Thanks a lot for the information! 🙂👍

rmay635703 wrote on 2023-11-06, 01:07:

There was a Microsoft MACH10 8086-10MHZ Overdrive card for pcxt

Yes, I vaguely remember hearing about that card before.
Didn't know about it using an 8086.
I thought it was an 80186 or something.

rmay635703 wrote on 2023-11-06, 01:07:

The trouble with the 8086 is that it cost almost as much as a 286 and lacked extended memory while still using slow multiplexed address pins.
It also lacked IRQs/DMAs so you would only get 1 16bit XTISA slot, more and it could only be accessed with ports.

Or the other way round, the 80286 got as affordable as an 8086? 😉 Just kidding, I think I understand. 🙂

The lack of Extended Memory being unsexy makes sense.
On the other hand, many users of the time were cheap on RAM expansion.
Or so I learned in retrospect.

rmay635703 wrote on 2023-11-06, 01:07:

They WEREN'T rare, IBM, Compaq and generic beige box places used them, Tandy used them in the last XT 1000'S but they were always stuck in a wierd mid range place that was hard to market early on. (Some early pcxt software broke due to timing even on an 8086) Later the 8086 was a low end CPU that was slightly faster than an 8088 and again hard to market

That may be true, I believe you. It makes sense, too.
It's just.. x86 started out as a 16-Bit architecture, it uses 16-Bit ALU and 16-Bit instructions.
And the 8086 was the full, 16-Bit processor.
Using it was by principle the cleaner approach.

And the Olivetti M24 and the Amstrad PCs based on 8086 were performing near twice as fast as an 8088 PC at same clock rate.

I suppose, an 8086 running at 8 MHz would have been an improved experience to the user.

Especially if DRAM could be 16-Bit interleaved and/or didn't have to use any wait states.

That was a price problem, too, maybe.:
Fast DRAM was either expensive or not available.
So an 8-Bit connection wasn't possible at high speeds on an 8088.

By contrast, using slower (common) DRAM on an 8086 with a 16-Bit wide connection maybe was possible (interleaved), resulting in similar performance.

But even if interleaved RAM wasn't an option, interfacing the PC BIOS that way surely was.
Because, the BIOS routines are being called almost constantly.

Over a slow 8-Bit connection, it's not really ideal.

An AT compatible has Shadow Memory feature for compensating that bottleneck, often.

Using an 8086 and a 16-Bit EPROM connection (odd, even) would be an similar improvement, as well.

Anyway, I'm just thinking out loud here, of course. No offense.
Now that I think of it, maybe the true joy of the PC/XT platform is seeing things crawl, maybe.
So we can "see" the PC thinking (on oscilloscope, on debugger, on POST cards) and how it interacts with the environment.
I suppose that's part of the hobby, too. Things happen closer to our time frame, which makes things seem more real.

Edit: I had previously assumed that the 8086 was second-sourced by various manufacturers
and not much more expensive than the 8088, thus.
However, if there was little demand for the 8086 at the time, then they were produced in lower volume, maybe.
That could explain the price difference, maybe. Would make sense to me, at least.

Jo22 wrote on 2023-11-06, 20:51:

Fast DRAM was either expensive or not available. So an 8-Bit connection wasn't possible at high speeds on an 8088. […]
Show full quote

Fast DRAM was either expensive or not available.
So an 8-Bit connection wasn't possible at high speeds on an 8088.

By contrast, using slower (common) DRAM on an 8086 with a 16-Bit wide connection maybe was possible (interleaved), resulting in similar performance.

But even if interleaved RAM wasn't an option, interfacing the PC BIOS that way surely was.
Because, the BIOS routines are being called almost constantly.

Over a slow 8-Bit connection, it's not really ideal.

It seems you are unaware about the slow bus interface of the 8088 and 8086. Those processors take 4 processor clocks per bus cycle. So even at 8MHz, you just have to manage 2 megabuscycles per second, which is 500ns per cycle. This is plenty of time, even for DRAM with 200ns RAS access time (yeah, RAS cycle time is higher than RAS access time, but not that much). On the other hand, you are perfectly right that the 8-bit bus interface of the 8088 is a serious bottleneck ("what use is a processor that can execute "ADD BX,1234h" in 3 clocks (IIRC) if it requires 16 clocks to fetch that instruction?"), but "DRAM was to slow" is not the issue.

Jo22 wrote on 2023-11-06, 20:51:

An AT compatible has Shadow Memory feature for compensating that bottleneck, often.

In contrast to the 8086, which requires 4 clocks per bus access, the 80286 can get down to 2 clocks per bus access. The AT-like systems with 6, 8 or 10MHz did not require shadow RAM, the BIOS ROMs could be read in sufficient time that at most 1WS was required. Remember that the original AT (but not the XT/286) used 1WS even for the main memory. Shadow RAM got introduced with systems clocked at 12MHz or 16MHz.

Jo22 wrote on 2023-11-06, 20:51:

Using an 8086 and a 16-Bit EPROM connection (odd, even) would be an similar improvement, as well.

The 8086 still only has half the memory bandwidth of a 0WS 80286 system. Systems with an 8086 and 16 bit BIOS access do exist, e.g. the IBM PS/2 Model 25 and 30. Those systems used the 16 bit bus for onboard RAM and ROM, and narrowed everything else (including the MCGA graphics chip) to 8 bits.

Thank you very much for your explanation, mkarcher! It's very well done, I think. 😃
And you're right, I wasn't aware of these details. The 808x was a bit before my time, I have to admit. Thanks again.