[rescue] interest in RS/6000 591 systems/parts?

[vance at neurotica.com]

On Thu, 6 Mar 2003, Francisco Javier Mesa-Martinez wrote:

> > 15.5 for the Ultra 2/1300 with 2M.
>
> I only see 11.2 for the 591, in specbench. I do not know if the 1MB
> cache can make an almost 4 point difference in the fp results.
>
> http://www.specbench.org/osg/cpu95/results/res9509/p066.html
>
> Anyhoo, the ultrasparc is pretty weak when it comes to FP anyways. The
> POWER2 is also a bit unbalanced (fp vs int performance). But it is still
> a good case for the MGZ myth, since the 77MHz part is faster than a
> PentiumPro running a 3 times that speed (in FP performance)

Well, one big difference is that the POWER2 can retire 4 (as far as I can
recall) FP ops/cycle, whereas the PPro doesn't even come close to that.
Neither does the SPARC.  POWER is *extremely* heavy on FP.  They kinda
suck at integer-heavy stuff.  Like Java stuff doesn't run all that great
on RS/6000's.

BTW, the figure I have for a 591 with no L2 cache is 12.8.

Peace...  Sridhar