BITMAIN launches 4th generation Bitcoin mining ASIC: BM1385

[2112]

Thats why you employ people that know what they are doing and use MPW runs. Not rocket science, is it?

It seems to be impossible for the Bitcoin miner designers to hire the knowledgeable people. Do you know of any Bitcoin mining ASIC project that doesn't look like a student project or a quickie hack job?

Thus far only ASICMINER acknowledged (very early) of being unable to hire or contract anyone with power/analog ASIC experience.

[Tupsu]

@dogie

Of course, an S5 isnt obsolete when the S7 has been released - you have your points here.

@Tupsu

Thanks for the pics- that looks cute. So it is quite possible.

S5 controller



S5+ controller



The same space, but the temperature difference is 10 degrees Celsius.

[2112]

I am surprised that they say that full custom poses a higher 'risk' - that's only true for very complex chips like cpu's, not for the very simple (and I mean very simple) functions found in SHA256.

The added 'risk' is because for the first time they are getting outside the 'standard cell' design flow.

I have doubt that the new design is truly full custom. Their previous designs were simple unrolled hashers. True 'full custom' optimized design would be rolled. And switching from unrolled to rolled would involve redesign of the I/O protocol.

My bet is on them purchasing a custom standard macro library: lower-power by lower-area and lower-noise-margins. Sort-of like bitfury did for his first chip: 55nm-drawn transistors in the 65nm-nominal process.

Such a 'extra-low-power' library may be violating some default DRC's (design rule checks) of their foundry. Thus the foundry makes them explicitly waive DRC conformance warranty with their mask order.

[Tupsu]

I am confused are you saying you can take an S5+ and 3 controller boards from standard S5 units and make them all run solo at 2.7th/s? If so that is pretty damn impressive and cool.

Yes. I did.

S5 controller with S5+ miner

[alh]

I am confused are you saying you can take an S5+ and 3 controller boards from standard S5 units and make them all run solo at 2.7th/s? If so that is pretty damn impressive and cool.

Yes. I did.

S5 controller with S5+ miner

That's pretty sweet. Looks like you put together more like what folks would have a thought an S5+ would be, before pictures of the "3 module monster" appeared. I thought there was a 16-pin -vs- 18-pin mismatch on the cables between the controller and the hashing board. What happened there?

Have you got any power measurements? I expect it would be roughly 1150W at the wall?

[brontosaurus]

I am surprised that they say that full custom poses a higher 'risk' - that's only true for very complex chips like cpu's, not for the very simple (and I mean very simple) functions found in SHA256.

The added 'risk' is because for the first time they are getting outside the 'standard cell' design flow.

I have doubt that the new design is truly full custom. Their previous designs were simple unrolled hashers. True 'full custom' optimized design would be rolled. And switching from unrolled to rolled would involve redesign of the I/O protocol.

My bet is on them purchasing a custom standard macro library: lower-power by lower-area and lower-noise-margins. Sort-of like bitfury did for his first chip: 55nm-drawn transistors in the 65nm-nominal process.

Such a 'extra-low-power' library may be violating some default DRC's (design rule checks) of their foundry. Thus the foundry makes them explicitly waive DRC conformance warranty with their mask order.

You have completely the wrong view of full custom, a rolled design would be a really dumb idea for a modern mining chip and very area inefficient, the customisation involves only two circuit elements, but I'm sure you know that. Not rocket science at all, no magic, and very little risk if you have some respect for semiconductor physics. DRC is there for very good reasons which again I'm sure you know, and only an idiot would even consider violating them.

[Abiky]

This new mining chip is more energy efficient. Every new model is even more efficient than the previous one. Can't wait for Bitmain to release the Antminer S7. This is something I would like to buy...saving my money to buy a pair of these 

[2112]

You have completely the wrong view of full custom, a rolled design would be a really dumb idea for a modern mining chip and very area inefficient, the customisation involves only two circuit elements, but I'm sure you know that. Not rocket science at all, no magic, and very little risk if you have some respect for semiconductor physics. DRC is there for very good reasons which again I'm sure you know, and only an idiot would even consider violating them.

The rolled vs. unrolled isn't a fully resolved choice. The losses and noise in the very long lines that drag the signals over 15 SHA-256 rounds are quite significant. I think the bitfury approach of routing hashed words in one direction and constant SHA coefficients in a perpendicular direction gives overall savings over trying to squeeze combinatorial optimizations after fully unrolling. Most of the combinatorial optimization gain is achieved by SHA-256 round pairing, i.e. 32 round-pairs instead of by-the-FIPS explicit 64 single-rounds.

