Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83; site sdccs5.UUCP
Path: utzoo!watmath!clyde!floyd!harpo!decvax!ittvax!dcdwest!sdcsvax!sdccs5!ee161agt
From: ee161agt@sdccs5.UUCP
Newsgroups: net.micro
Subject: 4 -> 8 -> 8/16 -> 16 -> 16/32 -> 32  What next, 64 bit micros ?
Message-ID: <1161@sdccs5.UUCP>
Date: Mon, 12-Mar-84 21:02:27 EST
Article-I.D.: sdccs5.1161
Posted: Mon Mar 12 21:02:27 1984
Date-Received: Wed, 14-Mar-84 19:07:39 EST
Organization: U.C. San Diego, Computer Center
Lines: 45

[]
	I have a crystal-ball study on my desk chock-full of predictions for
the semiconductor industry in the next decade.  Some of it I agree with. 
( they predict 4M-bit Drams by 1992 ) Some of it I don't.  

In particular I wonder about this:  
	"64 bits will be the common word-size of mP's by 1992."

I'm not saying it's not technologically feasable to build a 64 bit 
Microprocessor; I'm wondering who would want or need one.  Perhaps I'm short-
sighted, but I can't see much performance or utility gain from a jump from 32
to 64 bits.
			     32
Data :		32 bits  =  2   = 4,294,967,296 
			     64
		64 bits  =  2   = 18,446,744,073,709,551,616

	How many times do you use a number > 4 billion ?  Is it reasonable to
	make data paths 8 bytes wide to allow for such rare cases ?  I think
	you'd be better off doing this kind of math in double-precision.

	Floating point ?  I don't do too much crunching, but I would expect
	64 bits of double precision on a 32 bit machine would give you plenty
	of precision.

	Bus bandwidth:  I'm all for decreasing the memory bottle-neck with a
	over-large memory bus, just don't increase the word size of the mP to
	do it.  64 data & 64 address pins ?  That's 128 pins to start with
	without any multiplexing.  I'd sure like to see those chips, they'd
	have pins sprouting from everywhere.

Address :	same as above.

	A 32 bit, 4 giga-byte virtual address space might get a bit small in
	the next decade, but is that any reason to go to 18,446,744 tera-bytes.
	( don't even know prefixes above tera- )  How about 40 bit words or
	48 bit words, anything but 64.

I expect systems-on-a-chip long before 64 bit mP's.  I'd prefer a cache, 
memory-management, I/O, even memory on the CPU chip before they start expanding 
the data & address paths.

Any other views on micros beyond 32 bits ?

Paul van de Graaf 	sdcsvax!sdccs5!ee161agt		U. C. San Diego