Topic: CoinTerra announces its first ASIC - Hash-Rate greater than 500 GH/s - page 112. (Read 231002 times)

I wish I can understand what he was saying went from hours in a day to gallons to beer. End of the day he is saying their production cost for 1 chip is $500 lol.

This whole discussion has gone way over my head, but this comment definitely made me laugh harder than anything in a long time (including Inaba's trolling of the "Status of BFL" thread that's been going on recently).

What are you even arguing? Your own calculations show that Avalon is profitable at $1200 with only $100-$200 in Silicon. So please explain to me how $1/gh is production cost for CoinTerra?

That means CoinTerra has a production cost of $500 per chip, are you smoking something?

You can't just compare Avalon's selling price and then triple that and pretend that somehow means anything.

Avalon's chips are only 275MH/s and 15mm @ 110nm.  Moving to 28nm (110/28)^2 would make the same Avalon chip 1/15th the size (or about 1mm2).  Of course nobody makes chips that small but Avalon could make the chip 15x larger and thus have 15x the hashing power for the same SIZE so something on the order of 4 GH/s (15 x 0.275 GH/s) and 15mm @ 28nm.  Still 4 GH/s isn't 500 GH/s.  With Avalon design you would either need to make the chip much bigger (and thus the cost per chip rises but the cost is per mm2) or use multiple chips to get a 500 GH/s miner.  You can't simply say oh Avalon chips are cheap (<$13) so lets triple that cost ($39) and pretend shrinking it to 28nm makes it a 500 GH/s chip and that makes it ($39/500) $0.19 per GH/s.  

You pay per wafer, the wafer is a certain size, the size of your chip determines how many chips per wafer.  Wafer cost / number of chips per wafer = chip cost.  Bigger chip = bigger % of the wafer = higher cost per chip.  The cost per GH/s depends only on wafer cost AND GH/mm2.

Your math was just utter nonsense.  It is like saying a day is 24 hours and a gallon has 4 quarts so the cost of a beer is 4/24 = $0.16, any beer (regardless of if it is a small glass or giant flagon) can't possibly cost more than $0.16.

Ok I'm not going to argue anymore final point which has been made over and over again. Avalon Batch #1 $1200 say 96 chips that means $12.5 per chip. Say the cost is much more since it's 28nm lets triple the cost, $37.5 per 28nm chip add all the other costs say $200 a chip. 1 of their chips is 500gh/sec 2 chips to make 1TH equals $400 dollars costs, so they can def. sell for $1000 per GH.

The only reason for the price to be higher is that they are trying to recover their designing and developing costs, not the actual cost of the chips. Case in point again:

Intel Core i3-3225 22nm even smaller selling for $139.99 for a profit. Please enlighten me about these high magical costs of yours, actually please don't because clearly you are wrong or else Avalon chips wouldn't cost $12.5 and Intel chips $30-$100

The silicon itself has a cost.  The cost of miner is never going to be below the silicon cost, plus cutting, packaging, testing, etc.  Then you add on BOM (balance of material = everything in a miner other than the ASIC), assembly, labor, profit margin, defects.

$1 per GH/s is pretty close to the silicon cost.  Chip companies aren't going to sell chips at a loss.
[/quote]

So, a 300mm wafer has an area of about about 706cm^2.  And HashFast's chips look like they'll be less then 1cm^2, from their renderings.  You can't just divide the area by the chip size since the wafer has to be "pixelated" into chip shapes and the margins around the outside will be lost.

But, using this online calculator  we get about 600 dies per 300mm wafer.

Assuming a 10mm^2 die, you get $5000/(600chips*400Gh/s) you end up with $0.02/Gh/s

With a 20mm^2 die you get 140 chips and $0.08/Gh/s

With a 32mm^2 die you get just 58 chips and $0.24/Gh/s, which is the max on that calculator.

That's obviously not the cost of actual fabrication, of course. I was looking around and space on "multi-project" wafers for prototypes was around $15k per mm^2 with 18 week wait times.

I'd like to know exactly how much a 28nm wafer costs to actually run but it seems like you have to sign an NDA before they'll even tell you the price.

Until we see Cointerra's dimensions it is hard to say.  As for looking up the price of silicon, make sure you are looking at the cost of a printed wafer.  Blank wafers are used by FABs to print chips.  Those 28nm FABs costs billions to build and have a limited economical lifespan.  The price of raw silicon is pretty irrelevant.

Still I don't really care enough.  If people want to wait for sub $1 per GH/s assembled miners well let them.  I don't own any of these companies.

Your comparing apples to oranges.  Unless you OWN a fab your price isn't a blank silicon wafer, it is a printed silicon wafer.  No company with a 28nm FAB is offering small runs @ $500 a wafer.  

