A
Air Raid
http://arstechnica.com/articles/culture/amd-ati.ars/1
AMD + ATI: does it add up?
By Jon "Hannibal" Stokes , Anders Bylund
Monday, June 12, 2006
At the very end of May, Forbes ran a short piece reporting analyst
projections that AMD is looking to acquire graphics chip maker ATI. The
article was essentially informed speculation, based ostensibly on some
analysis of the graphics and processor markets and on the recent news
that AMD is expanding their fab capacity. Nothing in the article
suggested that the speculation had its basis in anonymous sources,
leaks, or rumors, so the piece didn't make much of a splash at the
time.
Last week at Computex, however, Intel allegedly began telling folks
behind closed doors that AMD is planning to acquire ATI. This news came
courtesy of Tweaktown, who cited a "trusted and reliable anonymous
source" for the claim. It wasn't clear from Tweaktown's report if Intel
itself had heard a rumor to this effect, or if the company was reading
the same tea leaves as the RBC Capital Markets analysts in the Forbes
article and coming to the same conclusion. Finally, last Friday brought
the following short note from our old friend Kyle Bennett at HardOCP:
Over dinner tonight in downtown Taipei it was explained to me that
Intel was making the rounds with their customers explaining exactly how
the AMD/ATI merger/acquisition was going to impact their business.
Closure of the deal is expected to pass in 2 weeks.
This looks fairly definitive, but only time will tell if the rumor
turns out to be true. As of this writing, the AMD-ATI
merger/acquisition is still in the realm of informed speculation and
anonymous sources. Nonetheless, that didn't squelch our excitement over
the possible deal here at the Ars Orbiting Headquarters. We think that
an AMD-ATI fusion is a match made in enthusiast heaven, and in this
article we're going to tell you why it would work from both a technical
and financial perspective.
Caught in the Crossfire
So what will it mean if the rumors are true? For one thing, it would
mean that AMD would finally have an in-house source of core logic
chipsets with the kind of features and quality that would better enable
them to compete with Intel in the consumer market.
Before Intel started its "platformization" approach with the Centrino
portable line, the company wasn't really thought of as a chipset maker,
even by many technical people. When people thought "Intel" they thought
"CPU," despite the fact that Intel has long had the lion's share of the
PC chipset market. Intel is a chipset powerhouse, and their ability to
offer a complete CPU + chipset solution with integrated graphics and
networking gives them an edge in the consumer market that AMD currently
cannot match.
AMD's lack of a complete, homegrown chipset solution has long been a
major weak spot, as those of us who've been burned by VIA chipsets can
amply attest. Back before NVIDIA and ATI got into the AMD core logic
chipset market, AMD users were stuck with whatever VIA produced. Even
those who later opted for an AMD-made northbridge still turned to VIA
for the southbridge (I/O hub) chip. Both NVIDIA and ATI eventually
entered the market and produced solid AMD chipsets, with the former
company occupying the higher end of the market and the latter left to
the lower end... at least until recently.
When ATI launched their multi-GPU Crossfire Xpress 3200 chipset earlier
this year, the graphics maker finally had a chipset that could compete
with NVIDIA's Athlon offerings. AMD's acquisition of ATI would bring
that new chipset line under AMD's roof, giving AMD a complete chipset
solution that could set the company on its way to competing with Intel
in quality, price, and features.
It's also important to note that Crossfire just isn't selling well on
Intel platforms. In fact, Intel won't even be supporting Crossfire on
their forthcoming series of enthusiast-oriented desktop chipsets, the
965 series. Right now, if you really want to do Crossfire, then you'll
be doing it with an Athlon FX rig. So if AMD and ATI join forces, ATI
won't really be losing anything in the way of Crossfire market share.
Seeing AMD's coherent HyperTransport licensing plans in a new light
In addition to offering AMD an in-house source of integrated graphics
and I/O hub hardware, an ATI acquisition could make a reality of some
of the more pie-in-the-sky elements of AMD's cHT coprocessor strategy.
I'm talking specifically about AMD's plans to do cHT coprocessor-based
media and physics acceleration.
