screen "2.5" mode

Quite pointless speculation, but yes... I think it would have fitted in nicely considering CPU speed and I think it would have definitely found some user base.

Most close to those resolution modes on MSX you can get, if you find your self an MSX1 with 16kB of VRAM and V9938... In that case you can get your self a 128x128 screen... Although it will be with 256-colors and other ways pretty useless also, LOL.

The Sega Master System VDP is 4 years younger, has much more logic gates and require a dual bank of 8KB VRAMs, to double the bandwidth. Not sure if it's a fair comparison.

Disadvantages that most people don't know:

1) It has only 512 tiles instead of 768. This means that 1/3 of the screen will have to be composed of repeated tiles. No problem for a dedicated game machine, but inadequate for a real general-purpose computer.

2) There's no free slot to access the VRAM outside of vblank. The video uses all the VRAM bandwidth. Again, not a big problem for a dedicated game machine, but inadequate for a real general-purpose computer. People complain that it's too slow to wait 29 cycles between accesses, imagine having to wait an entire frame for the VBLANK.

Being a dedicated games console, the NES also has astringent restrictions that make it unusable as a general-purpose computer.

1) There's only 2kB of VRAM, enough only for the SAT and name table. The graphics must reside in a VROM on the cartridge
2) There are only 256 tiles. This means that 2/3 of the screen will have to be composed of repeated tiles
3) You can also only access the VRAM on VBLANK. There are no available slots for the CPU when the screen is being drawn

There's no magic: it's the short blanket syndrome. Each machine was designed with focus on its application.

Horses for courses. In some situations wide pixels, and then four unique colours on a row, has advantages over square pixels with only two colours on a row. Similarly, larger square pixels (2x2) can have their purposes. Perhaps 192 bytes could've been used in order to have two bits specify for each of the 768 tiles what type of tile it would've been. sc2-style, sc2-style vertical, wide pixels, 2x2-pixels. The 2x2-pixels would've been a neat demo mode then if you'd have the right way to access those pixels (e.g. not like how sc3 pixels are accessed). In fact with 4x4 big pixels in any colour you'd save 8 bytes per tile, now what to do with those?

In fact, perhaps it'd been better whether screen 11/12 would have used a 2x2 area for its colour space rather than 4x1. As artefacts are more obvious with a wide line of pixels than with a smaller block of pixels.

So... four years on and still the same amount of VRAM. I can't help to think it shows how expensive memory chips were back in the days.

AFAIK, the type of PSRAM used by the SMS VDP didn't exist when the TMS9918 was designed.

Not only that, to keep up with the timing, each PSRAM is connected to the SMS VDP via a dedicated data bus, totalling 16bits.

One of the key cost factors on the 70s and early 80s was that the chip wouldn't surpass 40pins. More than that and the chip cost would escalate to the heavens, so such luxury was reserved only for chips targeted at mini-computers and expensive workstations.

This is why nearly all chips you find in a early 80s computer have a glass ceiling of 40 pins.

The SMS VDP belongs to a newer generation, when the SDIP-64 form factor became available. Many other chips of that era also took advantage of this, including the V9938, Y8950 and HD64180.

To save even more costs, the SMS2 VDP changed from SDIP-64 to the "zigzag encapsulation" that later became available.

That would have increased the (expensive) silicon area. Let's remember that the TMS9918A (yes, the 2nd revision that added the screen-2 and screen-3) didn't receive a smooth scroll feature because it was going to demand more silicon area than its target price would allow.

There's a message from the chip engineer explaining this on an ancient BBS backup. It's old, but with some luck and careful search you can find it.

So, adding different video modes fall in the same category. If you think carefully, screen-2 and screen-3 are just very neat tricks applied over the way the old TMS9918 (without A) worked:

- screen-2 expands the name table just by increasing its pointer after each 1/3 of the screen, and increases the color attribute pointer on each line, creating the famous 8x1 color attribute instead of 8x8 of a normal text screen.
(The ZX-Spectrum could have had this very same 8x1 color attribute trick, but they didn't notice this opportunity. Only the designers of the Timex 2068 clone noticed and implemented that)

- screen-3 is just the screen-2 tweaked to disable the pattern table and use only a fixed 4x1 pattern. Then instead of increasing the color attribute pointer each line, it's increased each 4 lines, creating a 4x4 pixel pattern.

As I explained in many other similar discussions here (this topic is incredibly recurrent), nearly all 80's hardware are based on nifty and very carefully planed tricks to save every possible penny.

Yes, RAM was one of the key cost factors during the whole 80s. To make things worse, when prices were beginning to drop around 1986, there was this famous fire in one of the biggest DRAM factories and prices skyrocketed again. I suspect the MSX2+ standard was planned to be released with more minimum RAM (even the DOS2 was planned to use that), but they had give up because of the expensive RAM prices.

If you notice, the MSX standard was also designed to use as much ROM as possible, leaving the RAM usage at a minimum. An MSX2+ would have required *much* more RAM if all the features it had were not in ROM. Same deal with both the MSX-DOS1 and MSX-DOS2.