Hot Chips, Tough Choices

Dave Andrews

Rich color, realistic 3-D, multichannel audio, and full-motion video are just some of the benefits that chip vendors are touting for a new wave of multimedia accelerator processors. However, these chips' market success will rest heavily with the silicon vendors' software partners. And in some cases, those developing the next wave of games to benefit from these processors have discovered that the path to multimedia nirvana can be a steep learning curve.

Nvidia (Sunnyvale, CA) recently announced the NV1, an integrated multimedia chip that combines GUI acceleration, 3-D rendering, video acceleration, and audio processing. SGS-Tomson Microelectronics will also sell the Nvidia chip as the STG 2000. Nvidia says that peripheral-card vendors like Diamond Multimedia Systems (Sunnyvale, CA), a company that has already announced it will use the chip in a PC board to be released later this summer, will be able to sell NV1-based add-in cards for about $200. Diamond and Taipei, Taiwan-based LeadTek showed preliminary NVI-based boards at the Computex show held last June in Taiwan. LeadTek says a 1-MB DRAM-equipped version of its board will sell for under $250.

However, Dean McCarron, an analyst specializing in chips and controllers at Mercury Research (Tempe, AZ), points out that the technology that gives Nvidia its edge is also a potential drawback. "Games developers have told me that, while worth the effort, the NV1 will be difficult to program to," McCarron says.

Part of the NV1's advantage is that it takes a radical approach to 3-D imaging. Instead of building 3-D objects from a multitude of flat polygons, the NV1 creates them using a much smaller number of "curved surface" polygons. As a result, less processing power is required for the kind of high-quality, real-time 3-D graphics that can make a PC game look very impressive.

Nvidia says the chip will accelerate applications written to other APIs like RenderWare, 3-D DDI, and 3-DR. But to get the highest level of performance, you have to develop directly to Nvidia's hardware API. "The NV1 intrigues us because it's inexpensive and it's well designed in terms of performance and throughput," says David Kaemmer, chief technology officer at Papyrus Design Group (Somerville, MA). However, he adds, "To take full advantage of the NV1 would take quite a bit of work."

Kaemmer says that Papyrus expects to release what is essentially a custom version of its NASCAR car racing game for Windows 3.1 in September. Kaemmer says the NV1's approach works well for NASCAR, because it lets developers render the curves in the road and racing car with fewer vertices. But he also says that Nvidia's degree of market success will help determine how much effort his company devotes to rewriting its programs.

Another company placing a heavy reliance on softwar e technology for its multimedia accelerator is Netherlands-based Philips and its TriMedia programmable DSP (digital signal processor), at the core of which is a VLIW (Very Long Instruction Word) architecture capable of executing up to five operations in a single cycle. Although VLIW specifies parallelism, unlike CISC or RISC technologies that rely on the processor itself to discover parallelism, VLIW relies on software, specifically the compiler. And that means developers will have to rely on the quality of their compiler and diagnostic and analysis tools, which Philips will supply, when writing applications for the TriMedia, slated to ship in volume next year.

It will be interesting to see which of the approaches--that of Philips, Nvidia, or a more traditional, OpenGL API-based one that's expected from the 3Dlabs/Creative Technologies partnership--will succeed in the quest for multimedia success. Mercury's McCarron predicts successful companies will have a strong balance of hardware, software, and thi rd-party developers.