Nvidia NV30 pursues raw speed

The NV30, Nvidia's next sally into the graphics processing wars, will go for raw clock speed, the company says: it will need it to carry out a much more complex processing load

Nvidia says that higher clock speeds will be a priority for its next-generation NV30 graphics processing unit (GPU), which is due to be announced at the Comdex Fall trade show next month in Las Vegas. The chipmaker, which currently dominates the competitive graphics chip market, is shifting to a cutting-edge 0.13-micron manufacturing process for the chip to ensure that the complex, highly programmable GPU can reach the high clock speeds necessary. By contrast, rival ATI is expected to stick with its 0.15-micron process with its upcoming R350 chip. A smaller process geometry means chips run more efficiently, consume less power, and can fit more transistors into the same die area -- but by the same token, smaller processes pose increasingly difficult technical hurdles. This means that while Nvidia pushes the megahertz envelope, it is limited by the need to keep the process stable. The company has said in the past that it would not sacrifice stability in order to reach higher clock speeds, a concern that has kept ATI with the more mature 0.15-micron process. Nvidia's manufacturing partner, Taiwan Semiconductor Manufacturing Co. (TSMC), has made a "breakthrough" in creating a stable 0.13-micron manufacturing process for the NV30, according to the chipmaker. "We're way out there. It's looking pretty promising," Dan Vivoli, Nvidia's vice president of marketing, told ZDNet UK. "The real benefit for us is getting the frequency up and the chip to run at a really high speed, in addition to the architectural enhancements," Vivoli said. ATI works with manufacturer UMC to produce its Radeon graphics chips. The top-end Radeon chip has more than 100 million transistors -- more than most PC processors -- and is capable of churning out 2.5 billion pixels per second. The chip clock speed is 325MHz, and the memory clock speed is 310MHz. Nvidia's current GeForce4 Ti 4600, by contrast, has 63 million transistors. Industry rumours have suggested that the NV30 chips could clock in at around 400MHz. Nvidia will need to come up with a major performance and features boost in order to top ATI's current 9700 chip, which was touted as being twice as powerful as Nvidia's GeForce4. Nvidia downplayed the challenge of ATI's performance gap. "If you look at the actual gameplay, the games you can really measure, it's maybe 15 percent faster than the (GeForce4 Ti) 4600. It's not that much faster for something that's a next-generation part." Programmability is the key
The upcoming chip is expected to set a new milestone in programmability -- the capacity of a chip to run programs that can bring the realism of real-time-rendered 3D graphics closer to that found in movie special effects. Nvidia was the first to market with a mainstream programmable GPU last year, in the form of the GeForce3, and the next generation will continue the trend, Nvidia said. However, programmability greatly increases the complexity of the tasks a GPU must carry out. "In this generation the frequency is really important. These are real processors now, running programs on them," Vivoli said. "Before they were doing graphics calls, and now they're doing loops and branches and all kinds of things. They need high frequencies and a lot of power." The shift to greater programmability will also mean that the NV30's architecture needs to be far more precise in order to carry out more complex computations without errors, Vivoli said. Programmability has emerged as a major theme in the graphics world, with Microsoft and the backers of OpenGL racing to add programmable features into the two dominant graphics APIs (application programming interfaces). Features such as vertex shaders are complex to build into games, however, and currently require developers to target a particular API and GPU and to write complex assembly code. To combat this limitation, Nvidia -- backed by Microsoft -- has introduced a high-level language called Cg. This makes shader programming considerably simpler for developers and, Nvidia hopes, will mean that the advanced functions of its high-end graphics chips will be more widely used in games. 3DLabs is leading the charge to introduce high-level programming capabilities into the next generation of OpenGL, called OpenGL 2.0. Nvidia hopes that Cg will give the company access to the potentially lucrative graphics market of Hollywood special effects, and Vivoli said that all of the major Hollywood studios were either committed to using Cg or were "dabbling pretty seriously" with it. He said that games developers had also shown huge interest, given that the Cg rendering software is still only available in an incomplete test version. However, Nvidia acknowledges that the most high-end games developers like id software, creators of Doom and Quake, will always want to write assembly code in order to have the highest level of control over the finished product. "They don't mind doing the assembly language," he said. "They're the true rocket scientists of the graphics world."

See Chips Central for the latest headlines on processors and semiconductors. To find out more about the computers and hardware that these chips are being used in, see ZDNet UK's Hardware News Section. Have your say instantly, and see what others have said. Go to the Chips Central Forum. Let the Chips Central editor know what you think by email. And sign up for the weekly Chips Central newsletter.