[Info-vax] Dave Cutler, Prism, DEC, Microsoft, etc.

[Michael Kraemer]

Neil Rieck schrieb:

>>
>>From the parochial viewpoint of the VMS installed base - what they needed
>>was faster VAXes.  Not Alpha.
>>
> 
> I'm not sure I agree with these points. In the book "DEC is Dead, Long
> Live DEC" it was stated that clock-for-clock comparisons between CISC
> and RISC showed that RISC out performed CISC two-to-one without any
> special tuning. 

This is not generally true.
In 1990 e.g. 68040-based workstations (from HP) were
roughly on par with Mips-based DECstations and somewhat faster
than Sparc's. The 68040 executed almost one instruction per cycle,
just like the contemporary RISCs, and the 68060 follow-up was
even superscalar. The RISCs, however, were easier to crank up
clock speed and that was it.


> Let's also remember that DEC/Compaq sold more Alphas than
> VAX.

We had that one before. How many more Alphas than VAX?
I presume much less than a factor of 2.

> Apparently Intel has sold more Itaniums than VAX + Alpha combined

And how many of them run VMS?

> so the world didn't exactly fall to pieces with the end of VAX.

Effectively VAX ended with the introduction of Alpha, 1992.

> What I really question is the speedup specs from Itanium (EPIC). Many
> have said that the EPIC compilers never lived up to the hype, 

How could they?
They are based on unrealistic assumptions,
i.e. perfectly predictable program execution and
indefinitely parallelizable program code.

> but that
> Itanium is still twice as fast as Alpha. 

No surprise here. Itanics are clocked roughly
a factor two higher than the last Alpha's
and have an obscene amount of cache.
So this isn't exactly an Epic breakthrough
but rather old-fashioned tricks borrowed from
the RISC camp
(HP snake CPUs of 1991 vintage come to mind).