In my previous reporting on AMD's plans to develop a coprocessor market
based on their coherent HyperTransport (cHT) bus technology, I've
expressed a mix of optimism and skepticism that the company can
successfully leverage cHT in both the server and consumer markets. To
recap that coverage very briefly, AMD has said that they'd like to see
a variety of coprocessors, each dedicated to accelerating a different
type of applications, that can be dropped into an open Socket AM2 or
more specialized cHT socket and function together with an Athlon or
Opteron as a tightly coupled coprocessor.
While I've argued that this plan makes a certain amount of sense in the
server space, where prices are higher and volumes are lower, the math
doesn't seem to work out in the consumer space. In particular, AMD's
aspirations to have someone develop a mass-market GPU that could be
placed in a cHT socket seem pie-in-the-sky when you consider the fact
that PCIe is fast and robust, and most importantly it has volume on its
side. Why would a graphics company that's already making money in the
consumer space specifically develop a GPU specifically for use on a
coherent HyperTransport bus, when the consumer market is full of
motherboards with free PCIe slots and completely absent motherboards
with free cHT sockets? The answer is probably, "they wouldn't."
But what if the market-based arguments against AMD's coprocessor plans
were nullified? This is exactly what would happen if AMD were to
acquire ATI. AMD could more easily the kinds of things they've recently
said that they want to do, and they wouldn't have to depend on the
willingness of third-party chipmakers to take a risk on an as-yet
unproven platform.
Accelerating graphics and audio with cHT
By purchasing ATI outright, AMD could make a GPU that's Socket-AM2
compatible and that shares a pool of DDR2 with the processor. It's true
that such a GPU would lose the advantage of ATI's very fast GDDR3
graphics memory, but the sheer size of the DDR2 pool and the
shared-memory architecture of the system would mitigate some of that
loss. Furthermore, the CPU/GPU integration wouldn't require special
support logic (i.e., would be glueless) and wouldn't have its own
dedicated pool of RAM, the overall cost of an AMD CPU + GPU system
could be lower than more traditional systems with PCIe-based GPUs.
ATI has recently demonstrated GPU-based audio acceleration, so AMD
could add that to their platform as well. With the proper operating
system and driver support, you could do a lot of interesting things by
way of dynamically dividing up the graphics and audio processing chores
among an Athlon FX's two cores and the GPU coprocessor.
In short, an ATI purchase would give AMD a ton of flexibility for
putting together a gaming/enthusiast platform that would offer an
excellent combination of price and performance. Such a platform could
also be attractive to universities and other research venues, where a
rack of low-cost CPU + GPU systems could significantly lower the cost
of doing simulations and other digital signal processing (DSP) type
tasks.
Combining the GPU and PPU to get the best of both worlds
AMD has talked up coprocessor-based physics acceleration, and an ATI
purchase could make this happen as well. Following a similar move by
NVIDIA, ATI has already announced plans to do physics acceleration on
their GPUs, so a GPU coprocessor could double as a physics coprocessor.
Such a cHT-based setup would also eliminate one of the major
disadvantages that GPU-based physics acceleration has when compared to
a dedicated, PCI-based solution like the PhysX PPU from Ageia.
As we've pointed out before here at Ars, what PhysX does and what ATI's
and NVIDIA's GPU-based "physics acceleration" solutions do are fairly
different.
The Nvidia technology is aimed at graphics effects, like particles,
cloth, water, and other second-order visual phenomena that don't affect
gameplay. In other words, when something happens in the game that has
an effect that the CPU doesn't really have to know about because it
won't affect gameplay like a rocket moving the fog out of the way as
it travels, or a cloak crumpling a certain way because it was moved by
a player, or water splashing in a certain pattern-then the GPU can
accelerate that.
However, the GPU does not feed very much data back to the CPU as a
result of its physics calculations, which means that nothing that the
GPU produces in the way of physics simulation can really affect
gameplay. So you couldn't use the GPU to simulate the partial
demolition of a structure that's holding some NPCs, because the CPU
wouldn't know enough about the results. The simulated demolition would
go from the GPU straight to the framebuffer and onto the screen, like
so many water or particle effects, but nobody would tell the actual
game code running on the CPU where the large chunks of debris (fallen
beams, pieces of wall, the inevitable barrels and crates) had fallen
and where the new gap in the wall appeared.
