> Sony, IBM and Toshiba have released limited data about the so-called 
> Cell chip that will be able to carry out trillions of calculations per 
> second.<> The chip will be made of several different processing cores 
> that work on tasks together. The PlayStation 3 is expected in 2006 but 
> developers are expecting to get prototypes early next year to tune 
> games that will appear on it at launch.
>
> When put inside powerful computer servers, the Cell consortium expects 
> it to be capable of handling 16 trillion floating point operations, or 
> calculations, every second.

Let's put this into perspective.  This article sounds a whole lot like 
the marchitecture for the VXA-100, the chip at the centre of the Voodoo 
4 and 5 graphics cards.  "With the power to perform over 1 billion 
operations per second..."

Hashcash, and all other "proof of work" algorithms that I am personally 
aware of, is inherently integer-based, and cannot be evaluated (at 
least, not at all efficiently) on floating-point hardware.  Indications 
of floating-point performance of any given machine are completely 
useless for predicting integer performance.  Most likely, the integer 
engine in the PS3 could be fairly compared to a PowerPC G3, as that's 
all you need to run even the heaviest of current game/physics engines.  
Remember the SNES and Megadrive, which used an 8-bit Z80 to run the 
actual game, and a 32-bit 68000 to do the graphics.

Most of the floating-point hardware in the PS3 will almost certainly be 
geared towards graphics, not general computation.  As an example, the 
ATI R300 GPU (from the Radeon 9700) is capable of about 6 gigaflops 
(this is an extremely wild estimate, so revise up or down by about an 
order of magnitude as you see fit), noticeably more than the CPU in the 
same PC is likely to achieve, but you rarely see it being used for 
scientific computations because that hardware is very specialised 
towards graphics operations.

Finally, "when put inside powerful computer servers" means the Cell 
people are probably talking about more than one Cell chip at a time - 
probably *much* more - but the reporter has failed to recognise this 
fact and/or make it clear to the public.  It's not at all unlikely that 
they're talking about multiple thousands, in which case it would look 
rather silly if they *didn't* come up with performance in the 
teraflops.  The hardware to do teraflops, however, is rather unlikely 
to find it's way into a consumer games console any time soon.

--------------------------------------------------------------
from:     Jonathan "Chromatix" Morton
mail:     chromi <at> chromatix.demon.co.uk
website:  http://www.chromatix.uklinux.net/
tagline:  The key to knowledge is not to rely on people to teach you it.