Topic: [ANN] Sfards:SF100, the first 28nm Dual-Mode Miner is accepting pre-order now - page 61. (Read 129811 times)

yeah pretty much agree we have 19 pages of what if.

And still waiting for a working demo of the gear they will sell.

 BTW I had 1 sp20 give me a steady .42 watts on a plat psu  with a solid underclock.

I like your lines of thought, bronto. But let's be careful not to get too far into the speculative world; it may be interesting and entertaining to do, but that's as far as it goes. At this point all we have is a datasheet. The rest is, to various levels, speculation.

I will admit that, based on the datasheet, my expectations for products based on this chip are tempered. Bitmain's BM1384 and Spondoolies's Rockerbox system designs approach ~0.4W/GH at the system level (never mind chip level) when underclocked/undervolted. For example an SP35 at 3400GH/s uses 1500W at the wall ... a system efficiency of 0.44W/GH. KNC's Neptune is also in that range.

If an Sfards system achieves 0.4W/GH at the system level it may not be a significant enough improvement over what's already available. It all depends on system cost, though, which is something we don't have a clue about. So I'm quite happy to remain patient and wait for an actual system demo, with measured specs, rather than speculating at this point.

I still think there is a home market.  It is harder an tighter margins but it's still there.

And they already have the data centers in place to put them.  I would expect them to make quite a few of their own. They made TONS of dragon's and they plugged them in until they sold.  They would take what remaining A1's they have out and throw in new miners with new chip.  So investing does make sense for them.

Who are they going to sell the majority of something that is 'cutting edge' to?

Not the home market that is for certain. They might just build and farm or like every other fabricator find a deal with someone who has the money to invest in petahash volume.

Much though I hate to be a party pooper, although it's very nice to see a new mining chip design, especially a dual mode one,
this design is really nothing new, SOI or not. Low w/(Gh/sec) figures can be achieved through simply lowering the clock speed
and/or chip voltage, if you look at the power specs you see that the device in SHA256 mode takes just under 78W at 0.8V at
870Mhz with a hashrate of 139GH/sec. This is it's most efficient point. Why?

At what seems like the lowest W/(Gh/sec), 0.31, the chip is running at 0.6V, 300Mhz and hashes at 48GH/sec. To reach 139GH/sec
performance you'll need to have (almost) 3 devices, each running at 15W. It's more power efficient for sure, but to reach ROI
you'll need to pay a lot more for your hardware. So what's the tradeoff? If you have one machine with, say 16 chips it will take
about 1.6kw at the wall, assuming you use DC to DC converters rather than stringing miners together. The 3 machines with chips
running at 15W will take about 900W, a difference of 0.7kw, or very roughly 500kWh per month. ( A bigger miner with 3 times as
many chips won't save much in build costs)

If your electricity costs 10 cents (US) per month then you save $50 in power, at 20 cents it's $100 and so on. Since the
hashrate of both solutions is identical, the network hashrate and bitcoin exchange rate doesn't come into the equation, so it's
simply a case of power saving per month to see how long it takes to pay for the extra two mining systems.

If each system cost $500 (which is unlikely as you'll see below) then at 10 cents/kWh it will take 20 months for the seemingly
more efficient 3 system solution to break even against the single one, or 10 months at 20 cent/kWh.

For some interesting reading about how chip voltage affects efficiency, see this excellent paper:

web.ece.ucdavis.edu/~anhttran/files/papers/atran_icce10_adder.pdf

It's for a 32 bit adder (which SHA uses as well as single full adders) but it will give you the general idea.


Many thanks to TheRealSteve for pointing  out that the die size is actually given in the data sheet on the packaging diagram.

I have to admit that I missed that, my estimate of the size of the chip is 12.5 -13.2 mm on a side, equating to a packaged device
cost (in volumes of 2 million pa) of $11.4. With the actual dies size of 10.5 mm on a side, that price comes down to $9.1, saving
$37 per 16 chip system. That does make a difference.

To build those 16 devices into a system (without a power supply), pay back $1.5 per device as amortised development costs of
$3 million over 2,000,000 units, test and package and pay it's share of Sfard's ongoing running costs will result in
$360 (ish), so expecting it to retail at $500 is unlikely. $550 gives Sfards a 33% margin which is much more realistic, giving a
cost of $0.27/Ghash. Yes, you would also have just under 30Mh/sec of scrypt which changes the picture quite a bit.

