20 Nov
2007
20 Nov
'07
4:03 a.m.
Subject: Re: *updating* 8088's
From: woodelf <bfranchuk at jetnet.ab.ca>
Date: Mon, 19 Nov 2007 20:39:49 -0700
To: General Discussion: On-Topic and Off-Topic Posts <cctalk at
classiccmp.org>
Chuck Guzis wrote:
> Well, there's clock speed and then there's "apparent clock speed".
> Let's see, a 16MHz 80386sx is equivalent to what, a 10MHz 80286?
Actually it's teh other way around the 386 was more efficient than
286 for the same clock speed.
So what speed does the memory run at is my question and
how wide?
The SX narrows teh bus bandwidth so the memory is cycled either
faster to make up for it or has fewer idle states (bus availability).
that meant using 60ns 30pin simms.
Playing around with homebrew micro design I am working
on, a 500 ns
memory cycle ( 2Mhz ) is about as fast as you can go with standard
parts about 150 ns access time. My limiting factor is not memory
speed but EEPROM and I/O chips dead slow speeds.
Those speed limits are consistant with pre 1978 parts for ram
and even in 1979 I had some semistatic rams that were 200ns.
Eproms were always slow and didn't break the 250ns barrier until
around 82 but the mask roms were quite a bit faster.
I build with 6 and 10mhz z80s and 12mhz 8085s I have and theres
little problem with finding static rams and Eprom (and EEprom)
that can keep up. I say little problem as I can find plenty of
parts that way too fast in non-DIP formats. If I need faster
I can easily find CMOS static rams in the under 25ns range
(486dx used 32kx8 and even 64kx8 15ns parts for cache) and
larger EE/Flash/Eproms in the sub100ns range.
Even back in 1982 I could get power hungry 2147(4Kx1) and 2167(16kx1)
parts in the 45ns range. Drams even first out 4164s were under Tcy
of 300ns. Old Eproms in the pre8K sizes (2716, 2732) were never fast
but 27C256s that do 150ns are really old parts and 27C010s I have
are 150ns for the slow parts.
Zilog peripherals can be found still at 4mhz and the 8085/8088
(82xx) parts were good to 5mhz with later ones (82Cxx) good to
125ns.
Allison
Cheers,
Chuck
20 Nov
20 Nov
8:47 a.m.
On 19 Nov 2007 at 23:03, Allison wrote:
Actually it's teh other way around the 386 was
more efficient than
286 for the same clock speed.
In 16-bit mode, I seem to recall that the 386SX was a travesty of a
CPU; a 16 MHz SX ran nowhere near as fast as a comparably-clocked
286. Early 386 boxes were nothing to crow about in 16-bit mode--and
a 32-bit software base pretty much didn't exist early on--and the
386SX was limited to 16MB of external memory, just like the 286.
Doubtless the 386SX board designs were low-budged also, which
probably figured into things. Some 286 vendors made a big thing of
the fact that a 286 could execute 16-bit real mode code substantially
faster than the 386SX. For example:
http://www.intersil.com/data/an/an121.pdf
On the other hand, the 386SX could execute 32-bit code. That is, if
you had any to run in 1989.
Cheers,
Chuck
10:03 a.m.
On Tue, Nov 20, 2007 at 12:47:51AM -0800, Chuck Guzis wrote:
On the other hand, the 386SX could execute 32-bit
code. That is, if
you had any to run in 1989.
There was plenty of 32-bit code being written in 1989... just not for MS-DOS.
-ethan
--
Ethan Dicks, A-333-S Current South Pole Weather at 20-Nov-2007 at 09:50 Z
South Pole Station
PSC 468 Box 400 Temp -40.7 F (-40.4 C) Windchill -63.5 F (-53.0 C)
APO AP 96598 Wind 6.8 kts Grid 73 Barometer 681.9 mb (10557 ft)
Ethan.Dicks at usap.gov http://penguincentral.com/penguincentral.html
1:34 p.m.
On the other
hand, the 386SX could execute 32-bit code. That is, if
you had any to run in 1989.
There was plenty of 32-bit code being written in 1989... just not for MS-DOS. 
21 Nov
21 Nov
1:05 a.m.
On Nov 20, 2007 5:34 AM, Cameron Kaiser <spectre at floodgap.com> wrote:
What about games, DOS/4GW, ... ?
I got my 386DX in 89. Very few, if any, MS-DOS games relied on 32bit
extenders at that point. It cut out too much of the market that still
had 16bit processors. It would have been nice not to push so hard to
have 635k low memory free in order to run the latest game.
