MSX and memory

It depends on your definition of powerful. All MSX 2+ machines have the same muscle (Z80). 512KB memory can be added through a Carnivore2 or a Multi Functional SCC+ card.

The Z80 has 64K of addressable space as a basis, like most 8-bit processors. The standard MSX memory lay-out is that the first 16K is used for the BIOS, the second 16K is used for the MSX-BASIC interpreter, and the last 32K is RAM, of which the top 3K is used for various system variables, and if a disk drive is attached, 6K more for the various data and code for disk operation. The rest of the RAM is available to the program, which is that 23431 bytes free you see in MSX-BASIC.

To get beyond that 64K you need to employ a mechanism called banking. You partition that 64K into blocks (let’s say four blocks of 16K each), and then swap those out as needed. Since the onset the MSX has the slots banking mechanism which allowed for extendibility but also has limitations. Then games started to employ megaROM mappers, and finally memory mappers for RAM expansions greater than 64K were introduced. The 128K RAM is provided using a memory mapper.

The problem with banking mechanisms is that at the core the Z80 is still a processor with only 64K of addressable memory. The banking gets around it, but also introduces complexity. The BIOS, the MSX-BASIC interpreter and the system memory must always stay accessible, so this means in the 64K space you only have a 16K block of BASIC program memory that can be changed.

But during the execution, different parts of the BASIC program need to be accessed, which means you constantly need to swap out what is visible in that 16K region of memory. When reading program data sequentially, you need to introduce checks to switch banks if you exceed the 16K region that is available. When reading data randomly, you need to constantly switch banks. And if you need things located in different banks of memory at the same time, things get even more complicated.

So as you can imagine, even though a single bank switch is quick, the overhead of switching those banks all the time can add up quickly if you’re not careful. The MSX-BASIC interpreter was written before the memory mapping mechanism was created, so this would mean a major rewrite of it to support the memory mapped memory, and it would likely run more slowly. This is why the MSX-BASIC interpreter does not support it at all outside of being able to use it as a RAM disk.

The full amount of memory is mostly usable for finely tuned assembly programs which have more flexibility to use the extra memory in an efficient way and avoid bank switches when they can. Though even then it can be quite cumbersome to work with, I’m certainly jealous of those 16-bit processors like the 68000 sometimes, which can address much more than 64K without the need for banking.

Lastly, about the video RAM, this is a whole other story as it holds graphics data and not program data. The amount shown in MSX-BASIC only concerns the memory available for program data.

These concepts also apply to most other 8-bit systems, and even sometimes in 16-bit systems.

Hi

I agree openmsx should be deleted....

How would you like the topic to be titled?

Does someone know why most MSX1's came with 64k RAM ? Almost no program, game or basic uses the first 32k of RAM, so to me it would sound totally logical to produce a machine with 32k of ROM, 32k of RAM and 16K of VRAM. Remember, memory was expensive in the early 80's.
Was it just marketing (better or same specs as the c64) ? Or were there other reasons ?