Overclocking P2B theory - help my brain

[Bob]

I come from the old overclocking days when we usually cranked the
multiplier, not the bus, so help me out here:

If I understand correctly, since we crank the bus now, I would get
better performance from (for example) an 850/100 as opposed to an
850/133 since the 850/100 has an 8.5 multiplier and the 850/133 has
a 6.5 multiplier ? I'm thinking that an 850/100 running at 133
is going to give me 1.1ghz and keep the PCI bus correct at the same
time. Do I have that right ? Is this realistic on a P2B and will I
need to jump though hoops to get my memory to run this fast ?

Also, I assume that I _want_ the slocket installation instead of a
true slot-1 CPU if I plan to overclock since the slocket will let
me choose a higher voltage manually vs. the slot-1 P3 that will
choose the voltage for me ? In other words, if you buy a slot-1
CPU, you can only run it at the rated speed ?

Thanks,

 

[Paul]

I come from the old overclocking days when we usually cranked the
multiplier, not the bus, so help me out here:

If I understand correctly, since we crank the bus now, I would get
better performance from (for example) an 850/100 as opposed to an
850/133 since the 850/100 has an 8.5 multiplier and the 850/133 has
a 6.5 multiplier ? I'm thinking that an 850/100 running at 133
is going to give me 1.1ghz and keep the PCI bus correct at the same
time. Do I have that right ? Is this realistic on a P2B and will I
need to jump though hoops to get my memory to run this fast ?

Also, I assume that I _want_ the slocket installation instead of a
true slot-1 CPU if I plan to overclock since the slocket will let
me choose a higher voltage manually vs. the slot-1 P3 that will
choose the voltage for me ? In other words, if you buy a slot-1
CPU, you can only run it at the rated speed ?

Thanks,

For info on the various models, Roland has an excellent FAQ:

http://homepage.hispeed.ch/rscheidegger/p2b_procupgrade_faq.html

While the boards share the 440BX chipset, they have different
voltage regulators and different clock chips. Depending on your
revision of P2B, a slocket equipped with VID jumpers would make
it easier to set the voltage to 1.8V, to get a Coppermine processor
to work on an older P2B. So, the slocket gives you that flexibility.
However, if you are handy with a knife and soldering iron, it is
also possible to hack up the copper to get the VID setting you want.
VID pins are accessible on the back of the S370 socket on the slocket,
and you can also get at the tracks that go to the SC242 connector
on the edge of the slocket.

For clocking, there are a couple of issues. Some of the clock generators
used in the P2Bxx family are really 100MHz only devices, and while
they have higher settings available, typically only the 112MHz setting
works. Basically, the rise time on these devices is limited.

For other devices, they work up to 133MHz or higher, and then the issue
is the resulting PCI/AGP dividers. The AGP has two settings, 1/1 or 2/3,
and (2/3)*133=89MHz - this is higher than the nominal 66MHz, and all but
a couple of ATI heavyweights, like the 9700 and 9800 can handle that.
(I've seen postings in Abxzone.com, that suggest some of the cards are
only comfortable to 75MHz for example.)

For the PCI, sometimes there is a 1/4 divider available at 133MHz,
which puts the PCI in spec. On the P2B-DS, an extra jumper has to be
installed, to get a desirable bit pattern to cause the clock gen to
give 133/33, instead of the default 133/44 provided by Asus. AFAIK,
the P2B is OK for this - maybe the manual mentions the clock ratios ?

As a P2B modder, you may find some of the info on this site useful:
http://www.tipperlinne.com/p2bmod

If you run the FSB at 133, then PC133 memory would be useful
If you have PC100, that 133MHz setting might not work. Crucial
sells CAS2 memory for the same price as CAS3, if you are shopping
for some.

As for the processor itself, you cannot assume that just any processor
is going to overclock an arbitrary amount. Have a look at the average
and spread on overclocking here:

http://www.cpudatabase.com/CPUdb

HTH,
Paul