Ideas for PCAT

[Newsdee]

It would make DOS faster, but not better. I mean, you could do it, but it seems like you're just making a very fast 8086. It might be easier to redesign the existing core to clock faster instead.

Not sure that is enough; from benchmarks results it looks like a lower clocked 286 can beat a very fast 8086. I'd suspect it has to do with bigger width of the data bus?

[dmckean]

It would make DOS faster, but not better. I mean, you could do it, but it seems like you're just making a very fast 8086. It might be easier to redesign the existing core to clock faster instead.

Not sure that is enough; from benchmarks results it looks like a lower clocked 286 can beat a very fast 8086. I'd suspect it has to do with bigger width of the data bus?

It's not just bus width, there's two generations of speed increases between the 8086 and the 80286 and many instructions operate much faster at the same clock speed.

[Malor]

The 8086 has the same data bus width as the 286 -- 16 bits. The 8088 is the cut-down version, with an 8-bit bus.

What I'm talking about is a crazy fast 8086, which is more or less what an 80386 that had only real mode would mostly look like. I think the instructions would have at least the improved timings of the 80286 as well, so at least hypothetically, it should match or beat a 286 clock for clock, and then clock up to 40ish MHz if it was somewhat like a real 386. (386-33s were pretty common, 386-40s were rarer.)

So, speed would definitely be there, but probably no high memory area, and definitely no expanded or extended memory, and no 32-bit instructions. It wouldn't be useless, but it wouldn't functionally be much different than making the existing 8088 core more efficient and more than doubling the current max 14MHz clock. It would still more or less be an XT, just a super fast one.

I haven't the faintest idea which approach would be easiest. MicroCoreLabs would have to make that call. I will say that neither approach looks that great for payoff versus time invested, however, as just about all the OSes moved to using the extra facilities on the 386 very quickly. DOS programs mostly did so indirectly, via the additional available RAM and EMM386.

[MicroCoreLabs]

Well said... Implementing all of the protected-mode functionality is a heavy lift - but without all of this we are left with a real-mode only 386 which can only run DOS. Furthermore, I wonder if there are many games/applications which run in real mode which sense a 386 is present and then take advantage of the 32-bit instructions... If not, then the core would spend its life running 16-bit applications as a super-fast 8086 which would be easier to implement by just speeding up the MCL86 using a number of tricks...

Basically Im speculating that a 386 core without full support for protected mode would be somewhat useless.

[Newsdee]

If not, then the core would spend its life running 16-bit applications as a super-fast 8086 which would be easier to implement by just speeding up the MCL86 using a number of tricks...

Indeed, but right now PCXT is using an 8bit bus and the XT chipset.. so maybe first we need a core whith an AT bus implemented, with also the extra interrupts etc.

From there upgrading the 8086 to a later chip would be just a modular replacement (in theory).

[Malor]

I'm not sure how accurate PCXT is internally; it may not actually be implementing the ISA bus, it may just be wiring components directly together, maybe with a wait state or two.

But the PC bus changed a lot from the XT to the AT. It's just not the extra 8 bits, there's a whole second cascaded IRQ chip, and there's the A20 line for the weird rollover 286 bug that ended up creating the High Memory Area. And I think the UARTs were improved as well. And I think the bus runs faster. There are probably more changes that I'm not thinking of right now.

Spark2k06 has said very explicitly he doesn't want to touch any of that, so you'd have to find a different designer who was willing to fork the core and take it in a new direction.

edit: all that said, the basic AT chipset configuration persisted until the advent of PCI, so it would cover quite a few years, and probably all the x86 CPUs that would ever be likely to get an FPGA implementation. That part of it might not be wasted time, but it might be easier to strip the chipset code out of the AO486 core and fix it up, rather than doing it again from scratch.

[Newsdee]

Spark2k06 has said very explicitly he doesn't want to touch any of that, so you'd have to find a different designer who was willing to fork the core and take it in a new direction.

Yes, I know. It's quite a bit of work to see what is possible, but it can be interesting to try.
I'm sure it could compete (or even beat) a fast 286 for older DOS games in real mode.