1

People keep saying that proof of work functions like scrypt prevent ASICs from working because of large memory requirements.

What's stopping ASIC designers from simply building ones that have lots of memory?

1
  • 2
    ASICs for scrypt are already announced (I'm not sure about credibility of these companies, though). Also, see this question
    – aland
    Jan 11, 2014 at 19:19

2 Answers 2

2

Theoretically, you could produce an ASIC for every possible algorithm. It's done all the time by NSA, CIA and their kind for cryptographic purposes. Why has no company produced an ASIC for scrypt or to look for Cunningham chains? Return of investment!

Developing an ASIC for scrypt is harder than for SHA-256, this is by design. So designing the electronics is going to be costly, producing and testing the first test batches is going to be expensive as well.

For such a project to be done commercially, either: a) investors see a potentially big market of buyers for such an ASIC, or b) they sense a big investment opportunity in creating such machines for mining themselves a scrpyt-based coin.

For both situations to happen, scrypt-based crypto-coins need to become more popular/used, or their market capitalization increased substantially. Now that Litecoin has become much more high-profile (and its market capitalization a big chunk), and with Dogecoin, Coinye and so many other scrypt-based coins hitting the news, that there's a lot of interest in designing an ASIC for scrypt.

What surprises me is not seeing a team of teachers/students at any college in the world designing the electronics, getting a grant to create a few batches and finally either publicly releasing the gerbers or raising seed-capital to start a company based on the effort. I've seen a number of companies start up like this, have no idea why no one has tried it.

2

Production cost, an ASIC is a dedicated chip that does only 1 task and isn't programmable. Production costs for ASICS run into millions of dollars as it involves chip design, clean room manufacturing, and low yields (an industry average is that 50% of manufactured chips don't work and have to be trashed, can't be fixed). To create an ASIC for scrypt you have to have financial backing, that's why BFL and others did pre-orders. Scrypt is risky because ASICs for em would be the first and there's no guarantee people will buy your chips.

2
  • Incidentally, such high rates of defects in chips are what Intel tried to solve with their infamous 80486SX, which really was just the same chip as the 486DX except it had a defect in the integrated math co-procesor (what previously were the separate 287 and 387 chips). By rebranding and selling at a lower price, garbage was produced into a product for the low-tier market. Then if you bought a "487" it was really a full 486DX which took the place of the other one :-S
    – Joe Pineda
    Jan 12, 2014 at 0:16
  • 1
    See Nvidia's 1.7% semiaccurate.com/2009/09/15/nvidia-gt300-yeilds-under-2 and AMD's tri-core cpu which are really defective quad cores.
    – John T
    Jan 12, 2014 at 0:19

Not the answer you're looking for? Browse other questions tagged or ask your own question.