Re: News on F21
On Tue, 11 Mar 1997, Arnim Littek wrote:
> On Mon, 10 Mar 1997 14:58:39 -0500, MISC@pisa.rockefeller.edu wrote:
> > If memory serves me right, F21 is 3/4 empty and is about 15K transistors.
> > that means that another 45K are potentially sitting there unused. How many
> > Ks of SRAM does that make? If Chuck's OKAD is about 8K, I believe that
> > almost anything useful can be that big and fit on chip.
>
> RAM is expensive, make no mistake. SRAM is 4 or 6 transistors per
> cell. An 8Kx8 SRAM is therefor 64K x 4/6 transistors, ie. MUCH more
> than the 45K leftover, as per your info.
Well, at 4 transistors/cell 45K of them makes 11K cells. 20 bits each,
this makes 550 words. Not much, but a small compute kernel can fit in
there to help with intensive computations.
At 500 mips internally, one can do 8x8 multiply/adds at 60M/sec. Not bad,
if you ask me.
Or one can delegate some blazingly fast algorithms transparently to FORTH
words. Given that the data access can then be on-page only, it is worth
even to make a fast memory move alone.
In that regard, I think that even as little as 50 cells will go a long way
towards true 500 MIPS computations.
And even 10 of them will be useful in certain circumstances, ant not very
difficult to implement (I hope). Mind it, F21 has 32 21-bit registers
right now.
--
Penio Penev <Penev@pisa.Rockefeller.edu> 1-212-327-7423
