Trying to explain some of the anomolies in behaviour of a system using 133MHz Front Side Bus, I decided to test out a hypothesis.

I like many of us, I had assumed that the P2B running at 133Mhz Front Side Bus would actually be clocking the PCI bus at one third, ie 44MHz. This would be higher than the 41Mhz (83Mhz FSB) and would cause problems with many PCI cards. However I heard reports from a guy (thanks Nikos) that he could run fine at 133Mhz FSB but not at 83Mhz, which puzzled me.

I started thinking, and the only explanation that I could come up with was that the P2B was dividing the FSB by 4 to get the PCI bus speed. If so, then this would be very good news for overclockers. As it takes the PCI bus out of the loop, it is only the SDram, processor and chipset which are clocked above spec, the PCI cards are no longer being overclocked.

I took the scope home this weekend, and while it was difficult (trying to sample 40MHz signals on a 60MHz scope is a fine art, as I found out), I did manage to get some very interesting results. It turns out that as I suggested 133MHz is indeed using PCI of 1/4 (ie 33MHz).

I kept the same (highest) time base, and counted the total number of clock cycles on the scope screen. This was the most accurate I could do, but the figures match and are consistant:

Jumper Settings for P2B	FSB (MHz)	Official PCI Bus Speed (MHz)	# cycles in 10 divs @ 0.005us/div	Measured PCI Bus Speed	Measured PCI Bus Divider
2-3 | 2-3 | 2-3	50	25.0	12-13	24-26	1/2
1-2 | 1-2 | 2-3	66	33.4	16-17	32-34	1/2
1-2 | 2-3 | 2-3	75	37.5	18-19	36-38	1/2
2-3 | 1-2 | 2-3	83	46.1	20-21	40-42	1/2
1-2 | 1-2 | 1-2	100	33.3	16-17	32-34	1/3
2-3 | 2-3 | 1-2	103	33.4	16.5-17.5	33-35	1/3
1-2 | 2-3 | 1-2	112	37.3	18-19	36-38	1/3
2-3 | 1-2 | 1-2	133	Not Given	16-17	32-34	1/4

So, in my eyes, it makes 133Mhz FSB a viable alternative to 100MHz, 103MHz and 112MHz Front side Busses. As long as you have SDram which works reliably at this speed, it gives you a whole slew of new settings to try most of which duplicate existing 66Mhz/100MHz speeds (i.e. 266/333/400) but also allows 466 (133*3.5, shame there was never a Celeron 233 *8').

The memory bandwidth/latency improvements are negligable, since you will probably have to increase CAS latency from 2 to 3 to get stable SDram. Even so a cache line load will take 75ns (7-1-1-1 @ 7.5ns) as apposed to 90ns (6-1-1-1 @ 10ns), and even a single word access will take 53ns at 133Mhz rather than 60ns at 100Mhz.

The big question in my mind - Is the P2B unique? Or do other motherboards which feature 133Mhz FSB also use /4 PCI clock?

Take care,

Mark.........
--
If the world were an oyster it would be mine.......

Well Mark, it looks like this oyster is yours, on behalf of AnandTech I would like to thank you for the email and the hot tip.