Dual CPU Matching Requirements

[Norm Dresner]

Okay, I've seen statements to the effect that for a Dual CPU board to
function "properly", the two CPUs must be matched for
1. Type --- pretty obvious
2. Speed --- I guess as long as the system runs at the slower of the
two speeds it's probably okay but the same is best
3. Revision/Stepping -- Huh? Why?


TIA
Norm

 

[Rob Stow]

Okay, I've seen statements to the effect that for a Dual CPU board to
function "properly", the two CPUs must be matched for
1. Type --- pretty obvious

Depends on how narrowly you define "type".

I have successfully run an Athlon XP and a Duron on the same Asus
motherboard. Worked fine with both CPUs running at 13 x 100 MHz.

2. Speed --- I guess as long as the system runs at the slower of the
two speeds it's probably okay but the same is best

Depends on the processors/motherboard/BIOS, but generally you are
pushing your luck if the speeds don't match. For example, I
have gotten away with speed mismatches between Athlon MPs on Tyan
S2460 or S2466 motherboards, but not on Asus or MSI motherboards
that uses exactly the same AMD 760 or 760MPX chipset. And, yes,
on the Tyan boards the processors both ran at the speed of the
slower one.

3. Revision/Stepping -- Huh? Why?

Intel processors are a lot fussier about this than AMD processors
are. With the Athlon MPs and the Opterons, so long as the
processors are the same speed, chances are they will play nicely
together.

I have added a second processor to about a half dozen systems
that initially had a single Opteron 24x processor and it has
worked every single time even though absolutely no effort was
made to match revision/stepping - and an AMD techie had advised
me that I needed to make no such effort. That same techie,
however, advised me that when SSE3 capable Opterons become
available this might not hold true any longer.

 

[Tony Hill]

Okay, I've seen statements to the effect that for a Dual CPU board to
function "properly", the two CPUs must be matched for
1. Type --- pretty obvious

Generally speaking, yes.

2. Speed --- I guess as long as the system runs at the slower of the
two speeds it's probably okay but the same is best

Depends on the system, though generally this is the case. In systems
where the bus is not shared (ie basically everything other than Intel
processors), then it's at least theoretically possible to run
unmatched speeds. Whether or not an individual system will actually
work with such a setup is another question though.

3. Revision/Stepping -- Huh? Why?

Really not an issue these days for the most part, that's more a
throw-back to the Pentium and PentiumPro days of yesteryear.

With a modern Xeon or Opteron you should have little trouble mixing
steppings, though ideally you should try and keep the same if you can.
Mixing steppings can lead to some REALLY obscure problems, but they
tend to be VERY rare.

Of course, you're forgetting one VERY important thing: voltage.
Generally speaking you MUST match the voltage for your processors.
Usually if you've matched the type and the speed than you're safe, but
there are some exceptions to this.

 

[Norm Dresner]

Generally speaking, yes.


Depends on the system, though generally this is the case. In systems
where the bus is not shared (ie basically everything other than Intel
processors), then it's at least theoretically possible to run
unmatched speeds. Whether or not an individual system will actually
work with such a setup is another question though.


Really not an issue these days for the most part, that's more a
throw-back to the Pentium and PentiumPro days of yesteryear.

With a modern Xeon or Opteron you should have little trouble mixing
steppings, though ideally you should try and keep the same if you can.
Mixing steppings can lead to some REALLY obscure problems, but they
tend to be VERY rare.

Of course, you're forgetting one VERY important thing: voltage.
Generally speaking you MUST match the voltage for your processors.
Usually if you've matched the type and the speed than you're safe, but
there are some exceptions to this.

At least for Intel Pentium-X (X=2,3, or4), I'd have thought that
matching type and speed would have been sufficient. Isn't it?

Thanks
Norm

 

[Tony Hill]

At least for Intel Pentium-X (X=2,3, or4), I'd have thought that
matching type and speed would have been sufficient. Isn't it?

That depend son how closely you define "type". For example, I suspect
that a PIII "Coppermine" chip, built on a 180nm process, will not play
nice with a PIII "Tualatin" chip, built on a 130nm process, even if
they are the same clock speed (there was some overlap, at least at
1.0GHz and 1.13GHz). Coppermine + an older Katmai PIII, built on a
250nm process, is probably going to be even worse.

The P4 would have similar requirements except that it is not
dual-processor capable at all. You have to get a Xeon for any sort of
dual-processor capabilities from Intel today.