What if that weird memory size "issue" was just the way the BIOS counts memory ? My 486 counts up to 7808KB of ram when there's really 8192 on board. Same with my 386, and this happens with other memory sizes too and probably other computers, but I couldn't tell what size precisely is 24MB in kB for example so I can't say.

But all these machines have something in common : they use an AMI bios. I'm not sure why the bios does that, but it's not an issue.

I recall there's also a memory hole between 15MB-16MB (some BIOSes have an option for that).
And there's shadow memory, of course. Incl. BIOS shadowing and VGA BIOS shadowing.

Scali wrote:

True, the SX core is the same as the DX core [..]

I think the same. I could be wrong, but I remember to have read that the 386DX has got all the features of the 386SX.
And that the original i386 incorporated both 32/32Bit and 16/24Bit bus adressing schemes from the very beginning.
So in theory, the DX could also run in "386SX mode" on a 286 or 386SX chipset, couldn't it ? 😀
If so, this would explain why the press in old PC magazines (~mid-late 80s) described the 386SX as "lame":
At least the original, non-enhanced SX (386SX-16) is/was just a pin-reduced version of the normal 386.
(Anyway, I'm speaking under correction here.)

Deksor wrote:

What if that weird memory size "issue" was just the way the BIOS counts memory ? My 486 counts up to 7808KB of ram when there's really 8192 on board. Same with my 386, and this happens with other memory sizes too and probably other computers, but I couldn't tell what size precisely is 24MB in kB for example so I can't say.

But all these machines have something in common : they use an AMI bios. I'm not sure why the bios does that, but it's not an issue.

I just looked 2 of my 486 pcs. One has Mylex eisa bios and 4Mb ram and it counts to 3968Kb and the other one has Amibios and it counts to 8064Kb with 8Mb ram which means there is 128Kb unaccounted for with both computers. I assume it is some kind of memory hole, but does anyone have any more information about it?

I assume it is some kind of memory hole, but does anyone have any more information about it?

Video BIOS caching.

The Serpent Rider wrote:

I assume it is some kind of memory hole, but does anyone have any more information about it?

Video BIOS caching.

It still stays the same even if you disable video bios caching in bios.

Some motherboards have a bios option to remap memory, otherwise the memory that sits between 640kb and 1mbyte is not accessible (except for shadow memory caching) and often not counted by the BIOS.

If you are seeing 128KB rather than 384kb missing, then the chipset is remapping some of the memory, but likely doing RAM shadowing of both the standard BIOS ROM and also the VGA BIOS ROM. RAM shadowing means that the chipset copies the contents of the BIOS ROM into standard memory and then puts a write lock on that memory. They do this because system memory is normally a LOT faster than ROM

If you are using DOS/Win3x/Win9x, you should leave the BIOS shadowing enabled.

Dos "mem" command shows all 8192Kb but it shows 133Kb out of those 8192Kb as "reserved", so for some reason it adds 5Kb to those 128Kb that were missing from the post memory counter.

Jo22 wrote:

So in theory, the DX could also run in "386SX mode" on a 286 or 386SX chipset, couldn't it ? 😀

In theory yes, because that's exactly what a 386SX is (just like an 8088 is an 8086 on an 8-bit bus).

Jo22 wrote:

If so, this would explain why the press in old PC magazines (~mid-late 80s) described the 386SX as "lame":
At least the original, non-enhanced SX (386SX-16) is/was just a pin-reduced version of the normal 386.

Yes... Technically the real 386DX-16 was pretty 'lame' already. The early 386 systems weren't all that much faster than equally clocked 286s.
Check this out for example: https://books.google.nl/books?id=nuXmVNll5JEC … c%20mag&f=false

The true power of the 386DX only came later, firstly, because the clockspeeds were ramped up to 33 MHz (I suppose the 386DX-33 is the 'archetypal' 386DX, much like the 486DX2-66 is for the 486). Secondly, because 32-bit software arrived, taking advantage of the new features of the CPU. And thirdly, because motherboards and chipsets improved, and caches were added to 386 machines.
And 386SX machines were actually slightly slower than equally clocked 286 machines at the same clockspeed.

The 386SX however, remained a budget machine. It basically ran on modified 286 motherboards, was always limited to a 16-bit bus, so 32-bit software didn't gain all that much. And because it was a budget solution, the motherboards generally didn't have cache either, and the clockspeeds were low (The 386SX-16 is probably the 'archetypal' 386SX, even though SX models up to 33 and 40 MHz do exist).

So over time, the gap between the SX and the DX grew bigger. Both in actual performance at the same clockspeed, and also in perceived status, because DX were high-end systems, and SX were low-end systems.