PDA

View Full Version : Q&A with AMD’s Rick Bergman on the graphics sweet spot


NV55
November 30th 08, 11:35 AM
Q&A with AMD’s Rick Bergman on the graphics sweet spot

Dean Takahashi | November 26th, 2008

Rick Bergman is the senior vice president and general manager of the
graphics products group at chip company Advanced Micro Devices. His
team of graphics chip designers –- formerly known as ATI Technologies
until AMD acquired the company in 2006 — has scored well with the
recent launch of the ATI Radeon HD 4800 series family. The so-called
“sweet spot” strategy caught rival Nvidia flat-footed. While Nvidia
created a high-end graphics chip that burned a lot of power, AMD
created a smaller chip that targeted the middle of the market. It
decided that it could put two such chips into a board to create a low-
cost, top-performance solution for high-end gamers. Meanwhile, the
single chips could be easily targeted toward the mid-range and low-end
of the markets. Nvidia, by contrast, has had to wait longer to
redesign its chips for the mid-range and low-end markets. The result
has been a big shift in market share from Nvidia to AMD, whose
graphics chip division is now profitable after a tough 2007. I spoke
with him at AMD’s recent analyst meeting at its headquarters in
Sunnyvale, Calif., after Bergman gave an upbeat outlook for AMD’s
graphics business.

VB: How long ago did ATI started thinking about its “sweet spot”
strategy?

RB: It was three or four years ago. We recognized we could not
continue on with huge die sizes. It was before the merger with
Advanced Micro Devices. We were still working on a chip called the
R600. We were thinking of what we could do for the R700. We decided we
couldn’t do another chip that was so big because of power consumption.
You can just plot how die sizes have grown generation after
generation.

VB: Did it take graphics chip designers longer than microprocessor
designers to realize that power was going to be a problem?

RB: The microprocessor designers ran into it first because they were
pushing frequency. When you do that, it pushes the circuits to the
limit and consumes a lot of power. We started from a lower frequency
base. Graphics processing units (GPUs) started requiring more and more
performance to do better 3-D images. We weren’t able to lower voltage,
and so it was clear we were going to run into the fundamental laws of
physics.

VB: You looked at this problem from a different perspective when you
were an underdog?

RB: About three years ago, we were around 50 percent plus market share
position. We were not in a tailspin. We were looking at where the
problems were. Where the growth was in the market. We had a chip
called the R300. That was the first 256-bit GPU. It let us get the
crown. It cascaded into multiple years of success for ATI. As we made
this decision, we had chips like the R580 which were the best in the
industry.

VB: How did the strategy gel?

RB: It wasn’t unanimous that we were headed down the right path. There
were many painful discussions. It started to become clear when we
taped out (finished the design of the chip). We knew our chip size,
performance and power. We felt like we had a winner.

VB: The merger between AMD and ATI came in the middle of all this in
2006. Did that stretch out the execution on your strategy?

RB: The acquisition was a little bit of a distraction. But our teams
stayed as focused as we could.

VB: There was a time when you were on the defensive.

RB: Yes, 2007 was a tough year for us. We were losing market share.
Our products weren’t as competitive as we would have preferred.

VB: Has the result played out as you expected? Nvidia shot for the
high end and you went for the sweet spot.

RB: They followed their script. We came in underneath with a superior
solution.

VB: Fusion — the plan to combine a graphics chip and a microprocessor
that was the impetus for the AMD-ATI merger — was being planned
alongside this sweet spot strategy. Now the first Fusion chip has been
delayed again until 2011.

RB: We’re trying to broaden the definition of Fusion. We probably made
an error two years ago when we said that Fusion is a combination of
the GPU and the CPU (central processing unit) on the same chip. That’s
just an implementation question. It may not make sense as we look for
more and more diverse solutions of the GPU and CPU working together.
There were challenges in doing it. When you start looking at more
advanced manufacturing technologies, where you can shrink the size of
each component more, it becomes easier to integrate them together.

VB: Will the 2011 product target a certain market?

RB: We are going after mainstream clients and a Netbook market. That’s
new for our strategy.

VB: In the meantime the sweet spot strategy will continue?

RB: They are independent. We will cascade the technology down through
the low end of our product line. The sweet spot strategy is designed
so that we can do that much more quickly, rather than wait through
long cycles of a redesign.

