It's gonna be very interesting to see how KnC will pull 100GH/s chips at ~250W each.
First, without using liquid cooling you need a heatskink+fan the size of a graphics card, for each chip, to have the necessary surface area for that much dissipation.
Second, I think it's a very very bad move to have big 100GH/s chips like that. What if one breaks??? You lose 100GH presto, especially on systems running 24/7!
Third, I'm still puzzled why are they only getting 0.4GH/s/W on 28nm? Even bitfury gets much better than that on 45nm and the Fast-Hash-One is more than 1GH/s/W too.
First, without using liquid cooling you need a heatskink+fan the size of a graphics card, for each chip, to have the necessary surface area for that much dissipation.
Second, I think it's a very very bad move to have big 100GH/s chips like that. What if one breaks??? You lose 100GH presto, especially on systems running 24/7!
Third, I'm still puzzled why are they only getting 0.4GH/s/W on 28nm? Even bitfury gets much better than that on 45nm and the Fast-Hash-One is more than 1GH/s/W too.
While this thread isn't about KnC, it's about VMC I'll entertain you.
The Heatsink/fan will be rather large for each chip, but it fits inside the dimensions of the case they've been advertising for a long time now so I'm not sure how this is relevant.
Chips don't usually break.... They also have a warranty on them. I'd say it has a similar chance of breaking as your GPU miner (which will be running much hotter more than likely)
They don't want to under deliver. When the actual chip comes out and they test it, it will more than likely be a lot better than that. But in the mean time, do the math on how much you will be spending on Electric. It will come up to be less than $85/month difference if in fact it is 1,000W vs 400W. By the time this small amount matters there will be next gen ASICs available.
Bigger dies are actually EASIER to cool than all the little small chips.
The only thing that matters is surface area vs heat dissipated. It has nothing to do with the amount of chips. It is much harder to dissipate 250W of heat in a 55x55 area than 250W of heat in many small chips, where the total surface area is bigger. The smaller the total area, the more quality is needed in chip manufacturing, thermal interfaces, etc. A recent example of this is when Intel changed to TIM on Ivy Bridge heat spreader vs Fluxless solder on Sandy Bridge which was much better. Unless the cooling is top notch, small surface areas will always work at an increased temperature (this is pure thermodynamics, as the thermal resistance is higher), reducing operating life due to increased electromigration, especially on 24/7.
It will only have the same chance of breaking as a GPU miner if they are manufactured with the same quality constraints as a GPU miner. Let's hope they don't cut any corners then.
Energy efficiency is what matters in the end, since any system will mine at a profit until it doesn't pay the electricity bill anymore. Using ~12W chips, at 1000W, you can get 1,333 TH/s.
This is 28nm. There is no 'next-gen' for the time being until smaller die sizes are available more generally. Delivering a 'next-gen' product on 28nm again just means the first one wasn't optimized enough. The main advantage of going now to 28nm is some kind of minimal future-proof.
here's the heatsink they're planning on using. (see here)
As you can see, they're gigantic, even larger then a typical GPU cooler and rated at 320 watts cooling capacity.
As far as breaking a chip goes, they do have a warranty, they haven't given any details on it but it would surprise me if they'd be willing overnight you a new module if it breaks, but it also seems unlikely that one would just break once you got it up and running, chips typically either work don't.
The chips are also designed with multiple domains that can be disabled if they're not working
As you can see, they're gigantic, even larger then a typical GPU cooler and rated at 320 watts cooling capacity.
As far as breaking a chip goes, they do have a warranty, they haven't given any details on it but it would surprise me if they'd be willing overnight you a new module if it breaks, but it also seems unlikely that one would just break once you got it up and running, chips typically either work don't.
The chips are also designed with multiple domains that can be disabled if they're not working
Thanks for the links.
Considering everything, their manufacturing cost will have to be quite high for each unit, as all the eggs are really in the same basket.
