Path: utzoo!attcan!uunet!steinmetz!disney!oconnor
From: oconnor@disney.steinmetz (Dennis M. O'Connor)
Newsgroups: comp.arch
Subject: Re: 60 Megahertz Microprocessor
Keywords: GaAs;  "Worlds fastest" is tough to define
Message-ID: <12596@steinmetz.ge.com>
Date: 14 Nov 88 18:42:13 GMT
Sender: news@steinmetz.ge.com
Reply-To: oconnor%sungod@steinmetz.UUCP
Organization: GE Corporate R&D Center
Lines: 41

An article by mark@mips.COM (Mark G. Johnson) says:
] In article <19070@uflorida.cis.ufl.EDU>,  cl0@beach.cis.ufl.edu writes
]   >According to the recent Communication ACM (Vol. 31, No. 10, Oct. 1988,
]   >p.1173), the world fastest microprocessor is now in existence and clocks
]   >in at over 60 Megahertz. Any one have more information of it? Thanks.
] 
] [...] the title [...] November 1988 --  belongs to Rockwell
] International [...] Galliun Arsenide MESFET implementation [...]
] at 150 million native instructions per second. [...]
] 
] Texas Instruments [...] Gallium Arsenide *bipolar* technology.
] [...] 200 MHz and 200 million native [IPS]
] when their die shrink pops out of fab ... in February [...]
] In February the non-shrunk die was running at 120 MHz.
] [...]
] Prediction: Whatever uP you choose to brand "worlds fastest", there will
] be lots of disagreement.  LOTS of disagreement.

This is very true. For instance, the the IEEE FP uniots for the
GaAs chips mentioned use bit-serial techniques to do multiplies.
For the TI chip, for example :

	Integer multiply : 39 cycles, 195ns
	Integer divide	 : 47 cycles, 235ns
	FADD single      :  6 cycles,  30ns
	FMUL single	 : 19 cycles,  95ns
	FDIV single	 : 38 cycles, 190ns

    Double precision floating point is even slower.

This isn't slow, but on-the-market CMOS chips can and have done better,
simply by throwing more transistors at the problem. Although I haven't
seen any CMOS 30ns IEEE adders, I've seen much faster CMOS multipliers.
So what's your measure of speed : Integer adds or floating multiplies ?

Remember that speed isn't GaAs's only interesting property to the
military. They are VERY Rad-hard, which is sometimes more important.
The best bulk CMOS process I've seen is only good to a megarad.
--
 Dennis O'Connor   oconnor%sungod@steinmetz.UUCP  ARPA: OCONNORDM@ge-crd.arpa
   "Never confuse the USENET with something that matters, like CAFFEINE."