Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!seismo!rutgers!mcnc!ece-csc!ncrcae!ncr-sd!hp-sdd!hplabs!cae780!tektronix!tekcrl!tekgvs!toma
From: toma@tekgvs.TEK.COM (Tom Almy)
Newsgroups: comp.sys.ibm.pc
Subject: Re: brain-dead 386?
Message-ID: <2604@tekgvs.TEK.COM>
Date: Thu, 27-Aug-87 10:03:54 EDT
Article-I.D.: tekgvs.2604
Posted: Thu Aug 27 10:03:54 1987
Date-Received: Sat, 29-Aug-87 12:08:27 EDT
References: <4149@spool.wisc.edu.WISC.EDU>
Reply-To: toma@tekgvs.UUCP (Tom Almy)
Distribution: comp
Organization: Tektronix, Inc., Beaverton, OR.
Lines: 32
Keywords: 386, 16 Mhz 286

In article <4149@spool.wisc.edu.WISC.EDU> jojo@speedy.WISC.EDU (Jon Wesener) writes:
>
>I have recently been hearing how AMD 16Mhz 80286-16 with CHIPS chips
>set is faster than the 80386 also running at 16Mhz.  [...]
>It would seem the 286 would still have 1/2 the bandwidth 

Well it depends on the benchmark program.  Most instructions will execute in
the same number of clocks on both processors.  If the AMD system has 0 wait
states it could easily run many benchmark programs faster than an 80386
system with one wait state.  To tip the scale in favor of the 80386 you can
execute benchmarks with lots of fast executing but multi-byte long instructions
which would show the advantage of the 80386 instruction queue.  Benchmark
the performance of VDISK -- the extended (or is it expanded?) memory fetches
are done by moving 32 bits at a time instead of 16, and the return to real
mode is done without reseting the processor -- VDISK is much faster.

To really see the speed advantage you will need to run benchmark programs
with different machine code for the 80386.  For instance, the famous
8086 killer C benchmark "long i; for (i=0; i<1000000; i++);" will be much
faster on the 80386 because of the 32 bit data registers and instructions.
"Large Model" programs can be recompiled small model on the 80386, saving
lots of execution time, because of the 4 gigabyte segment size.

Now the rest of the system (disks, display, in fact anything plugged into
the PC bus) will not be any faster with an 80386 because the data path is
a mere 16 bits wide.  I have found that EGA intensive programs do not run
noticibly faster on the 80386 than they did on my older 8Mhz PC/AT.  An
80386 system could be real nice if it didn't have to be PC/AT compatible.

Tom Almy
toma@tekgvs.TEK.COM
{many hubs}!tektronix!tekgvs!toma