By: bakaneko (nyan.delete@this.hyan.wan), January 29, 2013 9:47 am
Room: Moderated Discussions
Richard Cownie (tich.delete@this.pobox.com) on January 29, 2013 7:19 am wrote:
> bakaneko (nyan.delete@this.hyan.wan) on January 29, 2013 6:49 am wrote:
>
> > Selling server chips with extremely low margins
> > sounds dangerous. Qualcomm needs to be able to
> > survive bad times. And the more Qualcomm nibbles
> > at Intels high-margin business (which also
> > exist to protect itself in exactly these bad
> > times), the higher they have to aim for higher
> > margins as they take over Intel's risks. (Which
> > loopsided means that Intel would die faster,
> > because they have to sell for higher prices.)
> >
> > Qualcomm may very well want you to sell servers
> > at lower margins, but in the end they have to do
> > the same as Intel.
>
> Let's do a global replace:
>
> "Selling server chips with extremely low margins
> sounds dangerous. [Intel] needs to be able to survive bad times.
> And the more [Intel] nibbles at DEC/HP/Sun/IBM's high-margin
> business (which also exists to protect itself in exactly these
> bad times), the higher they have to aim for higher margins as they
> take over DEC/HP/Sun/IBM's risks"
>
> That argument didn't work in 1995-2005. It won't work now.
> Intel was able to sell server chips relatively cheap because
> they had massive desktop/laptop volume and could exploit the
> associated massive investment in R&D and manufacturing to
> attack the small, but lucrative, server cpu business.
Ah yes, I remember that time. The internet was lively, my
mom draw lolcats and it was hard to explain to daddy why he
should set his facebook to private.
Everyone was on it with a flatrate, but there was a dark
side to all of it: The evil RISC empire wanted too much
money for all the millions of servers and desktops, coercing
everyone into submission. Intel, which was in the market for
CISC microcontrollers for two decades came to the rescue.
[TODO: earthshattering final battle]
(Sadly the first internet got destroyed and was erased from
history books and the hive mind because of all the
embarrassing group behaviour. But we will never forget you
Intel, may you rest in piece. *kira*)
Other dumb comparisons?
Also Sparta won, IBM should have won, too.
> What's happening now is that the desktop/laptop business is
> not looking so great. The biggest volumes are in cellphones
> and tablets, and those high volumes are driving huge investments
> in both (ARM) core design, and manufacturing. And those people
> are itching to get some higher-margin business - and server
> cpu's look like a tempting target.
All they would do with your plan is to reinvent the wheel.
Just a waste of time and money. And your plan isn't even
looking beyond Intel. It's just "we can do what Intel can
do, too (and 10% better after reading the shiny brochure).".
Which is just dumb and never worked.
I'm not saying they can't do it, but that it is dumb.
Worse, they put all that hard work into it, just to make
less money and still run after Intel. And when reality hits,
they will look pretty dumb.
Just because it worked in 1995, doesn't mean it will work
now.
> > Which brings the question really down to questions
> > besides money. Why go with Qualcomm? What is so
> > much better?
>
> I expect ARM-based servers will be attractive first for those
> wanting maximum throughput-per-rack and maximum throughput-per-watt.
> Datacenter space is expensive; power is expensive; cooling is expensive.
> The people who know how to pack a quad-core SoC into a cellphone
> have expertise very relevant to the problem of packing lots of cores
> into a rack, and being able to run it cheaply.
But this is silly. You either need better/more servers or
you don't. The only thing you could poke Intel with is that
maybe their CPUs could have faster reaction times for certain
events. But I never measured it, so I wouldn't know.
> Of course Intel is grappling with the same problem - I just don't
> know if they'll be willing and able to mix it up with lean and hungry
> competitors who've been living and breathing throughput-per-watt
> for 20+ years.
ARM never breath throughput per watt. They breath low power
microcontrollers.
> bakaneko (nyan.delete@this.hyan.wan) on January 29, 2013 6:49 am wrote:
>
> > Selling server chips with extremely low margins
> > sounds dangerous. Qualcomm needs to be able to
> > survive bad times. And the more Qualcomm nibbles
> > at Intels high-margin business (which also
> > exist to protect itself in exactly these bad
> > times), the higher they have to aim for higher
> > margins as they take over Intel's risks. (Which
> > loopsided means that Intel would die faster,
> > because they have to sell for higher prices.)
> >
> > Qualcomm may very well want you to sell servers
> > at lower margins, but in the end they have to do
> > the same as Intel.
>
> Let's do a global replace:
>
> "Selling server chips with extremely low margins
> sounds dangerous. [Intel] needs to be able to survive bad times.
> And the more [Intel] nibbles at DEC/HP/Sun/IBM's high-margin
> business (which also exists to protect itself in exactly these
> bad times), the higher they have to aim for higher margins as they
> take over DEC/HP/Sun/IBM's risks"
>
> That argument didn't work in 1995-2005. It won't work now.
> Intel was able to sell server chips relatively cheap because
> they had massive desktop/laptop volume and could exploit the
> associated massive investment in R&D and manufacturing to
> attack the small, but lucrative, server cpu business.
Ah yes, I remember that time. The internet was lively, my
mom draw lolcats and it was hard to explain to daddy why he
should set his facebook to private.
Everyone was on it with a flatrate, but there was a dark
side to all of it: The evil RISC empire wanted too much
money for all the millions of servers and desktops, coercing
everyone into submission. Intel, which was in the market for
CISC microcontrollers for two decades came to the rescue.
[TODO: earthshattering final battle]
(Sadly the first internet got destroyed and was erased from
history books and the hive mind because of all the
embarrassing group behaviour. But we will never forget you
Intel, may you rest in piece. *kira*)
Other dumb comparisons?
Also Sparta won, IBM should have won, too.
> What's happening now is that the desktop/laptop business is
> not looking so great. The biggest volumes are in cellphones
> and tablets, and those high volumes are driving huge investments
> in both (ARM) core design, and manufacturing. And those people
> are itching to get some higher-margin business - and server
> cpu's look like a tempting target.
All they would do with your plan is to reinvent the wheel.
Just a waste of time and money. And your plan isn't even
looking beyond Intel. It's just "we can do what Intel can
do, too (and 10% better after reading the shiny brochure).".
Which is just dumb and never worked.
I'm not saying they can't do it, but that it is dumb.
Worse, they put all that hard work into it, just to make
less money and still run after Intel. And when reality hits,
they will look pretty dumb.
Just because it worked in 1995, doesn't mean it will work
now.
> > Which brings the question really down to questions
> > besides money. Why go with Qualcomm? What is so
> > much better?
>
> I expect ARM-based servers will be attractive first for those
> wanting maximum throughput-per-rack and maximum throughput-per-watt.
> Datacenter space is expensive; power is expensive; cooling is expensive.
> The people who know how to pack a quad-core SoC into a cellphone
> have expertise very relevant to the problem of packing lots of cores
> into a rack, and being able to run it cheaply.
But this is silly. You either need better/more servers or
you don't. The only thing you could poke Intel with is that
maybe their CPUs could have faster reaction times for certain
events. But I never measured it, so I wouldn't know.
> Of course Intel is grappling with the same problem - I just don't
> know if they'll be willing and able to mix it up with lean and hungry
> competitors who've been living and breathing throughput-per-watt
> for 20+ years.
ARM never breath throughput per watt. They breath low power
microcontrollers.