Hacking the S7 - improving efficiency through minor hardware manipulation

[NotFuzzyWarm]

Actually, 4TH would be more like 1100W.

Still nice shaving of the power. More #'s for me to play with to see how much hash & 'spare' miners I can fit in my power budget.

edit: re: my free power (at work) advantage. While playing with numbers turns out I'm using 23kw there which at our rate works out to just s tad under $1,600/mo. Anywho, after deducting that from what the entire farm brings in including what is at home, I'd still NET around $5k/mo. Think I might talk to my Partner about setting up an automatic Coinbase xfr to the company each mo to be nice, eh?

I can certainly afford the BTC2.7-2.8 per month and it'd help a bit towards our current massive R&D expenses piling up for this year. To date that is somewhere close to $450k and climbing fast. That's not counting the usual price of a very nice Ultrafast laser from Trumpf http://www.us.trumpf.com/en/products/laser-technology/products/solid-state-lasers/short-pulsed-lasers/trumicro-series-5000-femto-edition.html that is so far on a FREE 6/mo loan to us from them... OTH, I have nice shiny new toys to play, er, I mean do mid-Pico down to high Femto-sec time scales material interaction tests with. With a selectable choice of IR/Green/UV wavelengths no less.

[sidehack]

One of these days I really wanna hang out at your shop.

[NotFuzzyWarm]

One of these days I really wanna hang out at your shop.

If you are ever up here in area feel free  to stop by. Current work is to once again leapfrog ahead of what our few competitors can provide. Since the last industry jump I did back in 2007 they are slowly figuring how we do what we do so bloody fast and tiny. http://www.industrial-lasers.com/articles/2007/11/processing-ceramic-substrates.html (edit: note the article was our 1st release of Public information. In 2010 we skyrocketed speeds to now 5x those mentioned...) The LED lighting explosion going on everywhere is a direct result of what I came up with back then re: micro-machining ceramics for die heat spreaders in the emitter chips and other high power density chips. The actual emitter dies are smaller than the then-current thermal/electrical via sizes *were* back then. We shattered that limitation so more emitters in same package size. Figure another 2-3 years tops and they will be offering what we do today if they can work a way around our patents.

[hurricandave]

This will sound stupid but, would the controller care if you pulled the fans and ran them from a separate speed controller like a evercool twister or equivalent?

[Unacceptable]

One of these days I really wanna hang out at your shop.

If you are ever up here in area feel free  to stop by. Current work is to once again leapfrog ahead of what our few competitors can provide. Since the last industry jump I did back in 2007 they are slowly figuring how we do what we so so bloody fast and tiny. http://www.industrial-lasers.com/articles/2007/11/processing-ceramic-substrates.html The LED lighting explosion going on everywhere is a direct result of what I came up with back then re: micro-machining ceramics for die heat spreaders in the emitter chips and other high power density chips. Figure another 2-3 years tops and they will be offering what we do today if they can work a way around our patents.

[philipma1957]

This will sound stupid but, would the controller care if you pulled the fans and ran them from a separate speed controller like a evercool twister or equivalent?

yes  it would care but you could splice the tach wire into the controller and fool the unit.

not sure why you would do all that since you can set fan to 20 or 30%

[sidehack]

[farnsworth]Good news, everyone![/farnsworth]

So I spent the day learning how to program in PIC. I have established that I can write code to the micro, and I can see a toggling clock output at pretty much exactly the rate I expected. I'm working on a bit-banging I2C routine and if this were a normal day I'd probably not go home until it was finished and working. However, it's Friday which means running errands before shops close at 5 and then it's burger time.

Otherwise, well, I'd probably be reporting results by about 7PM. That's how well things are going, that I expect it would only take another 3 hours to establish that I have functional I2C communication.

After that, I'd need to clean up the delay routines (right now I'm just inserting the right number of NOPs, but I should use actual loops instead) and then set up a timer for about 5 minutes so I can feed it a starting (hot) value and then, after a few minutes and cgminer is up and running and the chips are warmed up, I can feed it a running (cold) value automatically.

I really need to get back to manufacture, but this is such a freakin' cool project and coming together pretty well so I'll probably try and finish the code up tomorrow and let a test board run over the weekend.

[yslyung]

itchy hand here, can't wait to see some awesome results.

[RadekG]

Thank you Sidehack for continuing in my work. By the way, I found custom firmware from Germany which promises undervolting from WWW panel, so it seems they found our missing key - secret command to control PIC through controller board.

https://www.zwilla.de/de/downloads/custom-firmware-antminer-s7/

 

[sidehack]

I figured it was there, probably part of the reason Bitmain never updated their repository for S7 stuff because they didn't want people to know about it. When I saw a UART line on the micro was tied into the TX line on the controller really gave it away. Wish I'd been able to look into this stuff months ago, but it's been a long time since I had the money to buy miners near release to figure them out. I think the AM Tube in August 2014 was about the last, which was fun because Novak and I were finding and solving problems with those before most of North America even had 'em yet.