Cointerra is talking about a 500 GH chip. Are you really trying to claim that their chip uses $500 of silicon? Or even $400 worth? I don't think that silicon price is the limiting factor in any of this; I just looked up the price of silicon and unless I did something very wrong it looks to me like that amount of money would buy you like a KG. Additionally, many of the costs you are talking about (e.g. testing) are probably about the same regardless of number of chips on a board.

As suppliers continue to ramp up production numbers the price of blank 12-inch silicon wafers continues to fall. Early in the year 12-inch blank wafers were about US$500, but have since fallen to $200, reflecting the nearly three-fold production increases by some suppliers. I wouldn't even care to guess what the prices are now probably a few dollars the most.

Again Avalon has hundreds of chips on their unit, and they were able to make a profit on it at $1200 and that was hundreds of chips so I'm sorry but i think you are way off with your pricing.

On edit:  forgot to reduce chip size converting from 55nm to 28nm, the corrected fabrication guestimate is ~$0.25 per GH/s instead of original $1.00 per GH/s.


The silicon itself has a cost.  My data is somewhat old but a 300mm wafer runs ~$5K in large volume (think AMD or NVidia), lets be generous and say for small runs that it is only double that or $10K.  I would guess it is probably higher but lets go with that.  If someone has a better small run 28nm wafer cost let me know. 


300mm wafer has 70,685 mm^2 (Pi*150^2).  Lets look at the most advanced chip with known specs, bitfury.  I am just using bitfury because it is the chip with smallest feature size and highest performance that has actually been demonstrated. The die is estimated at 14.44mm and it gets 2GH/s nominal.  The same chip at 28nm would be 14.44 * (55/28)^2 = 3.74mm.   Now for the cost per GH/s of silicon fabrication the size of the chip doesn't really matter.  This hypothetical Bitfury28 could be produced as lots of 3.74mm 2GH/s chips or fewer 14.44 8GH/s chips.  Regardless the cost per GH/s is roughly the same.

 So an entire wafer (assuming no loss due to incomplete chips or defects) would be 70,685 mm^2 / 3.74mm^2 * 2 GH/s = 37,690 GH/s.  Simply put we are estimating that all the chips on the wafer will have an output of ~37 TH/s.  The real number will be lower because there is wasted space on wafer but for an estimate this is close enough.

At $10K per wafer you are looking at ($10,000 / 37,690) ~$0.26 per GH/s just for the silicon.  You still have the cutting, packaging, testing, defect losses, and shipping costs.  Now that gets you a box of working chips.  You now have to put them into boards so you have BOM (balance of materials), company profit, labor, defects, etc.  If you want to ignore the small stuff look for high capacity high quality DC to DC Power supplies as one expensive component.  I don't see how you are getting assembled ready to user miners at less than $1 per GH/s.*  

Prices will come down but that is just silly.  Now don't take the numbers as gospel but they should provide a ballpark look at silicon costs.

As for Intel that is just a red herring, Intel doesn't use foundries, they build their own fabs for internal use from the ground up so they are getting both the chip designer AND the foundry's share of the profit.  

* This assumes no chip that is radically more efficient (GH/mm2) than competitors, improved process (14/20nm), or larger production runs (think $100M in chips) to bring down fabrication costs.

What do you mean it's def. possible. Look at intels numbers they sell chips for $100 dollars, or $30 atom processors to OEMs. Now of course these guys don't have that type of volume but it's still very doable. Perfect case in point is with Avalon batch 1, they sold them for like $1200 dollars and one module had hundreds of chips on it.

Keep in mind that fixed costs are high but marginal costs are low, so once the chip is designed it's not THAT much more expensive to put 16 chips on a board than it is to put 2, or like Bitfury design a mobo that takes 16 boards that each have a 500 GH chip on them or something. Not that I know anything about the feasibility of scaling this type of technology from a technical standpoint... This is more of a guess than a knowledge-based opinion. Maybe someone more qualified than myself can  shed some light?

Probably not.  Silicon wafer's aren't that cheap.  Even in high volume we are talking $5K a piece.   So there will always be a chip production cost.  Unless you get a design which is significantly more efficient (GH/s per mm) at the same process (28nm) you are probably looking at more than that just for chip production cost.  Of course that ignores the BOM, assembly, testing, defects, profit margin, and amortizing the NRE.  I don't know how low chip companies can go but I think $1 per GH/s is like hoping gold will go below $300 an ounce (current minimal production cost).

that would be great but is it possible?

People don't realize that multiple chips make up miners. They need to be at the magical number of 1th per $1000ish or less to be worthwhile against any competition and longevity as they seem to be hinting at.

Well they have said that the chips are 500 GH, not the devices. Maybe they'll throw 16 of them in a box and charge 7k for it? Don't really see how they will generate much in the way of orders otherwise.

I think it is going to be too late for a 500GHs device in late Q4

Yes Nothing To Show and No Proof

But probably it has a fee on foreign transactions