21264 information

[Greg Lindahl]

> Most of that is information I already have.  I was wondering more about
> the 21264's actual speed... the only response to that I've ever received
> was 2 BIPS, 1.4MFLOPS... I'm not sure if this is correct or not.

There are estimated SPEC95 marks which have been published. You really
need to wait until more real systems show up to find out this kind of
stuff, I'm afraid.

> Apparently the FPU is much improved over the 21164, which was somewhat
> lacking compared to other processors.

The 21164 was the fastest processor on the planet in fp. For example,
when I was pricing cheap cpus a while back, my choice was a 533 mhz
21164 vs. a 200 mhz PeniumPro. The 21164 was 3x faster in floating
point, as measured by the SPEC95fp benchmark.

Of course, things change every month.

The major change on the 21264 is that it has relaxed issue rules, so
gcc/g77 will probably be closer to Digital's compilers in performance.

> While some I've talked to said FP
> performace is inherently poor in RISCs,

I would avoid generalizations like this as much as possible. The usual
generalization is the reverse, but Intel continually surprises its
critics by throwing lots of manpower at CPU design.

-- g