home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.arch
- Path: sparky!uunet!europa.asd.contel.com!darwin.sura.net!wupost!gumby!destroyer!sol.ctr.columbia.edu!news.columbia.edu!watsun.cc.columbia.edu!lasner
- From: lasner@watsun.cc.columbia.edu (Charles Lasner)
- Subject: Re: dinosaur horsepower
- Message-ID: <1992Jul29.003032.24500@news.columbia.edu>
- Sender: usenet@news.columbia.edu (The Network News)
- Nntp-Posting-Host: watsun.cc.columbia.edu
- Reply-To: lasner@watsun.cc.columbia.edu (Charles Lasner)
- Organization: Columbia University
- References: <Bs3oGt.6vp@helios.physics.utoronto.ca> <13307@ns-mx.uiowa.edu>
- Date: Wed, 29 Jul 1992 00:30:32 GMT
- Lines: 31
-
- In article <13307@ns-mx.uiowa.edu> jones@pyrite.cs.uiowa.edu (Douglas W. Jones,201H MLH,3193350740,3193382879) writes:
-
- >The DEC PDP-8, in 1965, had a 1.5us memory cycle time, so it took 3 us to
- >do a single 12 bit memory to accumulator add (1.5us for the instruction
- >fetch and 1.5us for the operand fetch and add). You could say that this
- >was a 1/3 MIP machine, but if you want any precision, you'll have to do an
- >extended precision add. A memory to memory 24 bit add on this machine
- >takes 7 instructions. A similar 36 bit add takes at least 11
- >instructions.
- >
- >The PDP-8 was a second generation machine (transistorized, discrete
- >components, core memory). The maximum memory capacity of the architecture
- >was 32K 12 bit words. Although it was a small inexpensive system,
- >by the standards of its day, the speed of the logic was very typical of
- >the times.
-
- As some people say, MIPS is Meaningless Instructions Per Second, so that's not
- a fair comparison. The PDP-8 can do 512 different operate instructions in only
- one cycle each. Most of them are meaningful, not just a NOP, such as:
- SMA SZA CLA which is skip the next instruction if the Accumulator is currently
- negative or zero and then clear the accumulator. All of this in 1 cycle of
- 1.5 microseconds, so therefore it's 2/3 MIPS not 1/3 MIPS. In terms of 1965
- pre-TTL logic, this was typical timing. Note that the I/O instruction was
- stretched to take 4.5 microseconds so that 1 microsecond spaced pulses
- could be generated on the bus, which could be 50 or more feet long. In later
- designs, the instructions optimized out unused pulse states so that most
- instructions could shave 3 microsecods off the time. Virtually none of them
- needed all three states; a few did need two. But the original PDP-8 and
- LINC-8 didn't optimize, so all I/O instructions did take 4.5.
-
- cjl
-