CoinTerra announces its first ASIC - Hash-Rate greater than 500 GH/s

[Puppet]

They should offer even lower prices now. I want to buy a Terraminer IV. Gimme $2/GH and I'll get 10 units or something.

While it may be possible for them to sell you chips at a lower price if they choose to, i really doubt they can sell any 'systems' lower than $3/GH because of all the other stuff thats in the box (case, power supply, cooling (fans, radiator, pumps), pcbs, controllers, sockets, cables etc).

LOL. You dont really believe that do you? You can buy barebone PCs which contain far more PCB, cables, connectors and stuff for $150 in retail. A liquid cooling kit from Intel costs $45 retail.

[aerobatic]

LOL. You dont really believe that do you? You can buy barebone PCs which contain far more PCB, cables, connectors and stuff for $150 in retail. A liquid cooling kit from Intel costs $45 retail.

Of course you can buy a barebones PC for cheaper.  but we're not talking about a PC here.  A PC doesn't hash at the performance we want.   We're talking something 500-1000 times faster at hashing than the fastest PC you can buy.   Youre living in fantasy land if you think 'cheap' pc components are whats inside cointerra's box.   While i think youre on the right track with pc components, i think you have aimed very low on the spec.

For a start, every 28nm asic company's hashing chips run a lot hotter than any Intel chip.   And in this box we're talking 4 hot chips, not just one thats in a PC.   Intel's $45 liquid cooling is designed for 125 watts TDP.  Presumably its got a bit of extra margin in it, but its not designed for the 250-350 watts that the cointerra chips will be putting out - 24/7 in the bitcoin hashing world.  You'd need industrial strength liquid cooling not just cheapo liquid cooling.   Intel's not a purveyor of the best cooling solutions and never has been.  For those, you need to buy from specialist third parties. like CoolIT and AseTek (who make the Corsair and NZXT liquid cooling systems for instance)

Then there's the power supply - another expensive thing you underestimated.  Add up 4 hot chips running at say 300+ watts... And you get to what?  1200 watts minimum?  And what if we want to overclock them?.. Shall we allow for up to 1600 watts?  And what about the other stuff thats in there... the fans, pumps, controller and whatever else might be inside?   Shall we say its closer to 2000 watts max?  Or more?   Now look on newegg and find a couple of psu's that delivers that wattage and tell me you can buy them for $50.  Im seeing power supplies in the $300-400 range that deliver that power.  And 4 cooling systems is north of $300 too.  So we're on $600+ and we havnt included a controller nor any fans yet.  Nor a case.   In fact, im not even sure i can buy all the bits that we need for under $1000 (not including cointerra's hashing chips), so in short, youre way off base talking $150 cost.   Do the math!

[Puppet]

so you think it costs 1000 plus asic? in that case I'm not even going to.bother disprove you. what do you think the production cost of such an asic is?

[aerobatic]

so you think it costs 1000 plus asic? in that case I'm not even going to.bother disprove you. what do you think the production cost of such an asic is?

its not the production cost of the asic thats the problem, its the NRE and mask setup costs.  The fab charges these costs before it makes a single chip.  These have to be amortised by the number of chips you're going to make and is the main reason that these asic cmopanies charge for pre-orders - so that they know how many chips to order, and so that they have enough cash to place the order.  If we assume that the design cost is $1-2m, and the NRE for 28nm is $3-4m, then you have to amortise $6m over however many chips you make, and in the bitcoin mining world - and with these high performance 28nm asics - the volumes are relatively low.. so youre talking thousands not hundreds of thousands of chips.

lets say for arguments sake, you only make 6,000 chips in its lifetime (in cointerra's case, with each chip being 0.5 TH, thatd be a total of 3 Petahashes).  Theyve announced 2 PH so far at least so 3 PH isn't a bad estimate)

lets say that youre amortising $6m over 6,000 chips, so that means your NRE & other design costs are about $1000 per chip.. plus the manufacturing cost, which is probably a couple of hundred per chip (or more, if you pay extra for the 'rocket run' where they make them in 2 months instead of 3 months and charge you a lot more)

So clearly, it makes sense to make more chips and spread the NRE costs over a larger number of chips, but anyway, i hope im showing you that at $3/GH for complete systems.. theyre really not making much margin and it explains why most of the others are charging more like $6/gh.. and for them to get to $2/GH selling systems is practically impossible (unless they decide to sell bare chips)

[Puppet]

spare me the economics 101. you said they couldn't sell it any cheaper because of "all the stuff that's in the box", not because  of sunk costs.  in reality, they can produce these boxes for little more 10% of their price., and sooner or later they will (have to). sunk costs won't change a damn thing about that, they will only factor in their profits/losses.

[aerobatic]

spare me the economics 101. you said they couldn't sell it any cheaper because of "all the stuff that's in the box", not because  of sunk costs.  in reality, they can produce these boxes for little more 10% of their price., and sooner or later they will (have to). sunk costs won't change a damn thing about that, they will only factor in their profits/losses.

when youre making low volume production, the cost of the NRE is your biggest cost.  if you ignore that, how do you even afford to make anything?  it doesnt just factor in your P/L.  its a cost.  and its an up front cost.  its not a 'sunk cost'

[Puppet]

it's a sunk cost all right, and easily covered by the first round of preorders.

