Topic: CoinTerra announces its first ASIC - Hash-Rate greater than 500 GH/s - page 86. (Read 231002 times)

i suspect bitcoin asics.. particularly those from cointerra and hashfast, will be amongst the highest wattage per mm that you would consider.. since they're both designed to be clocked at near the max... and scale up or down depending on thermals.   they're both probably designed to have a TDP of 250+ watts !

Actually, PC cooling is not designed to cope with maximum theoretical load. Its designed to cope with application loads. There is a substantial difference, and thats what TDP is for (or worse, ACP if you are called AMD). The maximum power consumption of almost all CPUs and GPUs is above TDP, but no real world applications will sustain this power draw and your system usually doesnt melt even when running a thermal virus (think furmark) because the CPU/GPU will throttle if need be.

A well designed bitcoin asic would act much like a thermal virus, running as close to maximum power draw as possible, yet throttling a bitcoin asic is about the last thing you would want to do, except when there is cooling failure.

That said, per watt, cooling a bitcoin asic is in all likelihood, easier than cooling a highend cpu or gpu, because the thermal density of hotspots is substantially lower. A typical desktop cpu will have hotspots consuming 80-90% of the chip's TDP in an area no bigger than ~5% of the die, think needle tip.

While bitcoin asics overall may have a comparable "tdp" per mm², the power consumption will be much more homogeneous across the entire die and hotspots will have a far lower thermal density than CPUs.

Damn straight, heres my 64 thread quad Intel rig.

All the components in a PC are absolutely designed to run flat out 24/7, as they will if you have to render a video or do something else processor intensive for a long period of time.

i don't recall if they announced it, but they told their customers (including me).  but thats not unusual.. most of the recent asic vendors will have paid for a rocket run or super rock run, or whatever the other names for it are.  there's no way to get silicon faster than 60 days unless you do pay extra... and the more you pay, the faster it is.

-- Jez

So Cointerra gave a vague delivery date and you know they chose a conservative timeline with lots of room for slack and errors.
Other companies gave a vague delivery date and they chose an aggressive timeline with no room for slack and errors.

Cointerra is different why?  You want them to be?

Cointerra has announced they paid a premium for a rocket run or this is an assumption?
Usually the pricing and timeline of fabrication is subject to NDA so I would be surprised if they announced that.

i think the short answer is Yes - Cointerra announced a quite vague December date for their first batch (i think they said late dec) and presumably they gave themselves the whole of december as their margin for things to be delayed and still meet their timescales... probably wise, whereas some others gave the earliest possible date as their target date giving them no room for inevitable slippage outside their control.

they had paid a lot extra (super rocket run at glofo) to expedite the december batch, which is a small batch versus the january one thats much cheaper to produce (no expedite fees for silicon nor rush shipped parts) thus the jan customers got a much better deal than the december customers in terms of price - at the expense of delivery date - but also the jan customers have a lower risk of slippage than the dec ones - and I'm pretty sure many december customers switched out their orders into january ones to take advantage of the extra cost savings ($3/gh in jan versus $7/gh in dec).  for the given delivery dates both prices were competitive (i still don't see anything else in dec or jan at anywhere near those prices) but the jan ones for most people were the 'better deal'.  (I've got a small order in the dec batch and a much larger order in jan to take advantage of the significantly better price per gh)

although theres not a huge difference, I've personally bought all my mining gear with fiat because then i can measure the resulting roi in fiat and feel happier with the decision (bitcoin price rises help achieve roi) as the variables for trying to buy in bitcoin and measure roi in bitcoin could be frustrating with the network difficulty rising so fast.  i have also been buying bitcoins directly, as a hedge.  some would say thats the best exposure, but if you want to mine as well and be part of the bitcoin ecosystem, then buying mining gear that will be low cost per gigahash and low power for long-term usage, is the way to go.

-- Jez

So, if I buy a CoinTerra 4 miner and pay using fiat, that's okay? (Or I can pay in bitcoin, bought using fiat at today's rate, so my basis is still fiat.)

Or would it be a better idea to buy bitcoins now using my fiat and just hold? See, mining continues, but holding bitcoins keep the bitcoins as is, the value just changes. I just need to mine a little bit more bitcoins than I would have bought and hold to make it worthwhile to buy a CoinTerra4 miner.

knc and hf are with tsmc, ct and bm are with gf.  lotsa acronyms ;-)

I guess my point wasn't that CT could have a major delay obviously a major delay now kills any chance of an on time delivery.  I was just pointing out even with perfect execution they may be limited by the schedule.  We have no idea what CT contract with TSMC GF is.  Maybe TSMC GF is saying wafers in 10 weeks so there doesn't need to be a slip they just never were going to be faster than that.  That wouldn't be particularly unusual but due to the tapeout delay that puts it mid Dec and maybe their assembly house doesn't have the staff and scheduling the week before Christmas to produce 10,000 (or however many they sold) boards.   So I don't think it is a foregone conclusion that they need another stumble to miss their deadlines.  The tapeout stumble may have already done that and it will just play out in agonizing slow motion over the next month.


The last orders were more like two weeks late not a few days but still I agree they have done better than just about anyone else except maybe Bitfury.

As is January for Chinese New Year. This is a very bad time of year to try to launch new electronics.

