Siggraph

Unlike the slowly changing, monolithic market for CPUs, the development of graphics chips and cards looks chaotic, with chip designs, products, and companies coming in and out of the market at a near furious pace over the decades since the first graphics technology delivered in 1960.

Over the past eight years or so, however, two companies have come to dominate the market for discrete chips and cards: ATI (since 2006 a division of AMD) and NVIDIA. While the companies continue to slug it out, most everyone else had fallen to the wayside. Now, the two companies own 98 percent of the discrete GPU business, according to on Peddie Research.

Peddie also says that the computer graphics market has grown from $1 billion in 1980 to some $60 billion today, for a remarkable growth rate of 16.5 percent for 27 years.

Those numbers could only make it attractive for the 800-pound gorilla of silicon: Intel. While the company has made low-end integrated graphics chipsets for years, over the past few months Intel has slowly revealed plans for competing directly with ATI and NVIDIA for the high-end market.

At Siggraph, Intel researchers delivered papers detailing its proposed chip architecture–code-named “Larrabee”—which features a new approach based on lashing together multiple x86 core processors, an architecture they should feel comfortable with, since it’s been key to Intel’s CPU design since the 1970s. Their first products might hit the market as early as 2009.

A multi-core approach is also used by NVIDIA and ATI, which currently employ hundreds of cores in their massive chips to create a reprogrammable on-the-fly parallel graphics rendering architecture.

Nvidia researchers released a statement at the show challenging Intel’s approach; they claim that a larger challenge for parallel computing is helping developers decide how to divide a problem into parallel tracks and then design software to leverage a parallel processor, something they’re doing at a furious pace, buying companies that make software to enable more flexible programming of their GPUs.

Nvidia also questioned whether or not current applications will really run on Larrabee without the need for modification, not an insignificant challenge with the thousands upon thousands of apps out there that are tightly integrated to current GPU hardware.

“We’ll just have to wait to see what (Intel) actually deliver(s),” says Shawn Worsell, NVIDIA product manager.