[aerobatic]

how many pre-orders do they need to sell to cover their cost?

[Puppet]

probably somewhere between 2 and 5 million dollar. first batch is 2 PH, do the math

[willphase]

I think it would be nice if cointerra could confirm that tapeout has begun.  I know they don't post on these forums, but perhaps a blog post or newsletter would be good instead?  Or a tweet...?

Will

[ur0pl]

They should offer even lower prices now. I want to buy a Terraminer IV. Gimme $2/GH and I'll get 10 units or something.

They have the best team of elite engineers out of all the asic mining companies. That is what you are paying extra for, not for hashing power.

[iCEBREAKER]

They should offer even lower prices now. I want to buy a Terraminer IV. Gimme $2/GH and I'll get 10 units or something.

They have the best team of elite engineers out of all the asic mining companies. That is what you are paying extra for, not for hashing power.

Frankly sir, that's bullshit.

You are lapping up slick marketing, not analyzing actual facts.  Last I checked, a fancy pedigree hashed at zero GH/sec.

If they really "have the best team of elite engineers out of all the asic mining companies" why

-  did they make claims about the physical performance of their chip (GH/W/sec) before logical design was even complete?

-  have they fallen behind schedule on their supposed "first week of October" tape-out?

Did they even raise the $5 million needed for 28nm NRE?  Or did ScamGarden, LABscam, ACTIVEscamming, and IceDrill IPOs exhaust investors before Cointerra (and CryptX) had a chance to get started?

[aerobatic]

You are lapping up slick marketing, not analyzing actual facts.  Last I checked, a fancy pedigree hashed at zero GH/sec.

If they really "have the best team of elite engineers out of all the asic mining companies" why
-  did they make claims about the physical performance of their chip (GH/W/sec) before logical design was even complete?
-  have they fallen behind schedule on their supposed "first week of October" tape-out?
Did they even raise the $5 million needed for 28nm NRE?  Or did ScamGarden, LABscam, ACTIVEscamming, and IceDrill IPOs exhaust investors before Cointerra (and CryptX) had a chance to get started?

i guess we're hoping we'll hear more news this week (and if theyre behind by a few days, well thats annoying but also no different than anyone else)...  and yes, its my understanding that they have higher performance and lower power consumption than their peers, and that it was validated and simulated during the physical design
process

[bobsag3]

You are lapping up slick marketing, not analyzing actual facts.  Last I checked, a fancy pedigree hashed at zero GH/sec.

If they really "have the best team of elite engineers out of all the asic mining companies" why
-  did they make claims about the physical performance of their chip (GH/W/sec) before logical design was even complete?
-  have they fallen behind schedule on their supposed "first week of October" tape-out?
Did they even raise the $5 million needed for 28nm NRE?  Or did ScamGarden, LABscam, ACTIVEscamming, and IceDrill IPOs exhaust investors before Cointerra (and CryptX) had a chance to get started?

i guess we're hoping we'll hear more news this week (and if theyre behind by a few days, well thats annoying but also no different than anyone else)...  and yes, its my understanding that they have higher performance and lower power consumption than their peers, and that it was validated and simulated during the physical design
process

Simulated.
Wake me up when we see a real chip... KnC still hasnt delivered on their w/ghash, so why would anyone believe a non-taped out simulation?

[iCEBREAKER]

i guess we're hoping we'll hear more news this week (and if theyre behind by a few days, well thats annoying but also no different than anyone else)...  and yes, its my understanding that they have higher performance and lower power consumption than their peers, and that it was validated and simulated during the physical design
process

Cointerra has barely started their logical design process.  They can't write RTL worth shit and it's already taking longer than said though it would.

We don't even know if the $5 million NRE has been paid yet.

See how happy they were to get a simple FPGA (logically emulating one or two cores) working?  HashFast was at stage way back in June!

Cointerra is nowhere near finishing their physical design process.  You can't do simulations of the physical design until logical design is complete!

All their specs (logical and physical) are ballpark estimates at best and wishful thinking at worst. 

Cointerra just looks at what HashFast is doing, and (claims to) one-up them without actually following through.

Show me the tape-out!

[cypherdoc]

not only that, their design team is spread all over the world with the main office in India.

coordination is difficult.

[aerobatic]

Cointerra has barely started their logical design process.  They can't write RTL worth shit and it's already taking longer than said though it would.
We don't even know if the $5 million NRE has been paid yet.
See how happy they were to get a simple FPGA (logically emulating one or two cores) working?  HashFast was at stage way back in June!
Cointerra is nowhere near finishing their physical design process.  You can't do simulations of the physical design until logical design is complete!
All their specs (logical and physical) are ballpark estimates at best and wishful thinking at worst. 
Cointerra just looks at what HashFast is doing, and (claims to) one-up them without actually following through.

