> 1977: 1Mhz 6502 or 4Mhz Z80  (also the Vax 11/780; Mhz?)

Vax 11/780 ran on a 10MHz clock, but took 1uS to execute a 32 bit instruction,
so lets call it 4MByte/sec.

> 1981: 4.77Mhz 8088 in the IBM PC
> 1984: 6Mhz 80286 in the IBM AT
> 1986: 16Mhz 80386 in a Compaq
> 1990: 33Mhz 80386 (mine just turned 9 & still runs, with its third hard
drive;
>        the first drive cost about $10/MB)
>
> Mhz alone is not a direct measure of performance, even with equal word size.
> In 1983 I programmed a single-board PDP-11 clocked around 21Mhz in
assembler,
> and since it had no  multiply or divide instruction, it took
> about 480 microseconds to do a 16x16 unsigned integer multiply, which is
> far longer than the 30 microseconds or so that a 4.77Mhz 8088 would take.

Indeed.  Some examples I've worked with...

IBM 1130 (designed in the 60s, I dealt with these in the early 70s), had a
2.2MHz clock but in fact took a minimum of 8 clock cycles to perform one 16
bit memory cycle, so was about 270kHz (3.6uS core cycle time), or
540kbyte/sec.

intel 8080 (circa 1975).  2 MHz clock, minimum 4 clocks per 8 bit memory
cycle, so 500kbyte/sec.

Zilog Z80 (circa 1977). 4Mhz clock, 4 clocks/8bits, 1Mbyte/sec

Intel 8088 in IBM PC: 5MHz clock, 5 clocks/8bits, 1Mbyte/sec

Intel 286 in IBM PC/AX: 8MHz clock, 3 clocks/16 bits, 5.3Mbyte/sec

Intel 386 in typical clone: 33MHz clock, 2 clocks/32 bits, 66Mbyte/sec burst,
but far less than this sustainable.

Intel 486DX2: 66Mhz clock, 2 clocks/32 bits, 132byte/sec internal, but 33Mhz
bus clock gave 66MByte/sec peak burst

Intel Pentium 90MHz, 90MHz core clock but 60MHz bus clock.  life gets complex.
CPU could execute 32 bits in 1 core clock, but memory bus typically had
minimum of 3 clocks/64bit cycle for 160Mbyte/sec peak burst.

Pentium-II 400MHz.  memory bus is 1 clock burst at 100MHz/64 bits or
800Mbyte/sec peak.  Core caches have multiple ports and dual execution units,
way hard to peg a number on.

________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm

Reply via email to