thus Michael spake:
> Hello,
> 
> 
>> I realize that NetBSD doesn't support multiprocessor sparcv9
>> systems yet, but, I assume that that's only a matter of time.  I
>> might be able to help with this if I get a smp v9 system.
> 
> 
> Now that X works fine on sparc64 the next target would be SMP.
> 
> 
>> What I don't have a good feel for is how these type of systems
>> perform.  My jobs will be mostly integer work that is strictly CPU
>> bound.  The memory bandwidth needed isn't very high.  Disk I/O
>> doesn't really matter too much, ie, nfs across 10baseT is fine.
>> 
>> I have one existing sparc system (ss4, 110mhz) that compared to the
>> other systems (say a 200mhz 604e) is quite underwhelming.  Are the
>> Supersparcs and hypersparcs likely to be as slow as the microsparc
>> IIs?  And how does the Ultra 2 compare to the Supersparcs and
>> hypersparcs?
> 
> 
> I think /any/ of your candidates would be faster than a microSPARC on
> a work-per-clock ratio. These CPUs aren't superscalar ( Super- and
> UltraSPARC are, not sure about HyperSPARC ) and have no L2 cache (
> only small L1 ).

i have a SS20 with Dual 85MHz SuperSPARC II (those weren't supported by
SUN IIRC because of the huge amount of heat they generate) which is
faster than a 200MHz HyperSPARC dualie of a friend. as a rule of thumb
one can say that the SuperSPARC II is at least as fast as a HyperSPARC
with twice the clock.

furthermore, the HyperSPARCs have a weak cache interface (depends on the
model, tho).

the only way to get a HyperSPARC way beyond a SuperSPARC II machine (in
performance as well as in price ;) is to get a quad CPU HyperSPARC
(there were 200MHz dual modules, but i think you pay for them their own
weight worth in gold)...

so best bet would be a U2 with UltraSPARC II (300 or 400MHz, the latter
ones far more expensive than the 300MHz). avoid UltraSPARC I. and pray
for the NetBSD guys to get SMP running on sparc64 ;)

timo