By: Richard Cownie (tich.delete@this.pobox.com), January 30, 2013 8:34 am
Room: Moderated Discussions
David Kanter (dkanter.delete@this.realworldtech.com) on January 29, 2013 11:51 pm wrote:
> Honestly, the trend is away from standard ARM cores. Also, there's a lot of other IP that servers need
> that isn't in the normal tablet/phone space, like high performance wired networking and PCI-E.
Sure, there's work to do. Just as there was for Intel in the 1990s. But when
the economic incentive is there - and I think Intel's margins on server cpu's are
a pretty fat target - then many companies will do what it takes to solve those
problems.
This is exactly what Intel went through in the RISC vs x86 battle. x86 didn't have
decent server software; but once they had good enough cheap hardware, Linux and
WindowsNT came along to fix that. Pentium didn't have a sensible MP bus;
Intel spent the money to develop the P6 bus for cheap/easy 4-way servers.
x86 didn't have RAS features; Intel added them. Intel didn't have cores as good as
DEC or HP; they hired good designers, borrowed a few ideas, and got the much
more competitive P6.
To make the situation even more worrying for Intel, it seems that PC sales - the
source of their economic power - are now falling (8% YoY in the last Q IIRC)
while the ARM business is booming. They may not be able to maintain a huge lead
in core design and manufacturing capability forever if their core business
declines (to be clear, I'm talking about what might happen 4 or 8 years from now,
not some imminent crash).
What happens in the long term depends on economic forces (high volume tends to win)
and on well-defended technical or legal obstacles (e.g. patents, software network
effects). If x86 PC sales shrink significantly, while ARM-based products continue
to boom, I'm not sure how Intel will maintain dominance.
> The right way to think about this is to compare Intel's margins to TSMC+SoC vendor. I think
> TSMC gross margins are somewhere around 50%. SoC vendors get away with 20-30% gross margins,
> but only on super high volume products. For a lower volume product, they probably want a lot
> higher than that. Also, you need to pay for some of the 3rd party IP, e.g. ARM licenses.
>
> That's starting to look like pretty fat margins.
>
> > I don't believe that Intel is selling server cpu's at the marginal
> > cost of production. Do you ?
You have to compare it to Intel's margins. IIRC their overall gross margin is
around 60%. But what we really need is an estimate of their margin on $300-500
server cpu's. My suspicion would be that those cost something like $50 to produce,
so have something like 500% margin, or higher. Which I think leaves plenty of
space for a leaner competitor to build a competitive part with $50 cost, pay the
foundry $80 or so, and sell it for $150. But you probably have better numbers
than I do.
> Honestly, the trend is away from standard ARM cores. Also, there's a lot of other IP that servers need
> that isn't in the normal tablet/phone space, like high performance wired networking and PCI-E.
Sure, there's work to do. Just as there was for Intel in the 1990s. But when
the economic incentive is there - and I think Intel's margins on server cpu's are
a pretty fat target - then many companies will do what it takes to solve those
problems.
This is exactly what Intel went through in the RISC vs x86 battle. x86 didn't have
decent server software; but once they had good enough cheap hardware, Linux and
WindowsNT came along to fix that. Pentium didn't have a sensible MP bus;
Intel spent the money to develop the P6 bus for cheap/easy 4-way servers.
x86 didn't have RAS features; Intel added them. Intel didn't have cores as good as
DEC or HP; they hired good designers, borrowed a few ideas, and got the much
more competitive P6.
To make the situation even more worrying for Intel, it seems that PC sales - the
source of their economic power - are now falling (8% YoY in the last Q IIRC)
while the ARM business is booming. They may not be able to maintain a huge lead
in core design and manufacturing capability forever if their core business
declines (to be clear, I'm talking about what might happen 4 or 8 years from now,
not some imminent crash).
What happens in the long term depends on economic forces (high volume tends to win)
and on well-defended technical or legal obstacles (e.g. patents, software network
effects). If x86 PC sales shrink significantly, while ARM-based products continue
to boom, I'm not sure how Intel will maintain dominance.
> The right way to think about this is to compare Intel's margins to TSMC+SoC vendor. I think
> TSMC gross margins are somewhere around 50%. SoC vendors get away with 20-30% gross margins,
> but only on super high volume products. For a lower volume product, they probably want a lot
> higher than that. Also, you need to pay for some of the 3rd party IP, e.g. ARM licenses.
>
> That's starting to look like pretty fat margins.
>
> > I don't believe that Intel is selling server cpu's at the marginal
> > cost of production. Do you ?
You have to compare it to Intel's margins. IIRC their overall gross margin is
around 60%. But what we really need is an estimate of their margin on $300-500
server cpu's. My suspicion would be that those cost something like $50 to produce,
so have something like 500% margin, or higher. Which I think leaves plenty of
space for a leaner competitor to build a competitive part with $50 cost, pay the
foundry $80 or so, and sell it for $150. But you probably have better numbers
than I do.