Yet another ASIC mining company has entered the fray – and this time, with an extremely bold performance claim. Cointerra promises the smartest chip design and the best price-performance ratio we’ve yet seen in the ASIC market. But can customers afford to wait for it?
The ASIC mining business was all about getting to market first, and we have seen some embarrassing episodes along the way, with some firms repeatedly setting back delivery dates, and revising power-performance ratios.
But designs have also been getting more sophisticated. One of the biggest innovations has been in the process node, which is the size of a typical transistor on a chip. ASIC mining chips have moved from 110 nm, down from 65 nm, to 28 nm designs. The advantage of making transistors smaller is that you can fit more of them on a chip, which means a higher hash rate in the same space, and lower power requirements. KnCMiner garnered lots of attention when it announced a 28 nm design, with units scheduled for release in September.
But it isn’t just about packing more transistors into a smaller space, says Cointerra founder Ravi Iyengar, who says that he is using his chip design experience to optimize the ASIC’s inner circuitry.
“Inside the ASIC chips, there are a number of hash engines and you need to ensure that there is an intelligent distribution of workload,” he says. “You want to ensure that you’re not duplicating them, and that you’re not keeping any of the chips idle. They have to be constantly working, with good and real work.”
Iyengar knows his stuff. He left his job as lead CPU architect at Samsung to found Cointerra. One of his favourite tricks is holding up his Android phone and pointing out that he designed the processor inside it.
Cointerra is designing its ASICs with extra tricks for power efficiency, too, he says. The upshot of it is that he hopes to get to market with the best ASIC units yet in the second half of Q4. It is promising at least 500 Gigahashes per second on a single chip called GoldStrike1 (GS1).
Much of the design innovation is still Iyengar’s secret sauce, and he is building intellectual property around it, he says, but we do know at least one accepted industry technique he’ll be using: increasing the size of each chip (known as the die size).
Each chip must be put into a hardware package that makes it usable on a circuit board. The package does important things like communicating between the chip and the board, and dissipating heat from the chip. By increasing the die size, Iyengar will also increase the amount of silicon per package. The cost of the package also increases with its size, but not linearly, meaning that a larger die size contained in a larger package will cost less than several smaller chips encased in several packages.
Techniques like this, along with his ASIC design, will enable Cointerra to produce a chip that will consume significantly less than 1 Watt per Gigahash, the company says. If he can pull it off, that will indeed give him an advantage over KnCMiner, which offers around 2.5 Watts per Gigahash.
That’s attractive, but the firm will still be two months or more behind KnCMiner with its devices. In the crazy world of ASIC mining, that’s a significant window. As these machines come on stream, they will push up the hashing difficulty on the bitcoin network, making it increasingly hard to do the calculations necessary to solve a transaction block.
“When you make your chips extremely efficient in terms of performance per area, and you make it extremely power efficient as well, then scaling becomes far easier,” he says. The idea is that it might take tens of chips from competitors to reach a Terahash in computing power, whereas Cointerra can do it by putting two in a single box – and keeping the power consumption relatively low. So, in short, if you want to offset the increase in difficulty as ASIC miners pour onto the market, then buy something scalable so that you can cram more hashes into a box.
We won’t know how much these boxes will cost until Cointerra releases pricing information, which it promises to do within the month. Still, Iyengar promises a reasonable price. “Our rates will be quite low, because our cost of Gigahash production will be far lower. We can afford to really have our products discounted,” he says.
This price point will be important, because Iyengar needs the money from pre-orders for preproduction. He has raised $1.5 million from angel investors based in Europe and the US, but he wants around $4 million to get his first chips off the production line.
He is developing units targeting a broad audience, and says that there will be a single-chip box with a 500 Gigahash/sec rate for lower-end customers. “There are some high-end miners that want to pack a lot of hash power into a single box,” he says, adding that he can provide 2 Terahashes/sec in a box.
“We want to keep a balance in the market. We don’t want to screw up the small miners, because they are a priority in our minds,” he says. That said, he will be selling chips and boards to OEMs, in the second half of Q4, as they become available.
Earlier this week, Cointerra announced a partnership with Open-Silicon, calling it an “ASIC design and development partner”. Open-Silicon won’t actually be doing any of the chip design, but it will be ‘taping out’ the final design for Cointerra before it goes to fabrication partner and AMD spinoff GlobalFoundries.
Taping out involves producing the final photomask that will be used during the production process. It’s the end of a massive design process, punctuated by an expensive production cycle, and it explains the high barrier to entry for ASIC design. Producing this stuff isn’t easy, which is no doubt partly why other companies have had their problems.
The question is whether Cointerra can pull off its claim of 500 Gigahashes per second on a single chip. It’s unprecedented, and far beyond the performance of anything else we’ve been promised. Stranger things have happened – but then, companies in the ASIC space have also made bold claims before, and been forced to backpeddle later.
While Cointerra does its work, the bitcoin network is doing its own – and more of it. In the five weeks since the end of June, the network hashrate has more than doubled. The difficulty is going to skyrocket as these new 28 nm ASICs come on stream – we’ll be watching when KnCMiner begins shipping, to see how much it spikes.
Right now, if you order a unit from KnCMiner, you’re getting it in October, the company says. You can’t even order from Cointerra yet, unless you are a volume customer.
Which leaves one important question: Is it better to be best, or to be first?