> > Yep, I read that in a datasheet a few seconds after sending the
> > message ;-) It does have an on-die thermal diode, but the actual
> > monitoring has to be done externally.
> 
> IIRC it took the external logic to long to react on siemens systems. 
> 
> However to me this is a minor problem. I use OC fans and coolers that 
> enshure, that even in the case of a fan stop in a standard MaCase 19" 
> chassis the CPU does not overheat enough to cause problems. The only 
> scenarium that would be fatal is when more than one fan stopped 
> and/or a cooler would somehow come loose.

The issue here isn't the dead cpu and dead motherboard, but the resulting
fire burning down the entire building :-)

> But to be honest, without 
> hotswap CPU and memory (or even mainboard) real redundancy is not 
> possible. My customers do not take hotswap PCI solutions for the 
> price such systems are at.

Yes: I played with PCI hot-swap (writing some of the PCI driver support).
It was obvious we could demonstate removing a working board and replacing
it with a different one.  Dunno which faults that gave tolerence to!
More likely is that the added complexity would give additional failures...
> 
> What I am interested is the performance I get of serverworks boards: 
> having a GBit ethernetcard and a raid controller the system seems to 
> be more performant than a compareable system with other chipsets even 
> when they have much faster CPUs. The "performance" is the "feeling" 
> on the workstations connected to the server. 

Yes many x86 systems are completely unbalanced - Joe public seems to assume
that more MHz of cpu makes a better system.  Would anyone in there right
mind buy the P4 2GHz system that 'Tiny' have been advertising on UK TV.
It probably runs slower that my IPX - ok maybe not that slow..

    David