Hi Aerobatic, you are of course quite correct in that voltage affects the power dissipation, but in the case of SHA256 it would be very unusual to increase the supply voltage - as the games do to get faster running cpu's - because the pipelines can generally switch much faster than they are clocked at. In SHA256 the problem is that almost every transistor in the pipelines switches every clock cycle - the real problem is how to get rid of the heat that's generated without having to increase the die size (and so wasting silicon area), I'm sure I remember the guy from Bitfury posting stuff some time ago about lowering the supply voltage on his asic to run cooler, which on the face of it sounds sensible but then that has a knock on effect on threshold voltages and that in turn can result in unreliable operation due to increased susceptibility to noise.
For my part, I'd much rather get a device that operates reliably within it's rated parameters that has been designed from the start to be efficient rather than buggering around to try to squeeze out more performance. Intel don't design 'to the limit', that's why their chips can be overclocked so readily, the engineers have built in a margin for error. The Bitcoin asic designers don't seem to grasp this point which makes me high suspicious of some of the 'credentials' trumpeted on their websites.
I chose my username because my colleagues do view me as a bit of a dinosaur, but that's because I started out designing things when you only had a few tens of thousands of transistors to design with, not hundreds of millions and so you had to be very creative with your resources. I like doing things right that work the way they were specified first time around. A lot of the technospeak from the rig vendors make me shudder and I wish to goodness someone would come along with a mining product minus the bullshit, fantasy project timescales but with some solid engineering behind it.
As an example of what bugs me, I had a look recently at Cointerra's Team, and to my astonishment found they have a 'Chief Crytographic Advisor'. Why would they employ such a guy? Don't the engineers understand SAH256? Nothing against Dr.Hanke, I'm sure he's very good at what he does but if I was a prospective rig customer I'd much rather see a VP in their team that actually had some experience of volume product manufacturing because the chip design for SHA256, whether anyone wants to admit it or not is fixed and well within the capabilities of most EE graduates. So the chip design is a given. What's proven not so easy is the actual system integration and product engineering part.
Nothing against Cointerra by the way, they seem less suspect than most of the other suppliers but they're still way too expensive for what's on offer and are excluding a huge number of people who would like to participate in mining.
Sorry for the rant. It's old age that causes it!
hi Bronto... have to disagree with you there. There's plenty of 'innovation' that goes into designing these high end bitcoin mining asics. Those that are just doing a textbook SHA256 design or are licensing an sha256 block and cookie cutting their chip design are now suffering from extremely poor performing asics that consume too much power per GH and dont fit many hashing units on their die. They have no way to differentiate their products. whereas the guys who are doing it right, are getting huge improvements in power consumption, performance, and size efficiencies.
i think both bitfury, knc, hashfast and cointerra.. would say that they have applied significant r&d and innovation to their respective designs. They're not just dialling in how many hashing units are in the die.. nor whether to use a single cycle pipelined hash or an iterative one... though there are designs that use either style... instead, theyre making fundamental power, die size and performance optimisations to the architecture and implementation of the hash itself.. and the ways to keep the pipelines full. Bear in mind that any improvements in the power consumption mean you can hash faster (no pun intended) since youve got more headroom before hitting the thermal limits of the package... and any improvements to the die size mean you can fit more hashing units in the same space, which has obvious advantages.
some of the special sauce is in architecture and rtl design.. and some of it is in back end layout... optimising the data paths and making the most of the process available.
as for why companies might employ a crypto consultant like Dr Timo Hanke.. well, i dont think theyd necessarily be using him to invent new crypto algos.. but instead someone of his background and skillset can add value in designing the architecture and software to make the best chip. In his case, he's also a very strong software engineer and is consulting (he's based in germany) on their software and architecture effort. Im sure all other bitcoin mining asic companies employ or consult with someone similar, though not necessarily from a crypto background.
If you really think there's no engineering innovation possible, then there's no point designing a new bitcoin asic... and you might as well just license the ip and subcontract the whole lot. Some companies did that. look where they are right now (VMC/AMC... im thinking of you)
i actually really like analyzing what happens in this bitcoin mining asic field... as its the perfect HPC arena. Its a microcosm of new technologies designed in record time. It really focusses many aspects of engineering from asic to system design.. with significant attention to power consumption and cooling technologies. these guys are doing state of the art stuff, that may perhaps be teaching lessons to the HPC and Intels of the world. Certainly most bitcoin mining asics have been designed and fabbed in record time and many of these companies have broken world records for how to go from concept to silicon in literally single digit numbers of weeks... if only the rest of the silicon world worked on those timescales.
i should also re-iterate what someone else said... that in bitcoin mining, since performance is the ultimate goal, and since the life of the hardware is very limited (months, not years)... the hardware designs are optimised to go for extreme overclocking and maximum performance.. at the bleeding edge. if they didnt, they wouldnt have competitive performance. thats the nature of the beast. if you want it to run cool for hours, im sure you can do that. but it wont generate that many coins!
-- Jez