Show me the tape-out!

its not even worth arguing with you, every statement youve made is plain wrong (troll-worthy).  tell you what, lets just do a bet.  you tell me how much you want to bet that cointerra wont tapeout within the next X (shall we say 10? days..  if not, you decide the number of days) and i will seriously consider taking your bet.  we are expecting them to tapeout imminently.  yes, they are behind by days from when they said they would, but i dont think theyre as far behind as you are making out.

you know as well as i do that by the time of imminent tapeout, they will clearly have got far enough in their physical design process to be pretty confident of their power consumption and performance figures as accurately as anyone else could know theirs.

You are also under the mistaken impression that the asic design must follow the fpga design - which in knc's and hashfast's design was the case, but in cointerra's example   the fpga actually came after their asic design was complete and was used for a different purpose - not to prove the design - though it did that - but to debug the software effort.

you really think cointerra cant write rtl code worth shit?  thats your honest opinion?   based on what?   Do you think cointerra's specs have been pulled out of their ass?  dont you think there's every chance with the background and experience these guys have that theyre professional asic engineers who know what theyre doing.   and they appear to have performance and power consumption figures that are materially better than their peers.  they didnt achieve those by accident now, did they?  if their competitors couldve hit the same performance, dont you think they would've?

(cypherdoc, cointerra's rtl design was done in austin, its the physical design that was done in bangalore)

[RoadStress]

You are lapping up slick marketing, not analyzing actual facts.  Last I checked, a fancy pedigree hashed at zero GH/sec.

If they really "have the best team of elite engineers out of all the asic mining companies" why
-  did they make claims about the physical performance of their chip (GH/W/sec) before logical design was even complete?
-  have they fallen behind schedule on their supposed "first week of October" tape-out?
Did they even raise the $5 million needed for 28nm NRE?  Or did ScamGarden, LABscam, ACTIVEscamming, and IceDrill IPOs exhaust investors before Cointerra (and CryptX) had a chance to get started?

i guess we're hoping we'll hear more news this week (and if theyre behind by a few days, well thats annoying but also no different than anyone else)...  and yes, its my understanding that they have higher performance and lower power consumption than their peers, and that it was validated and simulated during the physical design
process

Simulated.
Wake me up when we see a real chip... KnC still hasnt delivered on their w/ghash, so why would anyone believe a non-taped out simulation?

Hasn't is the right word. They over delivered!

[aerobatic]

Simulated.
Wake me up when we see a real chip... KnC still hasnt delivered on their w/ghash, so why would anyone believe a non-taped out simulation?

Hasn't is the right word. They over delivered!

Agreed, KnC originally promised 2.5 watts/GH, then it was promised at 1.4 watts/GH.. and now in the final version, theyre pretty much on 1 watt / GH.  so theyve done well and managed to lower the power consumption of the final design (bucking the trend set badly by BFL).

[iCEBREAKER]

its not even worth arguing with you, every statement youve made is plain wrong (troll-worthy).  tell you what, lets just do a bet.  you tell me how much you want to bet that cointerra wont tapeout within the next X (shall we say 10? days..  if not, you decide the number of days) and i will seriously consider taking your bet.  we are expecting them to tapeout imminently.  yes, they are behind by days from when they said they would, but i dont think theyre as far behind as you are making out.

you know as well as i do that by the time of imminent tapeout, they will clearly have got far enough in their physical design process to be pretty confident of their power consumption and performance figures as accurately as anyone else could know theirs.

You are also under the mistaken impression that the asic design must follow the fpga design - which in knc's and hashfast's design was the case, but in cointerra's example   the fpga actually came after their asic design was complete and was used for a different purpose - not to prove the design - though it did that - but to debug the software effort.

you really think cointerra cant write rtl code worth shit?  thats your honest opinion?   based on what?   Do you think cointerra's specs have been pulled out of their ass?  dont you think there's every chance with the background and experience these guys have that theyre professional asic engineers who know what theyre doing.   and they appear to have performance and power consumption figures that are materially better than their peers.  they didnt achieve those by accident now, did they?  if their competitors couldve hit the same performance, dont you think they would've?

(cypherdoc, cointerra's rtl design was done in austin, its the physical design that was done in bangalore)

Cointerra said they were taping out the "first week of October."

That is a fact, not an opinion.

It's now the second week of October.  Again, that's a fact, not an opinion.

Here's another fact: we have not heard a peep out of Cointerra, much less seen a press release from OpenSilicon (their supposed physical design partner) about NRE and/or tape-out.

Now let's move from facts to my personal opinions.

I already explained why making claims which can only be validated by physical emulation are wishful thinking when logical design has yet to be completed.  Therefore their numbers are being pulled out of their ass.

No, I don't think Cointerra can write RTL worth shit. 

Their expertise is in low-power, low gate activity ratio SOC chips for toys like phones and set-top boxes.  A completely original (not 99% Samsung-licensed ARM IP) massive billion transistor Bitcoin ASIC is beyond their competence level.