The take-home point here is that a real PPU has to collaborate closely
with the CPU-the two must talk back-and-forth quite a bit because the
physics output of the PPU affects gameplay. In contrast, the more "eye
candy"-oriented physics output from the GPU does not affect gameplay,
so that output goes straight from the GPU to the screen.
With a coprocessor-based GPU/PPU solution, ATI could give AMD's
platform the best of both worlds. A cHT GPU could do the kind of
PhysX-style, gameplay-oriented physics acceleration that requires close
collaboration with the CPU, and it could also do the kind of
visually-oriented physics acceleration that GPUs from ATI and NVIDIA
are currently capable of.
After ATI's Computex demonstration of a three-GPU physics and graphics
acceleration setup, where one GPU is dedicated to physics and two
others to graphics, it's not too hard to imagine how such physics
acceleration would work in a Socket 4x4 system. The GPU in the AM2
socket could host both gameplay-oriented and effects-oriented physics,
while two PCIe-based graphics cards could handle graphics chores.
Crunching the numbers
Click this Mojo Ad
financial report, March 31, AMD had US$2.6 billion of cash and
equivalents on hand, versus US$615 million of long-term debt. That's
less than half the debt the company carried a year ago (US$1.6
billion), and while paying down their loans and building another chip
fab, AMD is still cash flow positive over the last 12 months to the
tune of US$568 million.
In addition, AMD's stock price has been flying high since last fall,
making a stock-swap offer all the more attractive to the shareholders
of any company AMD would like to acquire. The recent price fall is
working against this option, however, making at least some cash likely
necessary to close a deal.
So just to challenge the thesis a bit, how about an AMD purchase of
NVIDIA? With 40.1 percent gross margins, the company has solid pricing
power, and it trickles down to 13.3 percent bottom-line net margins.
That's awfully nice for a hardware company. But the market knows this,
and the stock currently trades for 2.2 times trailing twelve months of
sales. The market cap sits at US$7.5 billion, though the enterprise
value-what AMD would really pay-is "only" US$6.6 billion thanks to
Nvidia's US$900 million of cash-minus-debt. Think of it as buying a
US$50 wallet, knowing that there's a 20-dollar bill inside.
So NVIDIA is a first-class business, but AMD would have to borrow a lot
to make the deal happen, or dilute its own stock down into the
basement. There is no chance of that happening, not when AMD is finally
getting its financial act together. On to ATI.
No Canadian dollar jokes, please; these are all American bucks. At
today's price, you could buy every share of ATI stock for only US$3.9
billion, and its bank account would give you back US$600 million, so we
have a US$3.3 billion price tag here. It's cheap compared to NVIDIA at
only 1.5 times sales, but a bit of that discount seems deserved. ATI's
gross margin (total sales minus the cost of making its products) is
just 27.5 percent, and the company is making a net loss of 2.1 percent
of its full-year sales.
This acquisition, if it comes to pass, will put a hurting on AMD's
margins at least until the combined company can start making money out
of its technological synergies. Total sales, however, gets an immediate
shot in the arm. AMD reported US$6 billion of sales over the last four
quarters, and ATI brought in US$2.3 billion.
In short, a combined cash-and-stock deal looks feasible, and then it
will be up to management to find ways of turning ATI's sales into
profits. And the way to do that is to take advantage of the kinds of
technological synergies mentioned above.
The prospects
AMD has the money in the bank to buy ATI, and they have the fab
capacity coming online to expand their integrated circuit business
beyond CPUs. An ATI purchase would give AMD a solid core logic chipset,
and the unprecedented capability to go head-to-head with Intel at the
level of the whole platform. Indeed, by leveraging cHT, Socket AM2, and
the multipurpose DSP capabilities of an ATI GPU, AMD would have the
ability to put together unique system architectures that would be
tailor-made for the gaming/enthusiast and high performance computing
(HPC) markets.
AMD/ATI would no doubt continue to develop standalone, PCIe-based
graphics cards for the mainstream market, giving them a significant
place in Intel-based desktops. The possibility that in 2007 you could
build a system with an AMD GPU and an Intel CPU makes for an
interesting dynamic, to say the least. It could be that Intel and
NVIDIA will work more closely together to exclude AMD/ATI from Intel
systems, but at this point it's hard to make specific predictions. The
best we can do is to note that anything that's positive for AMD-and
this merger would certainly be positive for them-is unlikely to spell
good news for Intel.
