SUMMARY: Mixing diff speed HyperSparc CPU modules, does it work?

From: Kai O'Yang (oyang@mars.fcit.monash.edu.au)
Date: Thu Aug 24 1995 - 05:14:26 CDT


My original question:
> We've got a clone Sparc10 (Mbus at 50MHz) running Solaris 2.3. The CPU is
> Ross HyperSparc @67MHz (RT625). We now want to add a 100MHz HyperSparc module
> to the machine. The sales and tech support of my vendor said yes, it works...
> Has any sun-managers tried it? I know mixing diff speeds SuperSparcs doesn't
> work, how about HyperSparcs? Someone told me it probably would work, but
> Solaris won't target CPU utilisation onto the faster CPU and there might not
> be overall speedups... I wonder if anyone has actually tried it.

The answer is yes, one can mix different speeds ROSS Hypersparc modules in the
same machine, however, there is a catch (from Wayne Schmidt):

>You will need to give precedence to the module with the largest cache size. eg
>you've 2 modules, one is a 66 mhz HyperSparc with a cache size of 256 K,
> two is a 100 mhz HyperSparc with a cache size of 1024 K.
>You can run into problems where you can "blow out" the bottom cache. Since
>there are 2 MBUS slots for CPU's, there is a precedence set up within the ROSS
>modules. Cache access of the both modules becomes linked, and is accessed in
>my understanding, in a linear fashion.
> MBUS2 1024 K
> MBUS1 256 K
>The process that will cause a cache miss in MBUS2 module, will most probably
>cause a miss in MBUS1. While not catastrophic to the machine, it IS
>catastrophic for performance. You should check this with the latest modules;
>as of the 125 mhz duals,

Also, John Cheshire wrote:
>It worked fine. It is correct, however, to say that Solaris will not target
>the faster processor, it will be just luck of the draw.

Thanks to

floating@float.demon.co.uk,
jmende01@mcs.eds.com
edz@bsn.com
cbarnard@cs.uchicago.edu
fmrco!fmrco.com!wayne@uunet.uu.net

Kai



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:10:32 CDT