I'm having so much fun that I'm gonna continue what I'm working on, and also because that link isn't working for me.

[RichBC]

Always fun to have a Midnight Oil Burning project on the go.   I guess there also has to be a strong possibility that work on the 135 Chp S7 voltage control will translate to the S9?


Rich

[yslyung]

Thank you Sidehack for continuing in my work. By the way, I found custom firmware from Germany which promises undervolting from WWW panel, so it seems they found our missing key - secret command to control PIC through controller board.

https://www.zwilla.de/de/downloads/custom-firmware-antminer-s7/

 

it's fake me thinks & this guy is a cheat too. there's a thread but it didn't last long, kano & cryptoglance chimed in too.

asking $ for something that don't work & did NOT ask permission from cryptoglance to use their software.

[adaseb]

Thank you Sidehack for continuing in my work. By the way, I found custom firmware from Germany which promises undervolting from WWW panel, so it seems they found our missing key - secret command to control PIC through controller board.

https://www.zwilla.de/de/downloads/custom-firmware-antminer-s7/

 

it's fake me thinks & this guy is a cheat too. there's a thread but it didn't last long, kano & cryptoglance chimed in too.

asking $ for something that don't work & did NOT ask permission from cryptoglance to use their software.

Don't think that will actually change the voltage. They released something similar for the S3/S5 but all it did was adjust the voltage in the "advanced" tab but we all know that since the voltage was fixed it didn't do anything.

[sidehack]

Looking through the thread, that's kinda what I was figuring too.

Don't worry guys, mine will work and I won't be charging for it.

[RichBC]

Looking through the thread, that's kinda what I was figuring too.

Don't worry guys, mine will work and I won't be charging for it.

The other giveaway is that it makes no mention of only working on a 135 chip S7.

Rich

[philipma1957]

Looking through the thread, that's kinda what I was figuring too.

Don't worry guys, mine will work and I won't be charging for it.

I was thinking should we chip in and order an s-9 for you to play with?

An s-9 doing .08 watts would be very nice gear.

[yslyung]

Looking through the thread, that's kinda what I was figuring too.

Don't worry guys, mine will work and I won't be charging for it.

yours will defo work, no need to charge, i'm sure many are willing to donate willingly to you.

we know you always do a good job & sharing.

[sidehack]

Tentative success!

I was running into no end of trouble getting anything except delay loops to work, when after about 4 hours I realized I didn't know how to declare variables properly so the compiler was dumping everything into the same register. So I fixed that, and two hours later I have working bit-bang I2C with ACK/NACK recognition for two-byte data in a 164-opcode program. Not bad for never programming PIC before. This was fun.
Anyways. It has a 20mS delay loop after turnon, to let things settle out, and then passes a hot value to the DPOT. After about 3:15 or so it passes the cold value, and actually continues to pass the cold value again every 3:15 after that. I could just as easily tell it not to do that, but eh, no harm no foul.

Anyways I've got it pushed to my hacked-up board running 650mV hot and 620mV cold at 450MHz. It's worker 1BTCMUSEuM1uy4mCce9JGisMZD15RTG7em_S7LN on solo.ckpool for anyone wanting to watch the stats. I'm gonna let it run until tomorro after lunch, and then if it's looking good I'll reinstall the PIC, put the same code on the second board and let 'em both run overnight. If it's still looking good on Monday I'll upload an archive of HEX files for different cold-running values (with a corresponding hot-running value of ~5% higher) so y'all can play around. And I've got access to a full 3-board S7 to jack with too, should be fun.

The gold medal will of course go to reverse-engineering the PIC firmware to figure out what command is sent from the controller to adjust the voltage from software. But I'm not up to that.

[navigatrix]

Actually, 4TH would be more like 1100W.

I'm in. Check your tip jar...

Thanks for this.

[sidehack]

Well the one blade's been running solid at 620mV (9.3V, down from 10.0V stock) and 450MHz (1TH) for about 15 hours now. This afternoon I'll do some more testing of the absolute bottom end efficiency and overall power use, generate some charts and plots for y'all. I'll go ahead and work up a stock S7 as well.

Just wondering, would anyone be interested in a group buy of S7 or S7LN pre-flashed to the desired improved efficiency setpoint for a moderate markup (maybe $20-30), tested and reshipped within the US within one or two days of arrival at my shop?