Playing devlis advocate here...
AMD's primary business was low profit to them most of the time (but last 2
years)
Prop up one low-profit business with a second low-profit business?
As long AMD play it nice, NVidia has not much space left. Intel is known for
screwing their chipset (and other) partners more that once. Telling f**k off
to AMD is not their interest.
I don't see them directly telling nVidia to screw off, but if they use
this new combined AMD-ATI to try and gain a competitive advantage in
the PC graphics market, then they are basically telling nVidia to
screw off. If they aren't doing this to gain a competitive advantage
then what was the point in the first place?
AMD/ATI combined is stronger here than ATI alone. Plus AMD has varius IP
cross-licesing agreements with Intel.
AMD has some cross-licensing agreements, but none of them are likely
to cover video chipsets. Besides it's not a matter of being legally
allowed to sell the chips that is the worry, it's about getting
pre-release info and help. Right now ATI and nVidia both get access
to Intel's chipsets LONG before they are released so that they can
develop video cards that will work with these chipsets. If Intel
stops providing these early chipsets and support in getting the cards
to work with them, ATI could be left at a serious dissadvantage to
nVidia.
It's not so easy. And AMD has much more money for lawyers than ATI alone
(and past performance indicates that AMD is willing to use that)
AMD is explicitly forbidden from using Intel's processor bus
technologies. This is part of a long-standing agreement dating back
to the early 90's in an effort to prevent AMD from selling processors
that will work in the same motherboards as Intel chips. It isn't much
of a stretch at all to think this could apply to chipsets as well as
processors.
Right now ATI has a license that grants them the right to build
chipsets for Intel processors, but that license is definitely going to
be full of limitations. Intel HAS pulled companies licenses in the
past. Serverworks is a prime example here, Intel all but terminated
their license after they were bought out by Broadcom. I *FULLY*
expect to see history repeat itself here, and like it or not, Intel
has every legal right to do so.
Well, maybe they foresee the change of focus on the market. Look at this --
CPU's are less & less important for PC's perfromance. With stuff like
physics coprocessors enetering arena importance of CPU as key performance
component even decreases.
Independant physics co-processors are a lost cause. If the technology
proves useful (somewhat questionable) then they'll get integrated into
a CPU. Having separate chips to handle these sorts of math things has
proven to be a bad idea.
Video, on the other hand, is a different story. Integrating video
onto the CPU has proven to be excrutiatingly difficult. The problems
are two-fold: first, GPUs have a LOT of transistors, even more than
CPUs. Second, GPUs needs HUGE amounts of memory bandwidth while CPUs
need LOTS of memory and flexible memory configurations. While a GPU
can get by very nicely with 512MB of memory soldered onto a board,
that just isn't an option for a CPU. A server might need 64GB of
memory, while a desktop might only need 1GB. Both scenarios would
benefit little from the huge bandwidth offered but would suffer very
badly from the increased cost of the faster memroy.
In Austria, in the first half of XX centurey, there was a company which kept
allmost total monopoly in a production of horse wagons. They even had
various govement aids like high import taxes for foreign products. They were
so big that they had their own iroworks producing only for them. Then
50-ties came, and all was kaput. The market has vanished. They don't exist
anymore, of course.
CPUs is a business which made both Intel and AMD significant. But will it be
able to keep those companies up in the future (with all their R&D costs and
expenses)?
It's a business that has huge R&D and capital (if you own fabs) costs,
combined with low profit margins. Not an easy business to succeed in.
The end result is that we're left with little more than companies that
focus ONLY on building CPUs. Look at companies like Hitachi, TI,
Motorola, Digital/Compaq/HP, etc. etc. All used to build high-end
CPUs but they either got out of the market or spun that division off
on it's own. Sun and Fujitsu are still struggling at it, but mostly
failing. The only exception to the above is IBM, who are the
exception to most rules in the computer world.
Thats allost a given.
The main upsides are:
* Ability to create Centrino counterpart
That requires marketing much more than any technology, and marketing
is an area that neither AMD or ATI are hugely strong at. Both are
out-marketted by their main rivals (Intel and nVidia respectively).
* Better ability to play in commoditised market
Commoditized for even less profit?
* Better ability to play in the middle of the market where bread&butter of
the desktop PC is -- high performance integrated AMD/ATI solutions for stuff
like media center PCs, with quite good playability of games and stuff.
Embedded graphics on coherent HyperTransport link might enable peroformance
unseen in embedded arena.
Embedded graphics performance is 99% memory bandwidth, 1% everything
else. Unless you plan on dropping memory on the motherboard to
connect to your video card, you're actually better off with the
external memory controller as Intel does things. Ohh, and ATI already
tried an AMD Hypertransport compatible chipset with memory on the
system board... it was a miserable failure due to costs (motherboards
are the one part of the PC where profit margins are worse than CPUs
and GPUs).
* AMD's one of the few companies in the world which have apropriate inhouse
know-how as well as state-of-the-art software&hardware for high peroformance
IC design & development. If used properly that ability could possibly
translate into significant improvement in combined company GPU designs.
I'll grant this as the one real advantage of the whole deal, combined
with AMD having their own fabs.
* Combined company has finally whole platform in their hands -- look how
long it took to have decent chipsets for K7/K8 platforms. VIA & their
chipses which for a few years (since 586B southbridge) f***ing up data in
multi harddrive systems without even acknowledging the problem (only
releasing driver updates which never fully got rid of it) can't be takes
seriously. Now when AMD want's to change something they can just do it and
the chipsets will be there (as they'll make their own).
Honestly I think that this "advantage" is doesn't really exist. When
the Athlon64 was released there were plenty of chipsets immediately
available. VIA was still screwing the pooch as usual, but nVidia was
there right from the get-go. AMD has done their own chipsets in the
past and they've proven to be inferior to nVidia's solutions for the
most part. This goes to show that just because AMD and ATI would be
one company, they aren't necessarily going to be any better at making
chipsets.
Besides, with AMD's current processors the real magic in chipsets is
in the "extra" stuff, ie PCI-Express, network chips, SATA, audio, etc.
The memory hangs off the processor so it's out of the equation and the
processor connects to the rest of the system by Hypertransport, which
is an open standard and relatively easy to implement. What this means
is that having knowledge of the CPU doesn't really buy you much of
anything when building the chiset and vice versa.
