TMS 9900 versus VAX

[apersson850]

In another thread, @retroclouds asked for a comparison between the TMS 9900, with its microcomputer origin, and the superminicomputer Digital VAX 11/780.

Here's an attempt. As always, more things could have been included, but I hope I touched the more interesting parts.

TMS 9900 versus something big.pdf

[Reciprocating Bill]

Another interesting comparison is between the 9900 and the LS-11, a four-chip implementation of the PDP-11. In the Book "16-bit Microprocessors" (Titus, Titus, Baldwin, Hubin and Scanlon, 1981, Howard Sams) several small assembler benchmarks were used for performance comparisons. Here are the numbers for the 9900 (numbers I obtained long ago utilizing pad ram on a 99/4a), an 8086 at 4Mhz, and an LS-11. The benchmarks were:

 

- A bubble sort of 600 numbers arranged in the worst-case high to low order,

- A square root routine, tested with three values,

- a string search

- finding a value in a lookup table

- a memory transfer

 

Bubble sort (seconds)

   9900  15.3"

   8086  13.9"

   LS-11 10.5"

 

Square root of 10,000 (microseconds)

   9900  490

   8086  637

   LS-11 628

 

Square root of 16,384

   9900  370

   8086  480

   LS-11 458

 

Square root of 58,081

  9900  370

  8086  480

  LS-11 457

 

String Search (microseconds)

  9900  1075

  8086  767

  LS-11 979

 

Sine Lookup Table (from four quadrants of table, microseconds)

  9900  84, 86, 77, 79

  8086  36 to 43 microseconds

  LS-11 64, 74, 75, 85

 

Transfer 256 words in memory (microseconds)

  9900 4308

  LS-11 3265

  (8086 not reported)

 

Of course, later 16/32 bit processors just coming online at the time (e.g. the 68000, etc.), also discussed int he book, blew the above numbers away. 

 

 

 

[apersson850]

The VAX 11/780 CPU, KA 780 was implemented mainly by using standard Schottky TTL circuits. Smaller versions of the computer had different implementations. The 11/730 implemented the processor with the popular AMD 2900 bit-slice chips. It had less performance, but also less cost.

[+TheBF]

2 hours ago, Reciprocating Bill said:

Another interesting comparison is between the 9900 and the LS-11, a four-chip implementation of the PDP-11. In the Book "16-bit Microprocessors" (Titus, Titus, Baldwin, Hubin and Scanlon, 1981, Howard Sams) several small assembler benchmarks were used for performance comparisons. Here are the numbers for the 9900 (numbers I obtained long ago utilizing pad ram on a 99/4a), an 8086 at 4Mhz, and an LS-11. The benchmarks were:

 

- A bubble sort of 600 numbers arranged in the worst-case high to low order,

- A square root routine, tested with three values,

- a string search

- finding a value in a lookup table

- a memory transfer

 

Bubble sort (seconds)

   9900  15.3"

   8086  13.9"

   LS-11 10.5"

 

Square root of 10,000 (microseconds)

   9900  490

   8086  637

   LS-11 628

 

Square root of 16,384

   9900  370

   8086  480

   LS-11 458

 

Square root of 58,081

  9900  370

  8086  480

  LS-11 457

 

String Search (microseconds)

  9900  1075

  8086  767

  LS-11 979

 

Sine Lookup Table (from four quadrants of table, microseconds)

  9900  84, 86, 77, 79

  8086  36 to 43 microseconds

  LS-11 64, 74, 75, 85

 

Transfer 256 words in memory (microseconds)

  9900 4308

  LS-11 3265

  (8086 not reported)

 

Of course, later 16/32 bit processors just coming online at the time (e.g. the 68000, etc.), also discussed int he book, blew the above numbers away. 

 

 

 

That looks pretty good for our old friend.

I bet 9900 would beat the others if they measured context switching time.

I have had 40 stack manipulation dummy tasks running simultaneously on my Forth system and the interpreter still responds reasonably well.