64bit data 128Bit address...Re: Building a Z-80 (64bit!
Hans Franke wrote:
>>>> I therefore see address buses growing at 16 bits every 30 years.
That's
>>>> just over a bit every 2 years - slower than I expected but not much.
>>>> Someone (I forget who) said that memory chips double in capacity every
>>>> 18 months. This would give 16 bits in 24 years.
>
>>> Interesting szenario, especialy when connected to the Mores Law
>>> (didn't he tell this regarding integration ?).
>
>> Thanks. That's the one I'm thinking of - the amount of memory you get
on
>> the same area of silicon doubles every 18 months.
>
> Basicly I think its about integrations and transistor
> equivalents, but this is just linear to the size of memorys.
>
> In fact, to come back to the original question, Arfon
> just took doubling of address space and doubbling of
> data bus width for the same thing, but in fact they
> are two different functions - widening data bus is
> linear, while widening address bus is to the square
> (sorry, my mathematical english just stops here).
> This means doubble the data bus just doubbles the
> date transfer rate (the only thing the data bus is
> needed for) or w'=w+w. But doubbling the address bus
> is putting the address range (and thats what the
> address bus is for) to the square or r'=r*r.
>
> So, while a 256 or 512 bit data bus is usefull (and
> already in use in main frames - only called data path),
> even a 128 Bit address bus is just nonsense.
Hmm. "Square" isn't the best way of putting it. In English we generally
call that an "exponential" or "power" function: for an address bus of n
bits the size of the memory is 2^^n words (pronounced "two to the power of
n"). That's what I was getting at in my post.
I agree totally that data paths should be as wide as possible. It wasn't
just vaxen (Good post, Allison. A lot of very good points that I hadn't
thought of) and mainframes that did that - the PERQ was a 16 or 20 bit
machine but the memory data path was 64 bits wide (Help! Tony, you know
more about this!) and fetched 4 words at a time for the processor to chew
on, then 4 words for the video, 4 words for the I/O etc. Huge increase in
speed.
Address bus should be as wide as you think you might need plus a bit more.
As Allison said (again! Why oh why did I delete her post?) you want your
processor to be able to think about huge memories even if you can't build
them and have to swap to disk.
Philip.
Received on Wed Oct 28 1998 - 10:47:13 GMT
This archive was generated by hypermail 2.3.0
: Fri Oct 10 2014 - 23:31:30 BST