[rescue] More on Silicon Samsara

[rescue at sunhelp.org]

On Sat, 23 Jun 2001, Chad A. Chance wrote:

> So far with this Ross 150, it feels just like my other SS20 with 2 SM61s. My
> plan is to run BSD on it, so the extra speed will be nice since SMP is not
> to be found in the BSD world. I think the extra cache is really only good
> for server type tasks, for a desktop the lower amount of cache is offset by
> the lower latency of the cache(from my understanding, the HyperSparcs have
> less wait states).

There is also an additional factor to consider.  The SuperSparcs use
physically mapped caches, these will flush less frequently under heavy
context switching than the virtually mapped cache of a HyperSparc (because
cache collisions between different processes on physical addresses are
essentially random but collisions in process address space are more
common).

That is to say, a SuperSparc will gain more of an edge against an
"equivalent" HyperSparc where heavy context switching is involved.  Most
single-process benchmarks will not show this behaviour.

> I've really been in the market for SM81s but a killer deal presented itself
> on this 150(less than $100), so I snagged it, now if I could find another to
> go with it, for some SMP action....

MBus module prices have been inflated for years, hopefully this is the
start of their path down to reality.  I had dual SM81's in my SS20 for a
while, that was a rocking box (though poor Disksuite threading eventually
caused me to go to a faster uniprocessor, an Ultra 1/140).

-James