[Hardware] The market of ASICs (One GigaKey / Second?)

[Elektron]

On 12 Jul, 2004, at 12:58, Michael Meeuwisse wrote:

> The RC5 algorithm is quite easy to translate to a pipeline, which, as 
> every other pipeline on the planet, can push out a key check every 
> clock cycle. With today's technologies (i hope) i'm not 
> underestimating things when i say that one Gigakey every second 
> (that's not a typo, that's what... 1.5block every second? ) is 
> possible.
>
> But then there's something called math. (Technical stuff lower, skip 
> if not interested)
>
> The expanded key table is S[T-1] long, T depends on R (2*(R+1) to be 
> exact), R is usually 12. That makes S quite big. And because we wanted 
> this in a pipeline (my estimations: about 600 stages) it sums up at 
> S[25..0] + say... 4 extra vars (think L, think current key), times 32 
> bits, times 600 stages, divided by 8 for bytes, that makes 72KiB. Not 
> that much, right?
> Then there is bandwith to that small amount of memory. Every stage we 
> need to read and write a lot to this. I'm not going to lay it out to 
> you (and some people might want to argue this number: feel free) but 
> my calculations resulted in numbers around 240 Gigabyte every second. 
> And that's DDR'ed. And no, don't think that bursting numbers will get 
> it much lower. And don't even start about managing this amount of 
> bandwith required. Fact: it's impossible for external memory, internal 
> memory (in FPGA's, or other candy) isn't even an argument.
>
> Thus, that puts us in the position where FPGA's are not an option and 
> mainstream number crunchers don't compete. Is one gigakey / second an 
> illusion? No. It's possible, but at a price.
> ASIC's _have_ the possibility to actually do this kind of work. The 
> disadvantage? Custom made masks for the production of these ASICs cost 
> about a million a piece. If you're lucky, and if you're first design 
> works.
>
> Now for the economic side :)
> *Say* I have one million dollars. *Say* I think there's a market for 
> these little giants. And that I'm getting about 250 (is that 
> overestimated? don't have a clue actually) chips of one wafer, and I'm 
> selling them at 100 bucks a piece (you guys see how you interface it 
> to your pc for now). After 40 wafers, I'm break-even. After that, it's 
> all going to make me rich before I turn 20. :)
>
> 40 wafers. Say 50. 12.500 chips. At 100 bucks a piece. Is there a 
> market for this? Do people actually want to spend one hundred dollars 
> for a chip that's going to give them higher stats and a change for 'la 
> grande' price of 1000 bucks?
> Btw, I do find it sad to see that the projects 'against [something]', 
> which have a higher change of getting funded by people use obfuscated 
> algorithms (and floating point stuff iirc), making thinking of these 
> kind of implementations impossible. But that's offtopic. :)
>
> Would you spend 100 bucks on a chip that's only giving you higher 
> stats? Would you do it when you know you're helping developing 
> medicine for, say, cancer?
>
> Would love to see a reply to this,
>
> cheers, wacco
>
> Ps: for stats: hell no. Cancer: yup. I think i would.
>
> _________________________________________________________________
> MSN Zoeken, voor duidelijke zoekresultaten! http://search.msn.nl
>
> _______________________________________________
> Hardware mailing list
> Hardware at lists.distributed.net
> http://lists.distributed.net/mailman/listinfo/hardware
>
>
END