[rescue] Compaq Proliant 8000

[Charles Shannon Hendrix]

Wed, 28 Apr 2004 @ 16:25 -0400, Dave McGuire said:

> On Apr 28, 2004, at 4:01 PM, Joshua Boyd wrote:
> >>  My memory may be suffering from bit rot, but I never saw it marketed
> >>as a general-purpose CPU.  It was generally touted as a "Cray on a
> >>chip".
> >
> >All I know is that supposed historians record it as an attempt at a
> >general purpose chip.
> 
>   Hmm.  In that case, I'll trust my memory of what the Intel literature 
> said. :)

The i860 was not capable of being a general purpose CPU.

It had a funny way of handling interrupts, and an in interrupt usually
flushed several pipelines and caused a reload.  This ate as much as 2000
cycles.  

User programs could see the pipeline in an i860, which is interesting,
but it was nearly impossible to make use of.  Compiler writers didn't
fix the problems and learn to use the i860 features until other RISC
designs had surpassed even the i860's paper performance.

Sound a little familiar?

The i860 was Intel's golden-child.  The hype virtually buried the i960,
which was a much better design and could be a real general purpose
CPU as well.  It had a linear address space, and MMU, and other nice
features.  But Intel marketing had the chip crippled by removing the MMU
and it was marketed as a pure embedded processor.

The i860 was fine as a co-processor though, because you can perfectly
order the software and usually keep it in cache the whole time.  It was
used in some nice graphics and SCSI controllers.  You can also avoid a
lot of need for performance-sucking interrupts.


-- 
shannon "AT" widomaker.com -- ["People should have access to the data which
you have about them.  There should be a process for them to challenge any
inaccuracies." -- Arthur Miller]