I did not do a full analog modeling of both choices (rolled/unrolled) for SHA-256. But I've done something similar in the past that was bound by the speed of carry-look-ahead adders. I actually doubt that anyone here on this forum (maybe with exception of bitfury) did the required tradeoff analysis. My scientific will-ass guess is that Bitcoin miner has a possibility of being an example of one such circuits where leaving things rolled will be of great benefit. The very high toggle ratio (only -6dB below the theoretical maximum of a ring oscillator) will probably benefit from using some sort of SCL (source-coupled logic) or CML (current-mode logic) instead of the garden-variety CMOS bang-bangs that every CAD monkey throws at the Bitcoin mining problem.

People do fully unrolled hashers because the logic synthesis tools use heuristic place & route algorithms that don't converge or converge extremely slowly on the rolled designs.

As far as I understand the full DRC compliance at 28nm "mature" process is very, very conservative. I don't have any exact numbers handy, but the assumed gate  error ratios for a "digital" manufacturing process are way too high for Bitcoin miner that can easily tolerate a percentage point of errors. Violating some of the DRC to shed the unnecessary margins is one of the simplest ways to save power, after the obvious things like dropping JTAG and other testability overheads.

Re-reading your first sentence, I don't really understand the part

the customisation involves only two circuit elements, but I'm sure you know that.

Could you restate what you had in mind?

[suchmoon]

Before anyone gets too excited over this development I would wait until you know what the die size is and the operating frequency at quoted power consumption.

I am surprised that they say that full custom poses a higher 'risk' - that's only true for very complex chips like cpu's, not for the very simple (and I mean very simple) functions found in SHA256.

Good luck to them though, anything that puts a spanner in the works of KNC or 21 has to be welcomed.

Well when you're considering laying down nearly 8 figure sums without seeing the chip work first, you better be damn sure its going to work.

Oh, and 'nearly eight figures'?

6, 7, 9, or 10 presumably. 100k USD sounds affordable

Take a look at what S1 prices did in early 2013. Diff went way up really fast, coin prices went down, and the price for an S1 dropped about 90% in something like four months. You think my December S1 would have sold in March for the 4BTC paid for it, or more like 8BTC if priced in dollars? Only to a fool.

So I'd say that, in general, with some exceptions, someone who takes a sizeable chunk out of the viable life of a miner and then sells it at new price is not selling at expected fair market value, and someone that buys it probably didn't do his research to know he was being ripped off.

The point is not, and never has been, that selling secondhand hardware at fair market value is bad. The point is, and always has been, that successfully selling used hardware at well above market value is taking advantage of a fool, which is unethical. Implying that the proper strategy for making positive returns on mining is (and always has been) taking advantage of fools, well I like to think this community is better than that.

My respect for what you do for this community notwithstanding - I think you're wrong on this one. S1 showed up at the end of 2013 - so perhaps you meant early 2014 there. Yes, S1 and S2 depreciated quickly. However S3, S4, S5 and perhaps even the last runs of the S2 could be used for a couple of months and sold for near their original price without ripping anyone off, due to all sorts of interesting scenarios with difficulty and exchange rate changes. So that's 3:2 against your historic argument, and the future is anyone's guess so let's make it 50:50

Again, you're taking an idealistic view of this. Purchase price minus "depreciation" must be your sale price? That's almost never the case. As someone already mentioned, if they list it as an auction on eBay and describe the condition properly and don't advertise it as a free money making machine - that's a fair game. Buyers electric cost, risk tolerance, future outlook, $25 eBay coupon, preference to buy locally instead of from China, and a bunch of other factors can affect the end price.

[itop_james]

I am confused are you saying you can take an S5+ and 3 controller boards from standard S5 units and make them all run solo at 2.7th/s? If so that is pretty damn impressive and cool.

Yes. I did.

S5 controller with S5+ miner

It is great test , how long you run it like this , I am thinking the stability of using lower version of controller .

[@boutiuqe]

If cloudmining is making profit why don't owner just mine theirself ?
Why do they want to sell to you guys ?

[suchmoon]

If cloudmining is making profit why don't owner just mine theirself ?
Why do they want to sell to you guys ?

