Why does PC video memory base address change depending on video mode?

Question

It is rather well-known that, with VGA-compatible PC video adapters, in black-and-white text modes video memory is available at linear address 0xB0000, in colour text mode at address 0xB8000, while in graphics mode, it appears at address 0xA0000. Ralf Brown’s has a whole table showing which video modes use which segment bases.

It seems it would have been simpler to always place the video memory at the same address. Indeed, this design created some complications when a protected-mode video ROM interface was created; an operating system had to allocate three selectors to video memory to use it, instead of just one.

What is the reason for the changing memory mapping that we ended up with? Was it a deliberate design that served a useful purpose, or purely an accident, a compromise between an attempted do-over and backwards compatibility?

Answer 1

The history behind the various base addresses is as follows:

• the area starting at A0000h was set aside for video frame buffers in general (see Who set the 640K limit?)
• MDA used 4KiB starting at B0000h (out of 16KiB reserved, and 32KiB occupied in practice because of partial address decoding in the MDA)
• CGA used 16KiB starting at B8000h (see also Who decides what is the memory address that the CGA video buffer will be mapped to?)
• EGA introduced modes based at A0000h, using planes of up to 64KiB, depending on the amount of memory installed; this wasn’t backward-compatible with MDA or CGA, so a new address was chosen

I don’t know why the MDA and CGA base addresses were chosen in the B0000h bank rather than A0000h; but there is a useful side-effect to the choice: MDA and non-MDA modes can be used in parallel, which enables dual-monitor setups. Thanks to the non-overlapping display buffers, and to the non-overlapping I/O port ranges (3B0h–3BBh for MDA, 3D0h–3DFh for CGA; EGA adds 3C0h–3CFh), a variety of combinations are possible, starting with CGA and MDA in the original IBM PC (John Elliott’s “Dual-Head Operation on a Vintage PC” gives details). “Common” uses of this feature included displaying a running program on the colour screen and a debugger on the monochrome screen (even under Windows), or a CAD render on the colour screen and UI on the monochrome screen.

Aside from dual-monitor use, the combination of features provided on the MDA and CGA required different ports and memory addresses to be used. On PCs at the time, there was no central coordination of address assignment; each card in the system was equal on the bus, for both memory addresses and I/O ports, and chose which addresses it responded to. The original MDA also included a parallel port; even if dual displays weren’t envisioned, users buying a PC with MDA and later installing CGA might want to keep their MDA to connect their printer. If the MDA and CGA buffers or port ranges overlapped, the system wouldn’t behave correctly; it was probably simpler to map the adapters to separate ranges rather than add jumpers to choose which card “owned” the relevant addresses (as seen on EGA).

Dual-monitor use seems to have been an intentional decision; at least, it is mentioned in the original IBM PC Technical Reference, in the DIP switch settings for monitor type selection (the MDA setting is listed as “IBM monochrome display or both types of display adapters”; PCs with dual monitors used the monochrome monitor as their primary display). The comments on dual-adapter purchases in the internal IBM PC Q&A suggests that simultaneous dual-monitor use wasn’t a well-known feature however, at least not in 1981. Since the MDA also hosted the parallel port, PCs with both MDA and CGA weren’t unheard of, but most CGA-equipped PCs would only have a single, colour, monitor (and switches configured for CGA-only operation). (A separate card providing only a parallel port was also available.)

Bear in mind that the PC wasn’t really designed as a family of computers, and didn’t have an expansion-ready high-performance API of any kind, so many of its implementation details became features, simultaneously sub-optimal and locked-in for a long time.

Answer 2

TL;DR: Compatibility

Before VGA there was MDA, CGA and EGA and it's all about continued support.

But Why Different Memory Regions ?

Video in the IBM PC is based on video cards bringing their own memory. While it is mapped into CPU address space, Video access can only happen from the card itself. Thus avoid CPU side collisions (bus war) when operating more than one card, they had to use different address regions.

Was it Intended to Have More Than One Parallel?

Sure. While the on-board BIOS did only support one card at a time, the Technical Reference does note that configurations of MDA plus CGA are intended, for example on page 1-125 regarding initialisation:

(IBM PC technical Reference 4/1983 p.1-125)

Or when it's about switch settings needed for the BIOS startup when MDA and another Adapter is installed. This ensures that the BIOS will use MDA as default.

(IBM PC Technical Reference 4/1983 p.G-6)

Is It Useful?

Oh yes. Developers loved to have two screen setup, one for the application and a second one for debugging. Some debuggers did support the setup quite well. But the same goes for application software, where offering use of a second screen was one of the advantages the IBM had over other systems of the same time.

Development History

IBM reserved the space above A0000h for extensions, with 128 KiB at A000h reserved for Video (*1). In the beginning, with the 1981 PC, two cards with 16 KiB video RAM each were offered. They got each assigned half the 64 KiB address space:

• MDA starting at B0000h
• CGA starting at B8000h

While both could feature the same memory layout, having them at different address enabled parallel use (*2). Later EGA (1981) was added with the PC-AT. Beside emulating MDA and CGA at their respective addresses, it also offered new video modes needing a continous 64 KiB buffer. To facilitate this, while still allowing parallel use of a second screen with MDA or CGA (or both), it needed a new base address:

• EGA starting at A0000h

This meant that the originally reserved 16 KiB at A0000h were now assigned as buffer(*3), thus the video-BIOS ROM needed for operating the EGA had to be placed else were. 32 KiB at C0000h was chosen for this. Which also adds an issue for use of EGA cards in PCs, as only the October 1982 BIOS added a ROM search at C0000h.

VGA in turn simply inherited the address layout for compatibility.

---

*1 - More exact, 16 KiB at A0000h reserved and 112 KiB starting at A4000h for video buffer. One may guess that the first 16 KiB were intended for usage by ROM extension. See page 2-27 of the first (8/81) edition of the PC Technical Reference.

*2 - All of that being a constrain of each card bringing it's own video memory.

*3 - See page 1-14 of the April 1983 Technical Reference.

Answer 3

For CGA vs. MDA, the reasoning for using different base adresses is simple and obvious: These two video systems are designed to co-exist for dual-monitor solutions thus each need a unique memory range.

When VGA was introduced, IBM apparently wanted to continue support the dual-mode setup, so VGA had to occupy yet another distinct address. IBM's original VGA card could co-exist with the MDA card as well