PCI Lock. is it really needed?

[Wes Newell]

I think not considering the dividers used today.
Consider that running the default 200MHz has a divider of 6. So take it up
to 217MHz and the PCI speed is still less than 36MHz. I don't know of any
half way decent PCI hardware that won't run on a less than 10% overclock
of the bus. All of mine will run happily at up to about 38MHz. So as long
as dividers of 7 and maybe 8 later are provided I don't really see a need
for a PCI lock ever again. When the bus goes over half way to the next
level, just use the next higher multiplier. And the higher the bus goes,
the less fluctuation. With a divider of 7, a half way overclock to 249
would only put the PCI bus at 35.5MHz.

I bring this up because of all the hoopla about the new PCI locked
chipsets. Just doesn't seem like a valid thing to worry about any more as
long as addtional dividers are provided. Opinions?

 

[Jon Back]

With the announcement of Via's new Pro chipset and the NVidia chipset with
working PCI lock, I can tell you I will almost certainly upgrade my
motherboard to take advantage.

When I go above 215 FSB, I get hard drive errors. That is definitely PCI bus
speed related. The processor seems to have no problem.
Then again, it means I'd need to buy memory faster than PC3200, but its only
money, right? I've got another motherboard that would love to take my old
memory.

I have been so happy with my AMD64 3000+, I just can't leave it alone (does
that make sense?) For stability, it has only been matched by an old dual
Pentium II system with a Tyan Tiger motherboard.

JonB - TeamAnandtech

 

[Erez Volach]

I think not considering the dividers used today.
Consider that running the default 200MHz has a divider of 6. So take it up
to 217MHz and the PCI speed is still less than 36MHz. I don't know of any
half way decent PCI hardware that won't run on a less than 10% overclock
of the bus. All of mine will run happily at up to about 38MHz. So as long
as dividers of 7 and maybe 8 later are provided I don't really see a need
for a PCI lock ever again. When the bus goes over half way to the next
level, just use the next higher multiplier. And the higher the bus goes,
the less fluctuation. With a divider of 7, a half way overclock to 249
would only put the PCI bus at 35.5MHz.

I bring this up because of all the hoopla about the new PCI locked
chipsets. Just doesn't seem like a valid thing to worry about any more as
long as addtional dividers are provided. Opinions?

Generally, I would agree. But there is no arithmetic bias towards this
situation: there have been non integer deviders in the past, so it wasn't
impossible to employ them at 0.5 steps say a year ago... I think the current
limitations (where available deviders are sometimes not enough to keep PCI
within tolerance while O/C) are more due to motherboard (or chipset ?)
simplicity from the designers point of view. They just care little for
providing more choice in that department (pci bus deviders) or did not see
need for it when they sat the drawing table.
In the past, how many deviders have you found on a motherboard ? I can't
remember more than 2-3. usually just 2. on intel first PCIsets it was 2/3 or
1/2 for 50, 60 and 66 MHz FSB.
I think 1/3 was introduced with 75 and 83.3 MHz FSB, and it was enough for
100 FSB too.
On KT133 PCI and AGP buses are driven via seperate clock signal path, which
is not linked with FSB or memory bus signal. That' an effective "bus lock".
I guess that with the wide range of tolerance some advanced Athlons and
their chipsets (like Tbred Bs, bartons and NF2) the range of deviders was
simply not covering enough "target" FSBs that overclockers wanted ;-)
Who is to blame ?

 

[Ben Pope]

I think not considering the dividers used today.
Consider that running the default 200MHz has a divider of 6. So take it up
to 217MHz and the PCI speed is still less than 36MHz. I don't know of any
half way decent PCI hardware that won't run on a less than 10% overclock
of the bus. All of mine will run happily at up to about 38MHz. So as long
as dividers of 7 and maybe 8 later are provided I don't really see a need
for a PCI lock ever again. When the bus goes over half way to the next
level, just use the next higher multiplier. And the higher the bus goes,
the less fluctuation. With a divider of 7, a half way overclock to 249
would only put the PCI bus at 35.5MHz.

I bring this up because of all the hoopla about the new PCI locked
chipsets. Just doesn't seem like a valid thing to worry about any more as
long as addtional dividers are provided. Opinions?

My opinion is that you should throw away the dividers and run the PCI bus
locked.

Why bother with more dividers when, for a few more pence, you can lock the
clock?

The argument I suppose, is that if you are overclocking your ram and your
fsb, then why not the PCI bus too? Well, there isn't anything hanging off
the PCI bus that could benefit from overclocking it.

Ben

 

[Ian Hastie]

I think not considering the dividers used today. Consider that running the
default 200MHz has a divider of 6. So take it up to 217MHz and the PCI
speed is still less than 36MHz. I don't know of any half way decent PCI
hardware that won't run on a less than 10% overclock of the bus. All of
mine will run happily at up to about 38MHz. So as long as dividers of 7
and maybe 8 later are provided I don't really see a need for a PCI lock
ever again. When the bus goes over half way to the next level, just use
the next higher multiplier. And the higher the bus goes, the less
fluctuation.

You can't assume this. It's more than likely that the overclocked bus
will have an increased fluctuation itself. Whether the increase will be
cancelled out by the higher divider is pure speculation without actually
measuring it.

With a divider of 7, a half way overclock to 249 would only
put the PCI bus at 35.5MHz.

I bring this up because of all the hoopla about the new PCI locked
chipsets. Just doesn't seem like a valid thing to worry about any more as
long as addtional dividers are provided. Opinions?

Overall, it's an extra option. If your system works properly without
locking then why not. If you need the lock for stability or prefer not
to overclock the PCI bus then you can do that too. It's always foolish
to prescribe one set of rules for everyone. I think you're also
forgetting the AGP bus which can be very difficult to overclock at all.

 

[barry]

...
Why bother with more dividers when, for a few more pence, you can lock the
clock?

The argument I suppose, is that if you are overclocking your ram and your
fsb, then why not the PCI bus too? Well, there isn't anything hanging off
the PCI bus that could benefit from overclocking it.

Ben

Is there a program that can read the actual pci and agp bus speeds and
report them ?