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

[Neil Rieck]

On Nov 25, 7:44 am, JF Mezei <jfmezei.spam... at vaxination.ca> wrote:
[...snip...]

>
> One need not forget that the VAX instruction set is far more complex
> than the 8086 or IBM 360/370/390 or whatever it is called this week.
>

Not sure if you are saying this was good or bad. CISC made sense when
memory was expensive but things changed when memory got cheap. Add to
this the fact that the VAX could require something like 200 microcode
cycles executing the POLY instruction then you wonder how they ever
got any work done while serving interrupts. These instructions were
restartable but you could not interrupt them in the middle then pick
up where you left off. Now if the CISC instruction is broken up into
multiple RISC instructions, then an interrupt would let you pick up
you RISC-based-POLY-equivalent where the interrupt happened.

No, while I smile fondly on my PDP and VAX days, I know that RISC and
Alpha where the right moves for the industry. It is too bad upper
management at DEC did not agree (then built VAX-9000)

[...snip...]

>
> Also, one needs to consider manufacturing costs.  Building a Nehalem
> 8086 with quickpath requires a huge investment in sophisticated FAB and
> you need to produce a lot to get payback on investment. The 8086s have
> tons of volume to justify these investments.   IA64 gets "hand me downs"
> since it is one or two generations behind in process, so it can use
> older FABs.  The high development costs coupled with low volumes are not
> a recipe for success for IA64.
>

QPI (CSI) were developed at DEC for Alpha. Had not Curly + Carly
killed off Alpha, the whole industry would have abandoned FSB much
sooner. That said, Intel knows how to make a buck and they stuck QPI
in Core-7 because that is where their volumes AND where they meet AMD
in the market place. With the demise of Alpha, PA-RISC, ROCK, and the
possible demise of SPARC and ULTRA SPARC, the competition against IA64
is lower so they can take their time.

> The big question is whether the Alpha architecture, if it had been
> adopted by Intel in 1999, could have progressed at a faster pace and
> lower development costs than what Intel has been able to achieve with IA64.
>

I agree. Besides, IA64 required new compiler technology which never
really materialized so it was all for nothing. At least DEC had
produced some pretty neat code generators for Alpha.

> But as it stands, IA64 is late to the game and nobody is missing it.
> That says a lot about its lack of importance in the market.

True

NSR