VUPs to MIPS

From: Allison J Parent <allisonp_at_world.std.com>
Date: Thu Jan 8 07:57:01 1998

<With the current battle between RISC and CISC, MIPS now really is
<"Meaningless Information Provided by Salesman". Technically, by the
<current philosophy, an 8080 is a CISC processor. Go figure.

Actually it's a partial indicator of execution speed and only a partial
indicator of an systems capability.

8080 is CISC, instructions are complex(some are) and are expected to
take many clock cycles. RISC by definition is at most two clock cycles
to execute a given instruction. Generally RISC machines can perform
several arithmetic/logical operations in one instruction making them look
CISCier. They are also generally weak in addressing data as there are
few addresing modes and rely on lots of registers for pointers. The
problem with RISC is that compilers have to work hard to use the full
capability of the cpu. VAX is CISC to the max with an extremely rich
instruction set and addressing modes. FYI the VAX inherits most of the
base instuctions from the PDP-11 which is the also very CISC.

Now all this is nice when instructions are executed one at a time like
the 8080 or 6502 did. This became meaningless with later VAXen that
were more piplined so that current instruction execution overlapped the
next. The 8088 also does this albeit weakly. With the RISC machines also
doing super pipining the number of clock cycles became less meaningful
and the "MIPS" did as well. Adding things like caches complicates this
more as a cache flush or processor lock can really tie things up for long
periods of time affecting performance.

VAX VUPs are a more subjective standard as they are a measure of the
system rather than the bare cpu SPECint92 is similar in that respect as
doing real work generally makes the system a slave to outside influences
like the disk system data rates and memory system size. A good example
of this is that 6mhz Z80 can execute instructions at or about 1mips but
IO to the average floppy is slow enough to make that cpu wait about 1-2
seconds for 16k of data. The disk systems for machines like Vaxen are
designed to supply data at 20-1000 times that data rate. So if you
sorting data (say a data files that is 16mb in size) the disk systems are
also a big factor no matter what the system as the data and program will
not for in local memory. Even with huge memories in the 100s of megabytes
it's not always efficient(or possible with multitasking) to copy from disk
to ram and sort then copy it back.

Just a peak under the hood.

Allison
Received on Thu Jan 08 1998 - 07:57:01 GMT

This archive was generated by hypermail 2.3.0 : Fri Oct 10 2014 - 23:30:56 BST