Where did 33/66/100(99) FSB originate from?

[BCinSC]

Back in the days of the IBM XT, it was 4.77MHZ, then 8 & 12MHz for 286; the 386 was 16, 20, 25, and 33 (plus AMD's 40MHz, though I know there were some other bastard schemes out there with 6 & 10MHz XT, and 10 & 16MHz 286). The Intel camp seemed to settle on 33.333MHz for the 386/486 (and original Pentium?). At some point, the jump to 66.667MHz came along, and then 100 and now 133. How did that come to be? Why 33.333 and the subsequent double, triple and quadruple? How does the timing crystal work?

 

[AndyHui]

We got to 33, synchronous motherboard and CPU speeds, then motherboard makers found that it was hard to keep up sending 50MHz to the board while the CPU was racing along at 50MHz.....

That's how we got double clock with the Intel 486 DX2/50 and DX2/66. It was cheaper to produce a system where the motherboard was operating at 25 or 33MHz rather than doing something like a 486 DX/50. If you ever compared the prices between a DX/50 and a DX2/50, you will know what I mean. The speed of the processor was allowed to go up while the motherboard didn't need to. We got stuck with those speeds as Intel simply used a multiplier on the chip in order to scale the frequency.

The Intel Pentium brought in the 66MHz FSB as it was a big jump in performance over the 33MHz 486 bus....and we've been stuck with that sort of speed since the 1st Pentium/60 and 66s in 1992.

 

[BCinSC]

But why 33? And I forgot about the 486/50 on 50MHz board. Even had a couple at work way back when and yes, they were REALLY expensive and frequently incompatible with add-in cards.

 

[BCinSC]

No other comments?

 

[dflipflop]

No other comments, except to say that 25 years later, I had this same question without finding a good answer. My guess is that they could have just as easily picked 32 MHz or 34 MHz, and that it's somewhat arbitrary.

 

[Ken g6]

I'm guessing it's because of this from Wikipedia:

Manufacturers have difficulty producing crystals thin enough to produce fundamental frequencies over 30 MHz. To produce higher frequencies, manufacturers make overtone crystals tuned to put the 3rd, 5th, or 7th overtone at the desired frequency, because they are thicker and therefore easier to manufacture than a fundamental crystal that would produce the same frequency—although exciting the desired overtone frequency requires a slightly more complicated oscillator circuit.[14][15][16][17][18]A fundamental crystal oscillator circuit is simpler and more efficient and has more pullability than a third overtone circuit.Depending on the manufacturer, the highest available fundamental frequency may be 25 MHz to 66 MHz.[19][20]

So I'm guessing it was the fastest fundamental crystal oscillator they could get at the time.