AMD + ATI: does it add up?
By Jon "Hannibal" Stokes , Anders Bylund
Monday, June 12, 2006
At the very end of May, Forbes ran a short piece reporting analyst
projections that AMD is looking to acquire graphics chip maker ATI. The
article was essentially informed speculation, based ostensibly on some
analysis of the graphics and processor markets and on the recent news
that AMD is expanding their fab capacity. Nothing in the article
suggested that the speculation had its basis in anonymous sources,
leaks, or rumors, so the piece didn't make much of a splash at the
time.
Last week at Computex, however, Intel allegedly began telling folks
behind closed doors that AMD is planning to acquire ATI. This news came
courtesy of Tweaktown, who cited a "trusted and reliable anonymous
source" for the claim. It wasn't clear from Tweaktown's report if Intel
itself had heard a rumor to this effect, or if the company was reading
the same tea leaves as the RBC Capital Markets analysts in the Forbes
article and coming to the same conclusion. Finally, last Friday brought
the following short note from our old friend Kyle Bennett at HardOCP:
Over dinner tonight in downtown Taipei it was explained to me that
Intel was making the rounds with their customers explaining exactly how
the AMD/ATI merger/acquisition was going to impact their business.
Closure of the deal is expected to pass in 2 weeks.
This looks fairly definitive, but only time will tell if the rumor
turns out to be true. As of this writing, the AMD-ATI
merger/acquisition is still in the realm of informed speculation and
anonymous sources. Nonetheless, that didn't squelch our excitement over
the possible deal here at the Ars Orbiting Headquarters. We think that
an AMD-ATI fusion is a match made in enthusiast heaven, and in this
article we're going to tell you why it would work from both a technical
and financial perspective.
Caught in the Crossfire
So what will it mean if the rumors are true? For one thing, it would
mean that AMD would finally have an in-house source of core logic
chipsets with the kind of features and quality that would better enable
them to compete with Intel in the consumer market.
Before Intel started its "platformization" approach with the Centrino
portable line, the company wasn't really thought of as a chipset maker,
even by many technical people. When people thought "Intel" they thought
"CPU," despite the fact that Intel has long had the lion's share of the
PC chipset market. Intel is a chipset powerhouse, and their ability to
offer a complete CPU + chipset solution with integrated graphics and
networking gives them an edge in the consumer market that AMD currently
cannot match.
AMD's lack of a complete, homegrown chipset solution has long been a
major weak spot, as those of us who've been burned by VIA chipsets can
amply attest. Back before NVIDIA and ATI got into the AMD core logic
chipset market, AMD users were stuck with whatever VIA produced. Even
those who later opted for an AMD-made northbridge still turned to VIA
for the southbridge (I/O hub) chip. Both NVIDIA and ATI eventually
entered the market and produced solid AMD chipsets, with the former
company occupying the higher end of the market and the latter left to
the lower end... at least until recently.
When ATI launched their multi-GPU Crossfire Xpress 3200 chipset earlier
this year, the graphics maker finally had a chipset that could compete
with NVIDIA's Athlon offerings. AMD's acquisition of ATI would bring
that new chipset line under AMD's roof, giving AMD a complete chipset
solution that could set the company on its way to competing with Intel
in quality, price, and features.
It's also important to note that Crossfire just isn't selling well on
Intel platforms. In fact, Intel won't even be supporting Crossfire on
their forthcoming series of enthusiast-oriented desktop chipsets, the
965 series. Right now, if you really want to do Crossfire, then you'll
be doing it with an Athlon FX rig. So if AMD and ATI join forces, ATI
won't really be losing anything in the way of Crossfire market share.
Seeing AMD's coherent HyperTransport licensing plans in a new light
In addition to offering AMD an in-house source of integrated graphics
and I/O hub hardware, an ATI acquisition could make a reality of some
of the more pie-in-the-sky elements of AMD's cHT coprocessor strategy.
I'm talking specifically about AMD's plans to do cHT coprocessor-based
media and physics acceleration.
In my previous reporting on AMD's plans to develop a coprocessor market
based on their coherent HyperTransport (cHT) bus technology, I've
expressed a mix of optimism and skepticism that the company can
successfully leverage cHT in both the server and consumer markets. To
recap that coverage very briefly, AMD has said that they'd like to see
a variety of coprocessors, each dedicated to accelerating a different
type of applications, that can be dropped into an open Socket AM2 or
more specialized cHT socket and function together with an Athlon or
Opteron as a tightly coupled coprocessor.