VB: ATI has always aimed to be more aggressive at adopting new
manufacturing technologies, which can give a design more performance
and other benefits. Nvidia has been more cautious. It argues that you
don’t want to try a new design and a new process at the same time.

RB: For the last six process advances at Taiwan Semiconductor
Manufacturing Co., we have been first. I don’t understand Nvidia’s
thinking there. But when a new process comes along, we jump on it.
There are risks to that. We have our moments. When you switch to a new
process, a lot of things can go wrong. When you don’t switch and stay
with an existing process, there are a lot of tools that help you out.
Once you have proven a process node, you are sure your subsequent
chips will work fine. We have made it through issues with new
processes and gotten through them just fine.

VB: What will happen with the transition to The Foundry Company, where
AMD will separate its manufacturing into a separate company?

RB: Our strategy has always been to have multiple sources of
manufacturing for our chips. We have used United Microelectronics. In
the last few years, we used TSMC more. We have grown and are buying
[500,000] wafers in a year. It makes sense for us to have multiple
sources. We will work with TSMC, UMC and the new Foundry Company too.
That’s going to happen over time.

VB: Have you been behind on the general-purpose GPU front [where a
computer uses the graphics processor to do non-graphics tasks such as
physics processing]?

RB: We were the first ones to talk about GP-GPU in October 2005. We
talked about supporting physics calculations on the GPU and doing
projects like [email protected] What you really need for it to be software
is a robust software environment. Nvidia has its CUDA programming
language, but that’s an interim step. You need open, standards-based
environments such as OpenCL [an open programming environment created
by Apple and supported by others] and DX11 [Microsoft's graphics
standard]. Ultimately they will prevail. It’s a big investment for us
to support GP-GPU.

VB: Do you think an ecosystem will develop around AMD as it has for
Nvidia, with new companies making use of their GP-GPU technology?

RB: That’s a contrast. They want the ecosystem to form around CUDA. We
want it to form around OpenCL and DX11. As that happens, we know we
will win, since standards win against proprietary solutions.

VB: A lot of your announcements are trailing theirs.

RB: We’ve been quiet, but we have the same support. We have millions
of chips in the market with the GP-GPU capability. We think that our
competitor is behind, and they are throwing up smoke screens in the
core graphics area. They want to show they have something, and so they
put more emphasis on CUDA.

VB: Is the mainstream market shifting to having stronger integrated
graphics in chip sets? For a long time, Intel has offered low-end
graphics capability in its chip sets. But it looks like it isn’t
cutting it anymore.

RB: We used to compete against Intel in Intel chip sets. We did
reasonably well. The challenge is they are so closely tied together
it’s tough to break in. The pricing is challenging. The market is
becoming savvier about the need for stand-alone graphics or better
chip sets with better integrated graphics. The question is whether
someone can break the hold that Intel has on the market.

VB: Do you think that people have a strong appetite for high-end
graphics?

RB: If you look at the sales, it’s still healthy. The third quarter
was down a little from last year. But the market keeps getting bigger
and bigger.

VB: Is PC gaming healthy?

RB: There is a very compelling story on why PC gaming is so healthy
now. When you consider shipments of high-end gaming rigs and strong
sales of online games such as World of Warcraft, you can see how
strong it is. Sales at the high end continue to astound me. We address
that by putting two of our sweet spot chips into one board. The gaming
platforms tend to be the first place where games appear, but we are in
about 80 percent of the gaming consoles sold, since we’re in the
Microsoft and Nintendo consoles.

VB: Do you have any worry that graphics may be impacted in the slowing
economy?

RB: I don’t know if graphics will be singled out in terms of impact.
We are heavy on the consumer side compared to the enterprise. Everyone
in the world is hanging on to see how bad the storm will be in the
quarter and going forward.

VB: Is the future good for graphics?

RB: Yes, we’re feeling good. I wish the economy were better. We have a
strong product line. The amplitude of our gains would be higher if the
market was stronger. But the flip side is that you better have a good
lineup in a slow market or things will get ugly.

VB: I wonder what the Wii means for graphics. It’s such a low-end
machine in terms of graphics.

RB: In some ways, I think it is great news for everybody. The Wii is
opening up a lot of new households to games that haven’t played before
in past generations. Sales keep going up and that is good for anyone.


http://venturebeat.com/2008/11/26/qa-with-rick-bergman-head-of-amds-graphics-business-on-fighting-nvidia-with-a-sweet-spot-strategy/