Little known fact: you can actually address 4MB of memory address space with a stock 8086/8088 as well (without using paging/windowing schemes like EMS). This is possible because during each memory access the CPU exposes the segment that is being accessed (CS, DS, ES or SS) on two of the pins. So if you devote a separate 1MB address space to each of these, that gives you 4MB total. Whether you are accessing CS via an explicit access or an execution prefetch is also exposed which could in theory give you 5MB, but then you wouldn't be able to write to the execution segment so it would no longer be a Von Neumann architecture.

I don't know anything about the KP1810BM86M but it's possible that it's not actually different from an 8086 and was just described as having 22-bit addresses because of this trick (and possibly as a marketing gimmick). In any case, a 22-bit variant of an 8086 system would not be able to run normal PC/XT software.

Hello reenigne, thank you very much for your reply!
And your explanation, of course. I did not know such tricks existed.
Seems like the 808x still holds a few surprises for us.. 😀

As for the KP1810BM86M, I do not really know anything about it, either (hence the thread).
Perhaps that soviet chip really was an oddball version, who knows.

The tiny bit of information I found was provided by Wikipedia, the typical source of a layman.
According to the last paragraph, for a longer time a 8086/86 clone was about the only x86 chip available.
Except for some rare 80286 clones (U80601, КР1847ВМ286).

So it would make sense that an 808x chip was somehow extended/improved to meet the requirements of certain fields of application.
That is just hypothetically speaking, of course. If there is any datasheet floating around, it hopefully is in English.

reenigne wrote:

In any case, a 22-bit variant of an 8086 system would not be able to run normal PC/XT software.

Well, this can be disproved by example, as 286 with its 24-bit address space is able to run most of "normal PC/XT software" 😀

Azarien wrote:

reenigne wrote:

In any case, a 22-bit variant of an 8086 system would not be able to run normal PC/XT software.

Well, this can be disproved by example, as 286 with its 24-bit address space is able to run most of "normal PC/XT software" 😀

Not in protected mode, which is necessary to access addresses higher than 0x10FFEF. When a 286 is running in real mode with the A20 gate turned off (as will be the case when it's running PC/XT software), its address space is effectively 20 bits.

^This discussion reminds me a bit of both Concurrent DOS 286, a protected-mode OS, and stories about the old CALL 5 interface. ^^

Yes, the segment + offset addressing scheme of the 8086 is physically limited to 1 MB.
The only backward compatible way around it is via virtualization, which requires a 386+ with v86 mode. You can then create separate 1 MB address ranges for each process on the machine.
The same trick is still being used for running 32-bit applications on 64-bit OSes today: each process can get a 4 GB address space.

1810VM86 is just clone of 8086. look:
http://i64.tinypic.com/nb53ph.png
here is NO pins for A20+ lines...

There is at least one NEC V-series chip that can address 16MB. One of them is the V33A. It has 24-address lines, and it's an extended 8086.

Anonymous Coward wrote:

There is at least one NEC V-series chip that can address 16MB. One of them is the V33A. It has 24-address lines, and it's an extended 8086.

Ah yes, here is its datasheet: http://datasheet.datasheetarchive.com/origina … ANS15-59629.pdf
Apparently they've created a new mode where you use pages of 16k instead of just 16 bytes.
They also implemented address translation, via a page table, so the segment registers are now selectors, similar to 386 protected mode.
So they also create a form of 'v86' mode, which can be transparent to 16-bit software, which still 'thinks' in 16-bit segments and offsets, and is unaware of the virtualization that goes on.