Re: Open Source Hardware? (SPARC)




P.S. The UltraSPARC T1 has absymal FP performance due to a single,
shared FPU which bottlenecks nicely with any FP traffic; this is
supposed to be significantly improved with the next chip (T2 or
whatever).

The T2 is unlikely to scream for single threaded FP apps, the T2
pipeline is a bit to sparse for that but it should be much better
than T1. You will should also expect to see Sun publishing
SPECint_rate numbers for the T2 something they havn't don't for the
T1 because some of the SPECint benchmarks actually execute some FP
instructions.

I've seen Sun using that as an excuse for not publishing
SPECint_rate2000 numbers. eon I think they asserted was one with a
bit of FP in it. still that is only one of the components - I would
have thought that the effect of the geometric mean would be such that
one component being wildly different wouldn't affect the overall
metric all that much.

maybe someone with a try-and-buy system can go ahead and run it and
publish - that seems to be the only way to get around clause 5f of the
Solaris and Studio 11 EULA's - schwartz's open invitation to write
evaluations of try-and-buy systems.

I suspect Sun have to be careful about just how much FP they add there
lest they blow their power budget out of the water.

rick jones
--
Process shall set you free from the need for rational thought.
these opinions are mine, all mine; HP might not want them anyway... :)
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...
.



Relevant Pages