Problem with the scrypt side of things is that the die area of this function is about the same as the SHA256 one, eDRAM takes
up a fair amount of space. Is it worth it? According to the datasheet, in dual mining mode the device will run at 100 Gh/sec (SHA)
and 1.75 MH/sec (Scrypt), so for the 16 chip system:

28Mh/sec will earn about $29 per month before electricity costs, about 50kWh / $5 at 10 cents/kWh, so earning $24/month net

1.6 GH/sec will earn $95 per month (BTC = $220/ Hashrate 370 PH) before electricity, 810kWh / $81 at 10 cents/kWh, earning $22/month net

The total payback of $22+$24 = $46 will pay back the cost of a single machine in 12 months IF the network hash rate stays under 400PH.
Even if you got the miner at cost, it would still take 8 months to pay for itself, never mind making any money. That's also assuming the network hashrate stays under 400PH which I think most would agree is extremely unlikely.

HOWEVER, if you get your electricity at 4 cents/kWH then the net monthly revenue is $74 per month, and if you bought these hypothetical machines in volume (100+) then you might get  them at $450 - then you would get payback in 6 months. Before some forum 'expert' (you know who you are) tells you that in Iceland power costs 2 cents/kWh, you'll only get that rate if you buy 100GWh per annum quantities, and you still have to house your miners and pay someone to look after them. 4 cents is much more realistic.

Oh, and don't forget that 2,000,000 of these chips would add 200PH to the Bitcoin network and 3,5 TH to the Litecoin one .........

They have the money, they have the people and experience.  So they should be able to make a working miner.

What it ends up being, or what it will be on spec is unknown.  It does seem they are not in a hurry, they are taking it slow.  Likely focusing on profit not how fast to pump it out.

most likely they will.  they sprung from gridseed hubs,blades & 5 pack blade.  So they did make very good LTC scrypt miners last year.

No reason they can not make one  of good quality.  Still at the same point when and how much. As always time keeps on ticking ticking into the future….

To SFARDS:
DVDD_OSC    Digital VDD. 0.8V-1.10V
DVSS_OSC    Digital GND

This two are listed in section 7.2 table. But not present on schematics and ball map.

Could you create a  Github repo or DropBox folder to share your work ? Do you plan on making your designs available  to the community ?

So, basically.. Soon™

when do you think that they will ship a working miner?

Then I may be interested.

Regarding the Body Bias, it seems I've found my answer in the reference design schematics:
BTC_GNDS_0
BTC_VDDS_0
BTC_GNDS_1
BTC_VDDS_1

must be the bias adjust pins.

Since the higher body bias seems to give the best efficiency, we should use a fixed value.

good old stable from 2013.

And now, this post is getting hot until next news

as much as i like the idea of another home miner an sp20 spondoolies unit (new i think) at 1100w on ebay goes for 400 bucks or so....with my biz and equip
deduction on the equipment over 5 years make it 300 usd lets say.....i can't get it to work at that price at 12c kwh etc at least at current btc prices

now for large scale water cooled big glacier cooled thermo electric at 2c kwh data farms it may work just dandy

anyway by the time they got the unit out the door as a home miner (or even a data hall large rig) could also be too late

now again that is assuming the above stats in the ball park

(remember I just pipping in here i know nothing...i at one time drank the kool aid and got a 1st batch knc 300mh titan scrypt miner no roi no refund...so take what i say in stride!)

So if they are really making a 24 chips machine were looking at these specs

1.7 TH/s
@550W
~650-700 watts at the wall...and thats being conservative.

That is if you go with .6v

If they price this under $500 a pop they'll have a winner.

What Kicad version are you using? Stable 2013 or newer with library table?

OK, lets see what the 4th brings forth in terms of dev board et al - never know, if you decide to send off for some 1 or 2 chip breakout PCB's (OSH or similar), I may be interested as 4 layer can be workable cost wise.

It's a good starting point to build a board off of, though, as it saves the schematic component / PCB package mapping effort - same as with my Eagle part.