While I've argued that this plan makes a certain amount of sense in the
server space, where prices are higher and volumes are lower, the math
doesn't seem to work out in the consumer space. In particular, AMD's
aspirations to have someone develop a mass-market GPU that could be
placed in a cHT socket seem pie-in-the-sky when you consider the fact
that PCIe is fast and robust, and most importantly it has volume on its
side. Why would a graphics company that's already making money in the
consumer space specifically develop a GPU specifically for use on a
coherent HyperTransport bus, when the consumer market is full of
motherboards with free PCIe slots and completely absent motherboards
with free cHT sockets? The answer is probably, "they wouldn't."
But what if the market-based arguments against AMD's coprocessor plans
were nullified? This is exactly what would happen if AMD were to
acquire ATI. AMD could more easily the kinds of things they've recently
said that they want to do, and they wouldn't have to depend on the
willingness of third-party chipmakers to take a risk on an as-yet
unproven platform.
Accelerating graphics and audio with cHT
By purchasing ATI outright, AMD could make a GPU that's Socket-AM2
compatible and that shares a pool of DDR2 with the processor. It's true
that such a GPU would lose the advantage of ATI's very fast GDDR3
graphics memory, but the sheer size of the DDR2 pool and the
shared-memory architecture of the system would mitigate some of that
loss. Furthermore, the CPU/GPU integration wouldn't require special
support logic (i.e., would be glueless) and wouldn't have its own
dedicated pool of RAM, the overall cost of an AMD CPU + GPU system
could be lower than more traditional systems with PCIe-based GPUs.
ATI has recently demonstrated GPU-based audio acceleration, so AMD
could add that to their platform as well. With the proper operating
system and driver support, you could do a lot of interesting things by
way of dynamically dividing up the graphics and audio processing chores
among an Athlon FX's two cores and the GPU coprocessor.
In short, an ATI purchase would give AMD a ton of flexibility for
putting together a gaming/enthusiast platform that would offer an
excellent combination of price and performance. Such a platform could
also be attractive to universities and other research venues, where a
rack of low-cost CPU + GPU systems could significantly lower the cost
of doing simulations and other digital signal processing (DSP) type
tasks.
Combining the GPU and PPU to get the best of both worlds
AMD has talked up coprocessor-based physics acceleration, and an ATI
purchase could make this happen as well. Following a similar move by
NVIDIA, ATI has already announced plans to do physics acceleration on
their GPUs, so a GPU coprocessor could double as a physics coprocessor.
Such a cHT-based setup would also eliminate one of the major
disadvantages that GPU-based physics acceleration has when compared to
a dedicated, PCI-based solution like the PhysX PPU from Ageia.
As we've pointed out before here at Ars, what PhysX does and what ATI's
and NVIDIA's GPU-based "physics acceleration" solutions do are fairly
different.
The Nvidia technology is aimed at graphics effects, like particles,
cloth, water, and other second-order visual phenomena that don't affect
gameplay. In other words, when something happens in the game that has
an effect that the CPU doesn't really have to know about because it
won't affect gameplay like a rocket moving the fog out of the way as
it travels, or a cloak crumpling a certain way because it was moved by
a player, or water splashing in a certain pattern-then the GPU can
accelerate that.
However, the GPU does not feed very much data back to the CPU as a
result of its physics calculations, which means that nothing that the
GPU produces in the way of physics simulation can really affect
gameplay. So you couldn't use the GPU to simulate the partial
demolition of a structure that's holding some NPCs, because the CPU
wouldn't know enough about the results. The simulated demolition would
go from the GPU straight to the framebuffer and onto the screen, like
so many water or particle effects, but nobody would tell the actual
game code running on the CPU where the large chunks of debris (fallen
beams, pieces of wall, the inevitable barrels and crates) had fallen
and where the new gap in the wall appeared.
The take-home point here is that a real PPU has to collaborate closely
with the CPU-the two must talk back-and-forth quite a bit because the
physics output of the PPU affects gameplay. In contrast, the more "eye
candy"-oriented physics output from the GPU does not affect gameplay,
so that output goes straight from the GPU to the screen.
