dbeeman@jarthur.Claremont.EDU (David Beeman) (09/28/89)
Lately, there have been a number of requests on the net for information
on fast 386 motherboards, so I thought that I'd pass on what I've learned
and at the same time ask if anyone has discovered any better prices for
33 Mhz boards than I've been able to find.
The May 30, 1989 PC Magazine reviewed a large number of 16, 20 and 25 Mhz
machines, but this review is already somewhat obsolete. Since then, prices
on the 25 Mhz motherboards have dropped to the point where they are not all
that much more expensive than the slower ones, and quite a few "no name"
generic Taiwanese boards have appeared which are build around standard chip
sets like the Chips and Technologies or Zymos/POACH sets. BIOS
compatibility doesn't seem to be much of a problem, as most of these use the
AMI or Phoenix BIOS. The back pages of PC Week and the ads in Microtimes or
Computer Currents will give you an idea of what you should expect to pay.
Computer Shopper is another good place to look, but you should phone for
current prices, as they will be out of date. Places which advertise
complete systems will often be willing to sell you a motherboard. Probably
the best way to find a good deal is to visit a "computer swap" where you can
talk to a number of vendors. Typically these are small businesses in local
industrial parks who put together clones for retail stores. They will
usually be able to give you a better price than you would get if you walked
into the store. Usually you can get them to agree to let you visit their
wharehouse the next week and buy components at the swap meet price.
Among the 25 Mhz machines reviewed by PC Mag, about one third had AMI
motherboards and one third had Micronics boards. The performance ratings
of those with AMI boards tended to be in the top half of the ratings, and
those with Micronics were in the lower half. However, if you look closely,
you will note that in the 25 Mhz category, the speed differences are small.
I ran several benchmarks on AMI and Micronics boards, as well as a few
Taiwanese boards and could see no significant speed differences between
the AMI and Micronics boards. The most significant factors affecting
performance are clock speed and whether or not a high speed memory cache
is used. It didn't seem to make much difference whether 32K or 64K of
cache was used, but boards with the Intel 82385 cache controller were
faster than those with the Austek controller.
The boards that I looked at tended to fall in four main categories.
(prices are typical "best deal" prices with CPU, but no memory)
I. Generic Taiwanese 25 Mhz boards with no cache, typical of those used in
the Gateway 386/25 ~ $700 - $800. Prices for a 20 Mhz board will be only
about $100 less. Be sure to ask for a true 25 Mhz CPU, rather than a 20
Mhz part "pushed" to 25 Mhz.
II. Generic 25 Mhz boards with the Austek controller ~ $900 - $1000
III."Name brands"
Micronics 2525 with 32K Intel 82385 cache $1300
AMI with AMI 64K cache (XT size board) $1350 (full size ~$1900)
IV. 33 Mhz boards with cache
Micronics 3333 (Price was lowered on Sept. 1) $1750
Hawk (made in Fresno) 64K Intel 82385 cache ~ $1700
Some things to ask: What is the bus speed? Is it selectable? Some video
cards won't run at 10 or 12 Mhz. What are the memory expansion options
and what kind of memory does it use? It is common to have room for 8 banks
of memory which must be filled 4 at a time with either 256K or 1MB memory.
This means that your choices are to put in 1MB of 256K ram (~$170) or 4MB
of 1 MB ram (~$410) as a minimum.
Below, I summarize the results of some tests with the PC Mag ver. 5
performance benchmarks ( from SIMTEL20 PD1:<MSDOS.SYSUTL>PCBENCH5.ARC ).
This is a newer version than used in the May review, so the times (in sec.)
are different.
386 Conv'l
Instr Prime Fl Pt Memory
Motherboard Mix Sieve Mix Access
~~~~~~~~~~~ ~~~~~~ ~~~~~ ~~~~~~ ~~~~~~
Generic 25 2.56 0.86 5.33 0.62
Generic 25 w/Austek 32K 2.39 0.57 4.39 0.39
Micronics 2525 2.20 0.52 4.01 0.36
AMI 25 Mhz 2.23 0.56 4.06 0.33
IBM 70/75 2.19 0.52 4.01 0.33
Hawk 33 Mhz. 1.75 0.40 3.07 0.36
In addition, I ran some timings on a "real" numerically intensive computation
(a neural network simulation using the back propagation algorithm). The
results were highly consistent with PC Mag's Floating Point Mix benchmark,
so it looks like this is a reasonable test for these kinds of applications.
I still haven't decided which board to buy, so if anyone has any additions
or corrections to this (or knows of a good deal on a 33 Mhz board!!!) I'd
like to hear from them.
|~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~|
| David E. Beeman, Jr. | dbeeman@jarthur.claremont.edu |
| Professor of Physics | or |
| Harvey Mudd College | dbeeman@hmcvax.bitnet |
| Claremont, CA. 91711 | or |
| uunet!jarthur!dbeeman |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~