SUMMARY: Different MHZ modules running together

From: Fletcher Cocquyt (
Date: Thu Mar 21 1996 - 10:12:16 CST

Original post:

Subject: Different MHZ modules running together

Simple multi-processor solaris 2.5 question:

Is it advisable to run different clock speed CPU modules together?

What are the considerations/implications? Do you effectively get two
of the slower speed CPU? ie running a mod50 & mod60 results in effectively
two 50's?


Short answer seems to always be: don't do what Sun doesn't support and
as "Andrey S. Ryzhov, CS Tech.Supp.Engr." <Andrey.Ryzhov@Russia.Sun.COM>

It is *absolutely* not supported by Sun.
I got some reports about successful experiments of mixing MBus modules
from one generation (i.e. SuperSPARC+ likely won't work with
SuperSPARC-II but 50MHz and 60MHz SS+'s may work together at 50MHz).

You may try at your own risk, but I'd better change one of MBus modules.

Although, John DiMarco <> gives a more practical answer:

SuperSPARC modules without SuperCACHE can't be mixed with those with
SuperCACHE, and SuperSPARC and HyperSPARC modules can't be mixed.
Further, SuperSPARC modules without SuperCACHE run at the bus speed,
so mixing two such modules of different speeds will either result in one of
the modules being overclocked (if the bus is running at the faster of the two
speeds), or both modules running at the speed of the slower module.
Finally, some early 33/36/40MHz SuperSPARC modules without SuperCACHE don't
support multiprocessing at all.

Mixing two SuperSPARC+SuperCACHE modules of different clock speeds does
work, although the operating system will not take the different speeds into
account. You may find that adding a slower module to a system with faster
modules may make a given application run slower, because critical-path code
may be scheduled on the slower processor. Mixing two HyperSPARC modules
of different speeds will also probably work.

The resulting system will have throughput somewhere between that of a system
with all slower processors and a system with all faster ones. Program
run length will probably be nondeterministic.

By the way, all such configurations are unsupported.


Its my experience that what John says is 100% true.
Thanks for the info!

Fletcher Cocquyt (809) 299-2900
System Administrator Trout Trading Hamilton, Bermuda

This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:10:55 CDT