With a coprocessor-based GPU/PPU solution, ATI could give AMD's
platform the best of both worlds. A cHT GPU could do the kind of
PhysX-style, gameplay-oriented physics acceleration that requires close
collaboration with the CPU, and it could also do the kind of
visually-oriented physics acceleration that GPUs from ATI and NVIDIA
are currently capable of.
After ATI's Computex demonstration of a three-GPU physics and graphics
acceleration setup, where one GPU is dedicated to physics and two
others to graphics, it's not too hard to imagine how such physics
acceleration would work in a Socket 4x4 system. The GPU in the AM2
socket could host both gameplay-oriented and effects-oriented physics,
while two PCIe-based graphics cards could handle graphics chores.
Crunching the numbers
Click this Mojo Ad
Canadian graphics and chipset maker into its arsenal. As of the latest
financial report, March 31, AMD had US$2.6 billion of cash and
equivalents on hand, versus US$615 million of long-term debt. That's
less than half the debt the company carried a year ago (US$1.6
billion), and while paying down their loans and building another chip
fab, AMD is still cash flow positive over the last 12 months to the
tune of US$568 million.
In addition, AMD's stock price has been flying high since last fall,
making a stock-swap offer all the more attractive to the shareholders
of any company AMD would like to acquire. The recent price fall is
working against this option, however, making at least some cash likely
necessary to close a deal.
So just to challenge the thesis a bit, how about an AMD purchase of
NVIDIA? With 40.1 percent gross margins, the company has solid pricing
power, and it trickles down to 13.3 percent bottom-line net margins.
That's awfully nice for a hardware company. But the market knows this,
and the stock currently trades for 2.2 times trailing twelve months of
sales. The market cap sits at US$7.5 billion, though the enterprise
value-what AMD would really pay-is "only" US$6.6 billion thanks to
Nvidia's US$900 million of cash-minus-debt. Think of it as buying a
US$50 wallet, knowing that there's a 20-dollar bill inside.
So NVIDIA is a first-class business, but AMD would have to borrow a lot
to make the deal happen, or dilute its own stock down into the
basement. There is no chance of that happening, not when AMD is finally
getting its financial act together. On to ATI.
No Canadian dollar jokes, please; these are all American bucks. At
today's price, you could buy every share of ATI stock for only US$3.9
billion, and its bank account would give you back US$600 million, so we
have a US$3.3 billion price tag here. It's cheap compared to NVIDIA at
only 1.5 times sales, but a bit of that discount seems deserved. ATI's
gross margin (total sales minus the cost of making its products) is
just 27.5 percent, and the company is making a net loss of 2.1 percent
of its full-year sales.
This acquisition, if it comes to pass, will put a hurting on AMD's
margins at least until the combined company can start making money out
of its technological synergies. Total sales, however, gets an immediate
shot in the arm. AMD reported US$6 billion of sales over the last four
quarters, and ATI brought in US$2.3 billion.
In short, a combined cash-and-stock deal looks feasible, and then it
will be up to management to find ways of turning ATI's sales into
profits. And the way to do that is to take advantage of the kinds of
technological synergies mentioned above.
The prospects
AMD has the money in the bank to buy ATI, and they have the fab
capacity coming online to expand their integrated circuit business
beyond CPUs. An ATI purchase would give AMD a solid core logic chipset,
and the unprecedented capability to go head-to-head with Intel at the
level of the whole platform. Indeed, by leveraging cHT, Socket AM2, and
the multipurpose DSP capabilities of an ATI GPU, AMD would have the
ability to put together unique system architectures that would be
tailor-made for the gaming/enthusiast and high performance computing
(HPC) markets.
AMD/ATI would no doubt continue to develop standalone, PCIe-based
graphics cards for the mainstream market, giving them a significant
place in Intel-based desktops. The possibility that in 2007 you could
build a system with an AMD GPU and an Intel CPU makes for an
interesting dynamic, to say the least. It could be that Intel and
NVIDIA will work more closely together to exclude AMD/ATI from Intel
systems, but at this point it's hard to make specific predictions. The
best we can do is to note that anything that's positive for AMD-and
this merger would certainly be positive for them-is unlikely to spell
good news for Intel.