The most used feature of the 386 was V86 mode, and it was mostly used
to free up memory for 16bit MS-DOS programs.
Eric
On the other hand, the 386SX could execute 32-bit
code. That is, if
you had any to run in 1989.
There was plenty of 32-bit code being written in 1989... just not for MS-DOS. 
20 Nov
20 Nov
2:51 p.m.
On Nov 20, 2007, at 5:03 AM, Ethan Dicks wrote:
Thank you! You saved me the trouble of saying exactly the same
thing. :)
-Dave
--
Dave McGuire
Port Charlotte, FL
Farewell Ophelia, 9/22/1991 - 7/25/2007
On the other
hand, the 386SX could execute 32-bit code. That is, if
you had any to run in 1989.
There was plenty of 32-bit code being written in 1989... just not
for MS-DOS.
4:03 p.m.
On 20 Nov 2007 at 10:03, Ethan Dicks wrote:
There was plenty of 32-bit code being written in
1989... just not for MS-DOS.
No doubt (I probably used a lot of it). But how much was available
at the debut of the 386sx? The PC world took an extraordinarily long
time to fully embrace 32-bit code. In 1989, there was Windows/386,
but that was about it from Microsoft--and it wasn't wildly popular.
I remember packages such as 386max, but those weren't applications
per se, just operating enhancements.
Even Windows 3.0 was a full year away in 1989. OS/2 2.0 was 3 years
off; Windows NT was 4 years away (and I wouldn't call 3.0 a 32-bit
bit OS because it could load 32-bit VxDs). That was pretty lame for
a CPU feature that had been available to the public since 1986--and
known, more or less accurately, by vendors for at least a year prior
to that.
Maybe the "16 BIT S/W ONLY" chips didn't help either.
In 1989 most protected-mode code was 16-bit. I recall an Intel
application engineer moaning about this at the time. I suspect a
fair number of people grabbed a 386sx because they thought it would
make things run faster--or for the hype.
Cheers,
Chuck
21 Nov
21 Nov
12:57 a.m.
Doubtless the 386SX board designs were low-budged
also, which
probably figured into things. Some 286 vendors made a big thing of
the fact that a 286 could execute 16-bit real mode code substantially
faster than the 386SX. For example:
http://www.intersil.com/data/an/an121.pdf
On the other hand, the 386SX could execute 32-bit code. That is, if
you had any to run in 1989.
You do need to consider that that "application note" was a marketing
tool and the subroutines they designed for the application note of
course will make heavy use of instructions that are faster on a 286
than on a 386SX. On typical 16 bit code I didn't see a significant
difference on a cacheless 16MHz 286 and a cacheless 16MHz 386SX. I
don't currently own a 386SX machine, so I can't give current
benchmarks. ISTR that you could get a 386SX with an SRAM cache on the
motherboard. I don't recall that on any 286 boards, but I'm sure
someone else will point one out...
Eric
2 a.m.
Chuck Guzis wrote:
In 16-bit mode, I seem to recall that the 386SX was a
travesty of a
CPU; a 16 MHz SX ran nowhere near as fast as a comparably-clocked
286.
I find that incredibly hard to believe, looking at the instruction
timings for both processors. What are you basing this on? Memory timings?
The only thing I can think of that might substantiate this claim is when
you were working with 32-bit registers in 16-bit DOS. Each instruction
needed a 66h or 67h prefix opcode to indicate 32-bit. This was wasteful
on a 386sx because the 16-bit memory interface ate more time fetching
the additional opcodes than you were saving by using 32-bit registers
(!). A friend of mine wrote a 386 extender ("pmode", circa 1992)
specifically to get the full use of his 386sx because he was royally
cheezed when he learned of this.
Some 286 vendors made a big thing of
the fact that a 286 could execute 16-bit real mode code substantially
faster than the 386SX. For example:
http://www.intersil.com/data/an/an121.pdf
I don't think that's a fair comparison because that's not a system,
that's a CPU variant (80C286).
--
Jim Leonard (trixter at oldskool.org) http://www.oldskool.org/
Help our electronic games project: http://www.mobygames.com/
Or check out some trippy MindCandy at http://www.mindcandydvd.com/
A child borne of the home computer wars: http://trixter.wordpress.com/
5:20 a.m.
On 20 Nov 2007 at 20:00, Jim Leonard wrote:
I find that incredibly hard to believe, looking at the
instruction
timings for both processors. What are you basing this on? Memory timings?
Nope, the early 386 machines were pretty lame. Anyone else have the
same experience with the SX vis-a-vis a 286?
Cheers,
Chuck
9:39 p.m.
