Topic: PCB Design for Cointerra (Read 2988 times)

Agree with b 100%. While a single chip .5Th/s solution sounds great, the heat density is an engineering challenge that didn't need to be tackled when time to market is what really counts. While all the 28nm vendors to date are going to have to expend resources working on this issue, a passively cooled bitfury rig at 55nm is now the efficiency king and will likely remain so for at least a few months and if they can shrink that die and decrease power consumption further then they'll really put a hurtin' on the other vendors. I'll take their pedestrian looking "deck of cards rig" with it's impressive specs now over an aesthetically pleasing rig with a necessarily elaborate cooling system next January. Actually I'll take two.    

Ok ok....

Even if all the power will transform into heat, using the Cointerra chip will is by far better than a bunch of avalons or bitfurys.

Over that less 1 chip replaces 100+ of the others probably the assembly cost is much much lower.

I still think Cointerra chip will be the best option of the current ones, and we should have a way to develop our PCB cards to just buy the chips, I think we may save some money and also reduce the time to market.

All the best

TMC

Agree!  if they say its 1200 watts, then its 1200 watts of heat that needs to be exhausted all the time.  Assuming it has four chips in the box, then thats 300 watts per chip, which would require an excellent cooling system to be able to run 24/7 (but certainly not out of the area of high end pc cpu coolers)

This is a law of thermodynamics, you can't disagree with it

Here is what confuses you: a CPU advertised with a thermal dissipation of 100W will actually draw anywhere from 0 to 100W depending on the load. It does not constantly draw 100W. You can measure this with a clamp-meter on the 4-pin ATX12V cable. So if average computer usage leads to the CPU drawing 40W, then on average you will only observe 40W of heat coming out of a "100W" CPU.

A Bitcoin mining chip, unlike a typical electronic chip, runs at full load 100% of the time. So a Bitcoin mining chip with a thermal dissipation of 100W will always exhaust 100W of heat.

The outputs are going to be heat, voltamps and photons. Almost all will be heat, only a tiny amount will be used to send electrical signals or give of EM radiation unless it's an LED or radio amplifier.

Sorry but I disagree.

is 300W always generate the same heat, we have no problem because the heat generated cant be more than any other "chip" (cpu, video, digital converter) that uses 300W, so if what you say is valid is valid the same cooling capacity to cool any chip that uses 300W is the same... and is not. at same power usage different chips generate different volumes of heat.

I agree energy transform in something else 100% but is not all heat, also I also agree the Cointerra chip per GigaHash looks will use much less energy than Avalon, Bitfury and all the rest.

All the best

TMC

They may not have even decided on the pin-outs and packaging yet.

As an experienced PCB designer, I would like to get a datasheet from CoinTerra before designing a PCB, preferably with a reference design.  If they release a datasheet, and the project is feasable, I would be happy to make a PCB for it, but the design needs to be there.

I think his point was that it might actually use less then 300W of electricity.

Cointerra emailed me and their plan is to support 3rd party manufactures of boards for single chips. Who knows what that really means.

300W of electrical consumption will always generate 300W of heat, no matter how efficient or inefficient the chip is. This is the law of conservation of energy.

Anyways, I think, cointera is an expert, and they know what they are doing..

I think Cointerra chip may not require extraordinary cooling.

Remember heat is wasted energy depending how efficient is his chip is the heat, Until we can see one in action we cant say much I guess.

TMC

aerobatic, yes Bitfury's 400 Gh/s rig would need fans if put in an enclosed box. But three 80mm fans (spanning the width of the m-hash board) to push air horizontally is still "easier" than the heavy HSF or fancy water cooling system that a 500 Gh/s cointerra chip will require. You made a specific claim about ease of cooling a big hot chip, so I responded to this claim precisely.

I do agree with the rest of your post, that a system with a single big hot chip might be cheaper than numerous smaller chips (less PCB, fewer components, etc).