Shifting part of the risk onto the buyer and getting money upfront.

[Marvell1]

If cloudmining is making profit why don't owner just mine theirself ?
Why do they want to sell to you guys ?

Remember the guy who killed the golden goose ...  they don't want to be that guy

[generalt]

I am confused are you saying you can take an S5+ and 3 controller boards from standard S5 units and make them all run solo at 2.7th/s? If so that is pretty damn impressive and cool.

Yes. I did.

S5 controller with S5+ miner

So basically instead of just selling a single S5+ as a single 2.7th unit they forced people to buy 3 of them chained together.  I would've preferred the single 2.7th unit.  Sounds like they were in a big rush to dump the remaining stock of their older chips.  I'm sure their hosting facility has the S7's up and running already since they dumped their used S5 miners as well at new miner prices.

[philipma1957]

I am confused are you saying you can take an S5+ and 3 controller boards from standard S5 units and make them all run solo at 2.7th/s? If so that is pretty damn impressive and cool.

Yes. I did.

S5 controller with S5+ miner

So basically instead of just selling a single S5+ as a single 2.7th unit they forced people to buy 3 of them chained together.  I would've preferred the single 2.7th unit.  Sounds like they were in a big rush to dump the remaining stock of their older chips.  I'm sure their hosting facility has the S7's up and running already since they dumped their used S5 miners as well at new miner prices.

well they could have charge more for them as singles, but 3x the boxes and more logistics.

the strongest thing I get from this is they went all s-7 as I type.

they will not be selling any more s-5's or s-5=+ units.

and they are clever enough to slowly add s-7's  to keep hash growth under 2 or 3 % 

thus picking up big savings on power costs.

[generalt]

You're right.  I guess they would have to swap out the right number of 5s for 7s as to not effect the hash rate too much but yeah the power savings alone should be huge.  I guess there's a delicate balance.  They can't turn them all on and make the difficulty jump too much.  If they do that then they can't charge as much for the S7 to their customers.

[notlist3d]

You're right.  I guess they would have to swap out the right number of 5s for 7s as to not effect the hash rate too much but yeah the power savings alone should be huge.  I guess there's a delicate balance.  They can't turn them all on and make the difficulty jump too much.  If they do that then they can't charge as much for the S7 to their customers.

And it put's them at a big advantage compared to other mining companies.  We don't know for sure what they are doing.  But likely the first S7's will end up in the spots the used S5's were.

When it happens is hard to say.  But I'm sure they have a good business plan.

[Tupsu]

I am confused are you saying you can take an S5+ and 3 controller boards from standard S5 units and make them all run solo at 2.7th/s? If so that is pretty damn impressive and cool.

Yes. I did.

S5 controller with S5+ miner

It is great test , how long you run it like this , I am thinking the stability of using lower version of controller .

It is not - "lower version of controller"

At the moment,
13h 36m
10HW error
Freg   325

[dogie]

Wouldnt it be nice to have some kind of standard in the BITMAIN world?
I'm thinking of just changing the blades and the firmware to upgrade an S5 to an S7. It would reduce the prices a bit (lets say 10-15%, its shipping costs, VAT ..) and it would also kinda bond the customer to the company as well.

This was discussed a while ago in design, and the conclusion was that its not actually that desirable for the same reason that S2 upgrade kits never appeared. By the time the next generation appears, the previous generation is still mining away perfectly happily and so you can't immediately replace those hashing boards.

Ie if the S5 was modular to the S7, you'd still buy a full S7 because any S5s you had would still be fine to run for the foreseeable future. No one would scrap their S5, they'd simply run both or sell the S5.

So did you know for sure about the S2 uprade kits not being produced and held back a response?  They only failed do to Bitmain not following through on not one but both announcements.  It's a shame to let a good case and design that's accomodating and easily upgradeable die. 

No I didn't know, that was a management decision. The S2 design wasn't very good and was subsequently never repeated. It required a motherboard, slots that often broke, large areas of dead space for cooling and is probably the only Bitmain miner that was susceptible to shipping issues. And an included non server grade PSU which liked to die after a year.

[QuintLeo]

If cloudmining is making profit why don't owner just mine theirself ?
Why do they want to sell to you guys ?

 Because they make more money and amortise the cost of the chip design a LOT faster by selling miners to those of us that don't trust cloud mining.