Nope, the early 386 machines were pretty lame. Anyone
else have the
same experience with the SX vis-a-vis a 286?
That's an artifact of the design method for early cheap 386
motherboards, which consisted of a single step. Take a 286
motherboard and double the data bus width. It resulted in a 386 that
spent all of its time waiting for the next memory access.
Decent (not cheap) motherboards with a significant amount of cache and
interleaved memory were much better.
22 Nov
22 Nov
12:11 a.m.
Chuck Guzis wrote:
Nope, the early 386 machines were pretty lame. Anyone
else have the
same experience with the SX vis-a-vis a 286?
Could you quantify "lame"? I used a 386sx-16 for years and it was
always faster (visibly, not just via benchmarks) than the 12MHz 286 I
had access to. If you're going to claim a 12Mhz 286 was faster than a
16MHz 386sx you'll have to provide some evidence...
--
Jim Leonard (trixter at oldskool.org) http://www.oldskool.org/
Help our electronic games project: http://www.mobygames.com/
Or check out some trippy MindCandy at http://www.mindcandydvd.com/
A child borne of the home computer wars: http://trixter.wordpress.com/
12:27 a.m.
On 21 Nov 2007 at 18:11, Jim Leonard wrote:
Could you quantify "lame"? I used a
386sx-16 for years and it was
always faster (visibly, not just via benchmarks) than the 12MHz 286 I
had access to.
My own experience was the opposite. I had a 16 MHz 386sx that was
visibly *slower* than a 12MHz 286 that I also had. It was very
cheap and eventually just quit working reliably all on its own.
Perhaps that was an artifact of early SMT. OTOH, I still have a
couple of 386DX systems that have worked flawlessly since the day I
bought them. One uses DIP for memory; the other uses SIPPs.
So, perhaps my own was a victim of bad design. I never was even
tempted to purchase another, so bad was my experience.
Cheers,
Chuck
3:30 a.m.
Chuck Guzis wrote:
When I have my 12MHz 286 and my 16MHz 386sx on the same table, I will
run benchmarks just to make sure the both of us aren't going crazy :-)
--
Jim Leonard (trixter at oldskool.org) http://www.oldskool.org/
Help our electronic games project: http://www.mobygames.com/
Or check out some trippy MindCandy at http://www.mindcandydvd.com/
A child borne of the home computer wars: http://trixter.wordpress.com/
Could you
quantify "lame"? I used a 386sx-16 for years and it was
always faster (visibly, not just via benchmarks) than the 12MHz 286 I
had access to.
My own experience was the opposite. I had a 16 MHz 386sx that was
visibly *slower* than a 12MHz 286 that I also had. It was very
cheap and eventually just quit working reliably all on its own.
Perhaps that was an artifact of early SMT. OTOH, I still have a
couple of 386DX systems that have worked flawlessly since the day I
bought them. One uses DIP for memory; the other uses SIPPs.
So, perhaps my own was a victim of bad design. I never was even
tempted to purchase another, so bad was my experience. 
21 Nov
21 Nov
6:07 p.m.
On Tuesday 20 November 2007 21:00, Jim Leonard wrote:
The only thing I can think of that might substantiate
this claim is when
you were working with 32-bit registers in 16-bit DOS. Each instruction
needed a 66h or 67h prefix opcode to indicate 32-bit. This was wasteful
on a 386sx because the 16-bit memory interface ate more time fetching
the additional opcodes than you were saving by using 32-bit registers
(!).
This is one of the things that's always bugged me about those "extra" z80
instructions...
--
Member of the toughest, meanest, deadliest, most unrelenting -- and
ablest -- form of life in this section of space, ?a critter that can
be killed but can't be tamed. ?--Robert A. Heinlein, "The Puppet Masters"
-
Information is more dangerous than cannon to a society ruled by lies. --James
M Dakin
20 Nov
20 Nov
4:43 p.m.
On Nov 19, 2007, at 11:03 PM, Allison wrote:
I build with 6 and 10mhz z80s and 12mhz 8085s I have
and theres
little problem with finding static rams and Eprom (and EEprom)
that can keep up.
Did you say 12MHz 8085? Holy cow! Tell me more?
-Dave
--
Dave McGuire
Port Charlotte, FL
Farewell Ophelia, 9/22/1991 - 7/25/2007
6255
days inactive
6257
days old
15 comments
8 participants
participants (8)
-
ajp166＠bellatlantic.net -
cclist＠sydex.com -
ethan.dicks＠usap.gov -
korpela＠ssl.berkeley.edu -
mcguire＠neurotica.com -
rtellason＠verizon.net -
spectre＠floodgap.com -
trixter＠oldskool.org