Note by Chris:
Despite my love to keep things private as long as possible, especially waiting VTC hard fork, it seems it is not possible this time.
I got a bit forced to move on. I will try to keep it as concise as possible (almost impossible job).
Basically I gave you hints about a second product, yes, it’s indeed another ASIC.
Well, not like the others, it’s different by several key attributes which were selected after this years of experience in farm mining.
First of all, it’s not a 28 nm chip. If one retails its chip, of course, it needs the highest density, the best power consumption ( at maximum speed), and maybe the fanciest technology (sells better). He needs to pack as much power, which ultimately gives product price per unit of volume, to get profit.
However, for someone not looking to sell hardware but use it for farming, density is not real a key element, power consumption may not be either an issue if chip can be throttled to get better efficiency, and farm may get better price for electricity than end user.
There is another issue when moving to lower technology nodes, since development price and time increase, chip price increase and runs outputs are finite, designers will tend to pack maximum of logic in minimum on space. This may result in several draw backs like poor thermal performance, poor electrical performance, etc. (these are basic).
This may result in other design decisions. When hashing BTCs things were simple, SHA256 is a simple algorithm which moves things around without big memory requirements.
This was quickly exploited by new coins designers, which added RAM requirements to their algorithms (or selected algorithms with bigger RAM requirements). The key thing is that, if they want to have an ecosystem around their new coin, they will need to stop arbitrary increasing the RAM size to the point where CPU or GPU can still hash their coins (and this is a good thing for alternate hardware designers, since coin designers can’t run wild with RAM requirements, if your alt hardware behaves at least like a CPU, he will hash at least like a CPU – of course, it does not need to implement all instruction or registry set, or memory management, etc. Same stands for GPU like).
The problem with 28 nm or less is that memory IP (adding arbitrary memory to chip) is somehow expensive in terms of area (which renders in hash density per chip).
So 28 nm (close or less) will most likely use outside RAM and high density logic. When doing so, they will most likely use commercial level packed RAM like regular DDR3 chips or similar (which, btw, technologically are not 28 nm).
By doing so, their chip enter another trap, the DDR ram has a big issue, while it is perfectly able internally to access data at max speed, the job itself of accessing DDR data at max speed from outside (from a connected CPU or ASIC) requires complicated ASIC hardware/firmware tuning algorithm based if speed is a requirement, this means that usually ASICs with external RAM may get some penalty when attempting to change algorithm (even if ASIC allows changes).
Even if this expensive ASICs have the ability to arbitrarily reconfigure themselves, they will ultimately ran into bandwidth issues which will return as poor performance.
To cut things short, instead of chasing most expensive and advanced technology, we simply turned orthogonally and chosen a design around a memory IP and a fab specialised in memory solutions and added sort of reconfigurable logic around (almost the same kind as other designers). It still can access outside memory, but current design don’t use it for costs reasons.
It’s not the newest technology.
The new ASIC won’t be the king in Litecoin, but it will survive in a farm, at the same time is doing the best job in the jungle of new algs with low-med difficulty. All this at a decent price.
And now to return at the starting point, I got informed on short notice that our fab will perform next run on this tech earlier than expected, in mid January, then will get two months of hiatus (March).
The first plan was quite different, but it changed, will list some pre-asic shares for those wanting to grab this product.
The original plan was to get stock ourselves and replace existent shares via payout conversion, marketing events and other similar ways, but there is not enough time left to move on this way
This means there is another ASIC generation coming??