someone (someone@somewhere.com) on 2/7/09 wrote:
>
>The scalability stuff in Tukwila takes up a significant
>portion of the die area and power budget. Putting even
>a fraction of that overhead in the same die that Intel
>sells as desktop and laptop processors is just plain
>nuts.

I agree. In fact, especially with four cores, I see no
point in doing any cache-coherent QPI at all in that
market. You can do a fine workstation with a single
chip.

So what matters is:

>What about Beckton (Nehalem generation Xeon MP)?
>Well the fact is x86 server sales of 4 sockets and up is
>only worth about $1B per quarter, has flat sales growth
>and AMD takes a bigger share of it than any other x86
>segment.

.. but that's kind of a self-fulfilling prophecy.

Of course x86 stays in the quad-socket (and lower)
area, as long as nobody drives it any higher. AMD didn't
really scale past that (yes, maybe that changes to 8 soon
enough). And the very reason that AMD had a bigger pie
of that market was that Intel scaled even less due to the
whole shared bus.

Now, I suspect that Intel sees Beckton as a "Niagara
killer" rather than a POWER6 competitor, and that would
certainly explain focusing on 1-4 sockets (with duals
probably being the bulk, since that just makes everything
cheap and simple for some random box that is going to
have tons of CPU power anyway).

>Putting the scalability stuff in Beckton would
>probably cost it a couple of cores and maybe a speed
>bin. Beckton will compete with future AMD processors
>far more than IPF and extra cores and clock speed are
>more valuable than scaling well to beyond 4 sockets
>(which AMD server processors don't).

I really wish people saw x86 as competition to POWER.

It obviously takes more than Intel for that to happen:
they'd need to get somebody like HP on board, because
that market really wants more than just the hardware.

But look at it this way: if you're trying to convince
somebody to stray from the POWER fold, what the hell has
IPF to offer? Absolutely nothing. This whole argument that
IPF is competing against POWER seems broken to me. If it
is, it is losing.

POWER has better performance, a better infrastructure, and
a vendor who doesn't screw its customers over like clock-
work every decade.

So if you want to compete against that, you need to compete
on something. IPF doesn't have it. Performance? No. Better
infrastructure? It's now finally catching up. And an
architecture that everybody will have to worry might go
the way of PA-RISC and Alpha in another five years, and
continuing the whole HP/Compaq history.

In contrast, if HP competed with x86, they'd have at least
a couple of arguments on their side. Performance? Check.
An architecture you don't need to even wonder whether it
will stay around? Check. An architecture that people are
used to, and scales from your company laptops all the way
up? Check.

>That's why IPF QPI gets distributed directory and x86 QPI
>doesn't. Sorry, no craziness or conspiracy theory here.
>It's just Intel doing what the server market wants. :-)

Oh, I'm sure beckton will sell like hotcakes. The people
who buy Niagara will all love it. There's no argument that
the bulk of servers is all at the low end.

But if people really want to compete against POWER, I don't
think IPF is the best answer.

Linus