for Litecoin fans) yet though it appears LTek (sp?) is planning a next gen miner around that chip and I'm sure Innosilicon will produce a "reference" design and sell it like they did on the A2 Terminators.

 It appears that BitFury is ONLY selling it's current gen gear to BIG farms - or might only be selling the chips and letting the farm do the board design and/or board build work.

 KnC isn't selling to anyone else any more, not that I've ever been impressed with the reliability of their equipment. They also have bad enough legal problems I won't be shocked if they end up bankrupt sometime in 2017 (legal processes tend to take a while when the defendent can afford good lawyers).

 No clue yet what Avalon is doing on next generation if anything.

 No idea if the Alchemist folks are even thinking about Bitcoin or a next-gen Litecoin miner design.

 The folks that announced an X11 miner design haven't said anything since they canceled out "due to lack of preorders, going to look for venture capitol investment". Not sure if that one was vaporware or not, but it seems they actually refunded preorders which tends to make me think they were legitimate - just badly underfunded.

 I try to ignore vaporware announcements, can't think of any legitimate folks still in the mineing hardware business I haven't mentioned above.

Thanks for your tips. I don't want to buy old mining equipments. I give up on S5 dream, too. I should save my money for better investment plans.

At 15 cents you are just into loosing money territory with an S3. If you want to have a play, as I am, best value for money is a 2nd hand S3 with an undervolt mod & the frequency turned down.

I think that with 15 Cents power any existing  S5, or new hardware S6/7 etc will not ever pay for itself, particularly with Bitcoin halving less than a Year away.


Rich

I have 15 cent per kw electricity cost. Do you think I shouldn't mine with that price?

Funny that you mention the DC-DC converters as those were the original problem with the S3s .  When Bitmain first announced the S3, they were promising 504GH/s hash rate.  Later, they tuned that down to 478GH/s and finally on launch, they were only 440GH/s.  All because the DC-DC converters they obtained were unreliable and shoddy.  Sure, some of the units will hash at 504GH/s (one of mine has been for a year now).  Some of the others won't even hit the advertised 440GH/s (one of mine has to be down clocked to run stably at 400GH/s).

I guess it's 13-15 cents around as he's living in Turkey. I wouldn't mine for these rates though, there are cheaper places to mine such as East of Turkey. (They don't pay electricity fee )

what is not cheap power?

 21 cents a kwatt? = loser  with any gear.  as the power costs more then the coin mined as soon as your start.

so the question is how much do you want to lose.


the lower then 21 cents means more time before power costs kill your profit.

so what is your power cost.

if you tell me 20 cents I would say do not spend much and mine a little.

I was having same questions these days. I'm so unsure about buying new mining equipment. Should I buy Antminer S5 or should I wait S6's to come? My electricity is not cheap.

I decided to try my hand at Undervolting the S3, I would like to play with the S5 but am hoping the second hand prices will come down a bit first.. Although there are posts around that say you can change the core voltage using the Advanced setting field, like some others have found on my S3 & S3+ it seems to make no difference to the voltage.

I wanted to be able to easily adjust the voltage so I soldered in 2 x 4 channel 5K digital pots across the VFB resistors. These are setup with an Arduino, which also measures the voltage set..



In practice I have just rediscovered what other have found, but it's good to see the numbers for yourself.  

The Standard S3 with 218.75MHz clock had a core voltage of 0.8V hashed at about 440GH & took 321W at the wall giving J/GH of 0.73

After trying a lot of frequencies and voltages I decided it was possible to continue at 218.75MHz with the voltage reduced to 0.73V. This gives a small number of hardware errors .001% but improves the J/GH to 0.65. I soldered 4K7 resistors into the S3 an am letting it run, while continuing to play with the S3+ with the digital pots.

At my electricity cost this moves me from Break Even to about £1 Week profit. I am happy with this for the moment as it's buying BTC and my Wife is happy as it's drying the washing.

As would be expected as you drop the frequency and Core voltage the efficiency increases and the J/GH drops.

The best I have seen was at 125Mhz, 0.64V Core Voltage, giving a hash of about 254GH which is J/GH of 0.48. This was with 270 Ohms across the VFB resistor and at this point I run out of adjustment on the core voltage as we are very close to the TPS533355 0.6V reference voltage. However I am actually surprised that the chip was still hashing at this voltage.

I could probably squeeze a slightly better result by doing away with half the DC-DC converters and moving to 4 chips per, as the currents are now much reduced. I may try this later?

Rich