[NTLUG:Discuss] Alphaserver 2100 -- Digital +3y, AMD v. IBM (Intel irrelevant), lesser chips

[Bryan J. Smith]

Justin M. Forbes wrote:  
> This is all completely incorrect.  Power 4 has always had 32bit
> instructions,

I know.  But it is not 100% compatible with the 32-bit PowerPC ISA.
That's what IBM was trying to develop with their original PowerPC 620
long ago -- a chip that was both 64/32-bit Power and 32-bit PowerPC
compatible.

> These are not 32bit compatibility instructions, AltiVec is actually
> a couple of 128bit registers, a vector execution unit, and a few new
> instructions.

>From what I understand, the PowerPC 970 adds more than just AltiVec
to Power4.  That gives it it's 32-bit PowerPC ISA compatibility.

> US IV is simply dual core US III, and the US V team got the axe
> just after tape out I believe.

Yeah, that's what I heard too.

> They are selling Opteron based servers now, and supposedly are
> working with Serverworks on an 8 way chipset, for general use, and
> a 16-32 way chipset to which Sun would have exclusive rights for a
> certain period of time (unknown how long).

Hmm, sounds like Sun is the HP of the Opteron world then?  (does HP
not have exclusive rights on 16-32 way IA-64 systems?).

>  Opteron has its own interconnects up to an 8 way, but to go beyond
> that, you need a good high speed interconnect, and I think the
> Sunfire technoligy is being looked at there.

Well, you certainly have to "get interesting" with the Opteron.  The
Opteron 800 has only 3 HyperTransport links, so it is not ideal to
go directly HT-to-HT after 4 or 8 CPUs.

> The other option is partnering with Fujitsu whose current Sparc64 V
> competes quite well with Power 4, much faster than current Ultrasparc.
> ... Fujitsu's chips are solid, if not commodity.

I know.  I've used their 8+ way systems before (I'm quoted in their 2001
PrimePower marketing materials ;-).

> AMD will have dual core Opterons next year, and they
> will supposedly work in any Socket 940 motherboard you buy now.

That's the cool thing about HyperTransport!  It's so modular!

I'd like to get a dual Socket-940 mainboard, and then opt for
two dual-cores when they come down in price later on.

> Power is actually less efficient than Intel in power consumption per
> transistor, even at 90nm.

"Per transistor" is the key point.  When you have twice the number of
transistors because you have a variable opcode of 8-144 bytes, and other
x86 compatibility details, then Power comes out on top.

Thanx for the  info/update.


-- 
Bryan J. Smith, E.I. -- b.j.smith at ieee.org