Bitfury (2W per chip) and Cointerra (300W per chip) are the extreme opposites of the scale. I think the best tradeoff in terms of cost and ease of cooling is somewhere in-between (20-150W per chip). Cointerra is taking unnecessary risks by aiming at the absolute highest wattage possible than can be practically cooled. They should have gone with 2 x 150W chips for example. It would have removed a huge cooling challenge, while not necessarily increasing cost that much.

I can give you an answer as soon as there are some more info about this chip.
Ref docs, chip description, samples etc...

The big, hot chips should require much smaller, and therefore cheaper PCBs (depending on the number of layers, of course) while requiring expensive heatsinks. On the other hand the small, cool chips will require more expensive PCBs that will take longer to manufacture on a given SMT line.

Cointerra is going to have to release reference PCB designs. I'm sure they will before November. Their chip seems to have thousands of pins and will likely require PCBs with several layers.

i think BitFury's chip is awesome, dont get me wrong.  im very impressed by what he's done. its a phenomenal achievement.. even more so when you hear this is effectively his first asic!  the guy's a genius!

but i think youre talking chalk and cheese.  the bitfury boards that are operating without heatsinks and fans are in open air and are very hobbyist looking things - they can only rely on convection when theyre on someone's desk in an air conditioned room.   When you put them in a box, and make them legal to own and sell, id be shocked if they didnt need fans for their airflow.

Theres no magic here.  BitFury's 400 GH set of boards consumes 400 Watts.  Youve got to move that heat out of it, somehow.  it cant be done by convection without either heatsinks or fans once its in a box.  And its not likely it will stay unboxed when its sold for more serious use.

But these are two different strategies to achieve the same end goal.  there's nothing wrong with either strategy as long as you can hit the price and the power consumption targets.

1.  The Ant colony approach.  Lots of tiny low power chips combining their strength in numbers to achieve a goal.  These guys are best demonstrated by BitFury but its the same strategy employed by Avalon, AsicMiner, CoinLab/Alydian, LabCoin, XCrowd etc.  BitFury's is by far the best in class in this ant colony model.

2.  The 'big hot chip' approach.. like KnC, Hashfast, Cointerra (and to a lesser extent, presumably AMC/VMC and the rest of the 28nm crowd).   These guys need large heatsinks and/or water cooling... because the chips are designed to be run at their maximum thermal capacity (250-320 watts, according to the specs of the cooler knc are using).  but the good thing about focussing all the power into small concentrated spots is that your heat is localised and you know where its coming from and you can employ heatpipes or liquid to extract the heat directly from the chip to the outside of the box.  you dont even need much airflow inside the box as all the heat leaves the box much more efficiently.  heat transfers more efficiently in liquid than it does in air.

the question of price and performance will rear its ugly head.

ignoring what bitfury, knc, hashfast and cointerra are actually charging for their products right now.  What do we think they will cost, per GH?

do we think that 256 low power (bitfury style) chips, packaged in their QFP48's is cheaper than the equivalent hash power using KnC, Hashfast or Cointerra's single but very much bigger die presumably packaged in an intel style ceramic packaging?  clearly the latter is MUCH more expensive than the former, in terms of a single unit.  but the former needs 256 of them to hit the same perf target that the latter hits with one expensive chip and expensive package and expensive cooler.   So which one will be cheaper to produce in the long term?   (i suspect the big hot chips with all the baggage that goes with them, will ultimately be cheaper and lower power than the large quantity of smaller lower power chips that you will need to hit the same performance target).

but its a very interesting exercise and i cant wait to see how they all fair.  and of course, all of them will be working on their mark-2 chips utilising what theyve learned from their mark-1 chips so the next round will be even more interesting!

-- Jez

Actually the opposite. Cointerra's 500 Gh/s will need a big fancy heatsink or water cooling system. But look at Bitfury's 400Gh/s rig with 256 small chips. They need neither heatsinks, nor fans, because the heat is so well distributed that air convection alone does the job.

CoinTera is awesome. 1T# chips for around $5600 in 50 chip pack. $14k for 2T# - 2U miner is also not so expensive.. looks like, complete miner + rental Data Center hosting is the way to win the game.