It's cool because the registers are all in RAM, with a "workspace pointer" on the CPU pointing at where they are. This is slow, but a context switch is just changing that pointer.
It's not all that slow as a concept at that time when RAM speeds were as fast as CPU speeds. I think it's just that TI's implementation of the concept in that particular cost-optimised home computer was pretty bad -- the actual registers were in 256 bytes of fast static RAM, but the rest of the system memory (both ROM and RAM) was accessed very inefficiently, not only 1 bytes at a time on a 16 bit machine, but also with something like 4 wait states for every byte.
The 6502 is not very different with a very small number of registers and Zero Page being used for most of what a modern machine would use registers for. For example (unlike the Z80) there is no register-to-register add or subtract or compare -- you can only add/sub/cmp/and/or/xor a memory location to the accumulator. Also, pointers can only be done using a pair of adjacent Zero Page locations.
As long as you were using data in those in-RAM registers the TI-99/4 was around four times faster than a 1 MHz 6502 for 16 bit arithmetic -- and with a single 2-byte instruction doing what needed 7 instructions and 13 bytes of code on 6502 -- and it was also twice as fast on 8 bit arithmetic.
It was just the cheap-ass main memory (and I/O) implementation that crippled it.
Well, it has 256 bytes of RAM which is basically a really big register file, and everything else goes in the 16kb of "video RAM" which you can read and write by poking at I/O registers. So it is not easy to program.
It's arguably the only 8-bit computer which has a really different architecture from the others. You could otherwise imagine pulling the SID chip off a C-64 and putting it on a TRS-80 Color Computer etc.
Sharing the main RAM with video was a weak point in computers of that time period because the video system stole many of the memory access cycles. Some recent retrocomputers that revisit that period like
have a full-size memory bank and a video RAM memory bank which is accessed through a port which can be pretty efficient because you can auto-incremement the address register and just write 1 byte to the port to write 1 byte to video RAM and repeat.
The 9900 was exactly contemporary with the LSI-11 CPU. Both TI and DEC were taking advantage of new LSI gate-counts to move discrete TTL CPUs into one chip.
The 990 series of minicomputers were competing with PDP-11s (Though DEC had highest market share, I believe 33% of the whole mini market?)
The 9900 was condensed in 1975 and went into the low-end 990/4. The higher end 990/9 and 990/10 were always going to be discrete TTL as the 9900 didn't support memory protection or mapping to the 2MByte total address space.
TI was always conscious of not challenging IBM head-to-head in minicomputers. Internal memos always projected TI's plan for its minis to occupy a space well below the latest IBM mainframes. From 1980, the planned 990/12 would arrive just as IBM delivered more compute power in their low-end... this was intentional, supposedly because IBM was the chief driver of TI's transistor business!
Yep, but it lacks a MMU so memory protection and paging are going to require a lot of work. I think the only reason this is feasible at all is they're running the OS out of a ROM cartridge.
The PDP machines that Unix was developed on had MMUs, which they needed because the 16 bit processors couldn't address the multi-megabyte address space the hardware supported.
The PDP-10 had an MMU similar to a modern MMU with page tables and such, the PDP-11 had an 8-segment-of-8kb MMU like what the TRS-80 Color Computer 3 had except the PDP-11 had a real supervisor mode and if a user mode program tried to change the MMU configuration it would fault.
In fact the 9900 itself was used in an entire line of minicomputers that included a hardware memory map.
The 9900 is a single chip implementation of the CPU board in the TI 990. They even created a dedicated memory mapper chip to go along with the product line, though it is significantly different than the one in the minicomputer line.
(edit: the 990 was first built in the early 70s, memory mappers are quite old conceptually)
(edit 2: in fact the necessity of using a memory mapper is what killed the platform, and was one of the things that made the IBM PC team decline the 9900.)
Unfortunately I don't think that there is a reasonable way to perform real hardware-level memory protection with that chip alone. I'm working on a project documenting the genesis of the 99000 chips, which include a privilege bit in the status register, from the minicomputer line.
Essay forthcoming, and probably an OS. Maybe a year...
One chip that could be used as a memory mapper for the 9900 (but wasn't in the TI99/4A) was the 74LS670, which was used in the IBM 5150 PC to allow the 8237 DMA chip to access more than 64KB (a limit that wasn't a problem when used in a 8080 system).
The 16K memory block is connect to the TMS9918 video chip and is accessed by port read/write. The TI-99 out of the box only has 256 bytes of RAM memory on the 16 bit buss.
UNIX99 uses an after market card with 1Mbyte of RAM on an 8bit external buss used for peripherals. This card replaces the old 32K RAM card made by TI in the 1980s.
All due to TI’s desire to use the same chip standards across all their machines big and small, IIRC.