[Gllug] iPhone 4 or Android phone

Nix nix at esperi.org.uk
Sat Jul 17 16:32:22 UTC 2010


On 17 Jul 2010, Richard Jones said:

> On Fri, Jul 16, 2010 at 10:41:35PM +0100, James Courtier-Dutton wrote:
>> On 16 July 2010 22:05, Nix <nix at esperi.org.uk> wrote:
>> > I think that's spread-spectrum clocking, making the clock signal very
>> > slightly irregular (roughly). It is an RF noise reduction technique, but
>> > quite different from async clocking, in which there is no single clock
>> > signal per se at all.
>> >
>> > IIRC people aren't pouring money into async clocking because of radio
>> > noise, but because it would let CPUs run much closer to full utilization
>> > all the time. However it's bastard hard.
>> 
>> Some of their biggest customers are ones where the RF noise is a major concern.
>> For example, smart cards that have a crypto function. Lower RF noise
>> makes for harder side channel attacks.

More to the point, everyone sane uses spread-spectrum clocking in any
case. If modern CPUs (at least high-power / fast ones like
Intel-compatibles) didn't use it, you'd know!

>> The cost benefit trade offs for 'full utilization" or not, do not
>> favour async yet.

Not remotely. They're still in the lab.

> I'm quite convinced that "James Courtier-Dutton" is an Eliza-like AI
> experiment which scrapes content from random Google searches.  Needs a
> bit of tuning IMHO.

You win the Internet :)

> I have actually built both synchronous and asychronous hardware for a
> research laboratory (VMEbus was fun!).  As Nix quite rightly says,
> asynch circuits are not commonly used, not because of some airy-fairy
> "cost benefit trade offs", but because it is "bastard hard" to make
> them work at all.

Well, technically that is a cost/benefit tradeoff, with a rather large
cost. The cost is a huge decade-plus research program, the benefit is a
faster machine using less power and making less noise.
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug




More information about the GLLUG mailing list