KNC were good, but far from perfect.

i always put 'time will tell' or similar words... at the end of my conjectures... since i agree, something that has gone wrong with one asic provider, may also go wrong with other ones (or different things could go wrong.. there are a lot of variables).  although the later asic companies have the ability to learn from earlier ones and try not to make the same mistakes (and make different ones ;-)   also, i think that this highlights some of the different strategies of both hf and ct for their choice of partners.  hf used a more nimble physical design partner - and worked much more closely - to excellent effect - whereas ct used a more slow and conservative partner whose staff were working much further away and this slowed down the process.  on the face of it, the more nimble people that hf used should've been a huge win for them - but the negative to that was they weren't a full service company and had specific areas of competence.. so the things that ct didn't have to worry about (like choosing and designing substrates, which were done by their partner) became  more of a problem for hf (i don't know exactly why it was such a problem!).   knc also had the benefit of a full service physical design partner, and knc also managed to avoid falling prey to substrate problems or other supply chain problems.. (all three of them use more or less the same design methodology of fitting a lot of cores on a die, and then mounting multiple dies on a substrate in a single package designed to run hot and be well cooled).  the three architectures are remarkably similar.

its clear that any of the other asic companies could stumble at any time during their production process.  hashfast was always held up as a model of speedy and aggressive execution, but seems to have suffered from the parts of the system that shouldn't have caused problems, and imho, they've been very very unlucky!  since only one other asic company has so far done anything close to what hashfast is attempting to do - namely, KncMiner - you can only hold them up as an example of a company that did get everything right.  they executed almost flawlessly.  they delivered within days of when they said they would.. and had everything ready to go... the chips, the software, the hardware, the boxes, the logistics... and even the hosting (a few days late but still done).  When people as smart and hungry as hashfast missed their dates, it makes me feel very sorry for them.. but it also makes me feel even more appreciative of what knc managed to do.  (yes, i have knc systems as well... i have a very heterogeneous mining operation)

Icebraker's nonstop trolling aside which is both annoying and excessive, you say "instead of months between them, there may now only be days or weeks".

Key word is "may", CT is already 30+ days behind tapeout.  Did their original Dec timeline have 30 days of slack?  Were they able to secure a 45 day rocket run like KNC or are they looking more like getting raw silicon back in 70 days plus another say 20 for packaging and another 10 for assembly and volume shipping (KNC didn't really get any shipping volume until 4th or 5th day).  So I am not 100% convinced that they absolutely will make Dec even if they do manage to avoid a major delay.  Dec even with no delay might be predicated on an early Oct tapeout. 

If anything HF stumble gave them a chance of closing the gap, it is entirely possible their OWN stumble ensures they won't.  One thing which may hurt but HF and CT is end of Dec is a flaking time of year.  In manufacturing industry they often have block vacation, reduced hours, shipping timelines slip because FedEX and UPS are at 300% of normal volume.  It isn't just the companies themselves but also every company the companies rely on as well.  Both CT are now in that boat but HF as least still has the possibility (we will see) of getting units out the door by mid Dec and bypassing at least some of that.  CT at best is going to be right in the middle of it.

iceBreaker, i keep reminding you, hashfast and cointerra have different products and price points... and originally they had significantly different delivery schedules, but events have conspired to push them much closer together than anyone could've imagined.

i think KnC and Hashfast have both proved that there's a lot more to shipping on time than just having your silicon back quickly...  and that you need everything else to go right - not just the silicon - to be able to ship on time.  Hashfast has (alas) now announced a two month delay - which has effectively thrown away the lead over cointerra/bitmine/blackarrow that they had, and will be now start to ship in mid december.

Although Hashfast taped out two months earlier than Cointerra and probably a month or so before BitMine and BlackArrow, they should've been one to two months ahead with delivering boxes... and yet, the other parts of the supply chain caused their current woes.   Its a real shame, as their silicon is more or less on time (ok, two weeks late maybe) but the rest of the parts of the system are letting them down and have affected us all (I've got orders with hf and ct in batch 1, so I'm very eager to see them both ship as early as possible and i had assumed that hf would be significantly earlier than ct and now i think there'll be a lot less between them).

Cointerra are doing everything they can to keep their schedule on target, which means having all the non-silicon parts of the supply chain - the boxes, boards, power supplies, and all the components that go onto the board ready to go (just like KnC did) so that the second the chips arrive back from the fab (in late dec?) they're able to test, box and ship.    If ct's execution goes as smoothly as KnC's did, that should mean ct are days and not weeks late on their announced timeline (by End of December).   We shall see if they can keep it up...   Only time will tell if CT execute more like KnC or more like HF...  !

Icebreaker, you can stop your trolling now.  instead of months between them, there may now only be days or weeks separating the deliveries of hf and ct gear.  I'm still expecting hf to deliver first.  things will have gone very wrong if they don't.

one things for certain, both hf and ct have learned a LOT from the experience and won't make the same mistakes on their second generation chips...!  they both made mistakes, and both could've been a LOT earlier than they are, had things gone well for them.

Too late for that.. this is just as good

Not really.

Cointerra is over two months behind HashFast, and hasn't even hired a more competent design team for their next chip.

Cointerra's first chip will be competing with HashFast's second one, sometime in March/April.

Meanwhile, TSMC will be pumping out multiple batches of HashFast G1 wafers.

It makes sense why the February batch does not get reduced in price -- look at the price of bitcoin.  With bitcoin at almost 3x of what it was just recently, the profit will be 3x as much in fiat.  Anyone who is in line for Dec and Jan are now holding preorders that are beginning to gain value above the original price.  In fact if they kept the price the same for February, and the difficulty rises 100% between jan and feb, then that is a value double that of the previous month batch.  I bet i'll make my ROI on my january order within a month of when I receive it, even with ~15TH/S network hash rate.. only because its 3x easier now with this BTC price.

btw, cointerra has announced their tape-out.  more details on their web site.