update: was [[rescue] Did I kill the octane?]

[Francisco Javier Mesa-Martinez]

On Mon, 29 Dec 2003, Dave McGuire wrote:

>
>    Hmm, near-identical FP performance.  I seem to recall someone
> mentioning that the big difference between the R10K and the R12K was a
> set of bug fixes to the caching systems which dramatically increase hit
> rates.  Does anyone else remember that?
>
>    We should find some R-series family information to see what the
> differences really are.

As far as I know the R14K and the R12K are pretty much the same
micro-architecture. The R14K utilices a different process/die since it was
re-routed (I assume for speed), and it can handle DDR external caches
which I believe the R12K and R10K could not use. The external clocking and I/O
specs seem to be identical to the R12K (which are 2x the speed of the
R10K), both chips seem to be pin compatible -but knowing SGI they will put
some weird change so the 2 chips can not be interchangeable just like
that. Other than that it seems that the R14K is just an R12K that can be pushed faster.
So on a clock per clock basis both chips should perform about equal,
some rough SPEC scaling seems to agree with such an speculation.

I will try to dig some papers, but this is from old recollections. I could
not find any real specs from SGI's website (which is annoying since their
tech specs have gone to hell)