Quote from: notbrain on March 23, 2013, 05:29:53 PM

The faster you do them serially, the more heat you produce. So a smaller manufacturing process would be necessary, at ridiculous clock rates. Liquid cooling from the get-go. You'd still need to do parallel; but there will always be room for faster and more efficient.

I've always wondered about this - why hasn't some small group within AMD or Intel or ARM gotten together and done something like this on the side? (If only I had gone into the EE track instead of web dev... Grin

) I think the process is just too complicated and expensive. Until an alternative use of SHA-2 256 hashes could be conjured up, it's a small market and the market cap of those companies is such that it really doesn't interest them. Maybe at BTC = $1000+?

The Avalon ASICs use a reference spec for their chips, BFLs are custom (correct me if I'm wrong), so there are a few ways of tackling the problem. Would be interested to know what your contacts at ARM say about it and what a 35nm process (or smaller?) would cost--still a "beyond seed funding" project. Maybe make a board/HDK that could be re-purposed for other things but also support a series of SHA-2 256 chips that could be put to mining use as well. FPGA-on-steroids. IANAEE

For some reason (and fortunately for FPGA and GPU miners) there seems to be gross mismanagement at BFL (and convicted felons under employ) and slow-but-steady progress at Avalon (graduate students in China doing the best they can without a solid biz plan).

ARM only takes $600m in revenue, and $150m in net, so if we hypothesise that Avalon are looking at close to $3m gross for ONE round then it would likely still interest them.

Because they've historically used the smallest processes possible and gained an advantage from it, while targeting lower power applications... I think if anyone could do it, it would be ARM.

Quote from: casascius on March 23, 2013, 05:39:33 PM

Welcome.

I don't know enough about manufacturing to be able to thoroughly understand how the process of going about any kind of large-scale ASIC production, but I wanted to throw out one thought that I had lingering that I feel would be very applicable to any ASIC endeavor undertaken by any large producer of silicon chips. I think that by all accounts I understand Bitcoin fairly well.

If I were hypothetically a producer of silicon chip products and in charge of my company's game plan, and also had the benefit of the Bitcoin knowledge that I currently possess, this is what I would be inclined to do.

First, I would make the ASIC bitcoin mining chips as a proof-of-concept, just as you have described, to get my foot in the waters of bitcoin mining.

Then I would evolve my IC design so that it became a one-wire design. By that, I mean that other than pins for power and ground, I would make it so that someone could derive all of the benefits from all the bitcoin mining core(s) on my chip through a single wire. This wire would implement some kind of command set and offer bidirectional communication.

Finally, I would take that design, and throw it in unused silicon space of other IC products I was producing, that happened to have at least one N/C pin available to steal for this purpose.

Essentially, I would be including FREE bitcoin miners in every single product I made, on the condition that there was room on the chip and one spare pin to expose the functionality, isolated of the remainder of the IC (other than power robbing). Whether any given buyer of that chip actually implemented the bitcoin miner would be completely up to them, and if the bitcoin mining pin were left unconnected, the mining function would just be dormant/disabled. Of course, the capabilities of the free bitcoin miner would be appropriately adapted for the environment the chip was going to be placed in... if it was likely to be in a low-power or heat-sensitive location, the free bitcoin miner would be trivially configurable through its command set to not cause the device to overheat or drain power inappropriately.

The expectation is, of course, that those who bought the chip and made them into marketable devices would have the option to make their device also be a free bitcoin miner if their business needs, resource constraints, and ideology all aligned to permit it.

Voila, I would be contributing to the hash rate of Bitcoin with a truly minimal resource cost, as well as shipping silicon products that had a niche competitive edge over others on the market, but meanwhile without burdening implementers who bought my chip and had no interest in bitcoin mining. I would feel like I was contributing to the stability of the world by making sure bitcoin mining chips covered the globe indiscriminately and weren't just being accumulated by the single largest bidder.

I like this, but its a hippy ideology in the sense that it makes the world a better place, but not the business's pockets.

Quote from: Gator-hex on March 23, 2013, 05:36:49 PM

Sadly ARM only design and license, no fabrication = no product.
Shame, because I'd have liked to see a British company compete.

I think its actually an advantage. Sure they don't have anything in house, but im sure the (every) fab that manufactures their products would be happy to help with a small project like this. After all, you'd play golf with your boss even though you hate him and gold right?