Why is Advanced Micro Devices laying out billions of dollars to get into the often-painful world of making computer graphics chips? Because Moore's Law never sleeps.
If AMD's shareholders approve the US$5.4 billion purchase of ATI Technologies, AMD will become the largest maker of standalone graphics chips on the planet and one of the largest producers of integrated chipsets. It may also give AMD early access to Intel product plans.
The key to the deal, however, is that it will give Sunnyvale, Calif.-based AMD the internal know-how to devise processors with integrated graphics and communications capabilities. That so-called system-on-a-chip would effectively provide all the intelligence a handheld device or low-end PC might require. In addition, the company will be able to do chipsets with integrated graphics and video chips for digital TVs--products that Intel makes now.
When it announced the ATI deal on Monday, AMD said it hoped to come out with processors with integrated graphics, and with chipsets with integrated graphics, by 2008. AMD already makes processors with built-in memory controllers that connect the processor to memory.
"They want some of the integration possibilities that ATI gives them," said Dean McCarron, the principal analyst at Mercury Research. "There's a number of reasons against this deal, and there is not as much fit in the short term. But if you look long term, you can make a bigger argument for it."
"Visual computing is playing a larger role in what we are doing going forward," Hector Ruiz, AMD's chief executive, said during a conference call Monday with analysts.
Graphics integration is inevitable, given the relentless pace of the chip industry, analysts and others said. Under Moore's Law, the number of transistors on a chip should double every two years.
That's a lot of transistor to digest. Many times, designers consume the "new" transistors by increasing the internal memory pool, or cache, or by beefing up the performance of existing blocks on the chip.
But many times, they integrate existing chips into the main processor. Early PCs had a thing called a math coprocessor, which years ago was absorbed into the processors. The 512K cache on the first Pentium IIs consisted of separate memory chips located near the processor and cost US$20 to add. The cache on the current Core Duo chips is eight times as large, is part of the larger chip, and the whole thing costs less to manufacture.
Graphics have already been put into chipsets, and many in the industry have been waiting for graphics to appear more frequently in processors.
"It has always been thought that the graphics chip would get integrated into the processor," said Dave Epstein, a partner at Crosslink Capital. Epstein founded Raycer Graphics, which he sold to Apple Computer in 1999.
Where and how you add graphics depends on the market, Epstein added. Putting graphics in the chipset simplifies processor design, but the manufacturer then has to make different chipsets for different segments of the market.
While integrated parts don't typically perform as well as standalone chips, the difference often become irrelevant after a few years. Back in 1998, Intel scared graphics chip companies when it decided to sell standalone graphics chips.
The chips sold miserably, and Intel got out of the business the next year. However, it continued to make chipsets with integrated graphics. In a few quarters, Intel became the largest graphics chipset vendor in the world, a title it still holds today.
Other companies have tried similar combinations in the past. Cyrix came out with the MediaGX, which had integrated controllers and graphics. Intel also designed, but never released, Timna in 1999, which had integrated graphics and a memory controller.
Neither product succeeded, but not because integration wasn't a good idea. Cyrix, which got bought by National, rarely landed design wins with customers, and the performance of its processors typically lagged.
Play video
» Blades for superior performance
There are currently no comments for this post.