A First Look At Fermi And More From Nvidia

Nvidia kicks off its GPU Technology Conference (GTC) in San Jose, Calif., Wednesday with Dan Vivoli, head of the graphics chip maker's Professional Solutions Group, introducing Nvidia CEO and co-founder Jen-Hsun Huang. The Santa Clara, Calif.-based company hosted a similar but more gaming-oriented event called NVISION a year ago, but tightened the focus of this year's conference to GPU computing based on Nvidia's proprietary CUDA programming language for graphics processors.

Nvidia CEO Jen-Hsun Huang, delivering the opening keynote at GTC Wednesday, holds up a graphics card built around the company's next-generation GPU architecture, code-named Fermi. Nvidia has not revealed a shipping date or pricing for Fermi-based products, which would succeed the GeForce GTX 280 and other products based on the chip maker's current, 10th-generation G80 microarchitecture. But several Nvidia partners exhibiting products at the conference suggested April as a likely time frame.

Fermi represents a "radically different way of designing GPUs," according to Jen-Hsun Huang. The next-generation chip, seen here, features three billion transistors arranged into 512 CUDA processing cores that are in turn arranged as 16 streaming multiprocessors with 32 cores apiece. Fermi's CUDA processing cores were designed for heterogeneous "co-processing," according to Nvidia, so that programmers can exploit the relative strengths of both CPUs and GPUs for faster, more powerful computing performance.

The namesakes of two legendary physicists collide in this visual representation of Nvidia's next-generation Fermi GPU architecture incorporated into the company's high-end Tesla brand of graphics products. Fermi will be ported across Nvidia's other main product lines -- GeForce and Quadro -- but could ramp first in the high-performance compute (HPC) and supercomputer markets where Tesla and CUDA development is strongest.

Camera operators prepare to dazzle GTC attendees with 3-D magic during the first part of Nvidia CEO Jen-Hsun Huang's opening keynote, which focused on the visual computing innovation for which the company is most famous. Later portions of the keynote were presented in more straightforward fashion -- in recent years Nvidia has sought to push its mission far beyond gaming and graphics, developing its CUDA programming language for general purpose-GPU (GP-GPU) computing and growing an ecosystem of academics, developers and system builders focused in areas such as medical imaging, geology and cloud computing.

An attendee watches the 3-D-enhanced presentation of realtime applications of Nvidia graphics technology during CEO Jen-Hsun Huang's opening keynote at the GPU Technology Conference. Highlights included rendering of photorealistic still images via GPU-enabled ray tracing, a demonstration of physics-driven fluid mechanics and perhaps best of all, a mini-game that involved firing virtual crash test dummies at a computer-generated wall to shattering effect.

Jen-Hsun Huang touts Fujifilm's latest creation, the first commercially available digital camera with 3-D still image and video capabilities. "We believe 3-D is the next step after HD," opined the Nvidia CEO, who put his evangelizing money where his mouth was by dealing out several Fujifilm 3-D cameras as prizes to lucky audience members at the GPU Technology Conference.

TechniScan's David Robinson explains how his medical imaging company has used Nvidia's CUDA-enabled GPUs to build parallel-programmed breast tumor detection technology that can provide early detection of cancer in women and render images more than twice as fast as CPU-powered units. The TechniScan product uses a pair of Tesla T1060s graphics cards to scan and develop a diagnostic image in less than 30 minutes.

Nvidia CEO Jen-Hsun Huang, left, and Realtime Technology's (RTT) Ludwig Fuchs get geeky about RTT's GPU-powered visual customization engine for Ferrari that allows the car maker's customers to visualize features such as seat upholstery and dashboard design at the click of a mouse. Fuchs, CEO of the Munich, Germany-based 3-D modeling software developer, wrapped up his presentation by leading Jen-Hsun through a stunning live demo of RTT software that projected virtual Ferrari tire rims onto a computer display based on a physical rimless tire on stage that was being digitally filmed in 3-D.

Jen-Hsun Huang displays Hewlett-Packard's HP Mini 311 netbook, which is built on Nvidia's Ion graphics platform and an Intel Atom processor. The outspoken Nvidia boss made headlines with his criticism of larger rival Intel at last year's NVISION event, but mostly held himself in check on the first day of this week's GPU Technology Conference. Still, bad blood with Intel over Ion -- Nvidia alleges that Intel has made it difficult for the Ion platform to succeed in the netbook and nettop spaces -- seems likely to have spurred Jen-Hsun to quip later about the early emergence of Ion-based netbooks that "Ion almost stands for rebellion."

Jen-Hsun Huang holds up an Nvidia Ion motherboard for netbooks and nettops that includes GeForce 9400M on-board graphics at the center of the board and a low-power Atom processor from Intel. Vendors such as Acer and Lenovo have built ultrasmall desktop PCs based on the Atom platform and now Hewlett-Packard is shipping an Ion-based netook, the HP Mini 311. Nvidia boasts that Ion is a far superior graphics platform for Atom-based PCs than Intel's own integrated graphics for Atom systems.

Jeffrey Nichols, right, of the Oak Ridge National Laboratory, and Nvidia's Jen-Hsun Huang discuss the new supercomputer project at Oak Ridge that will utilize next-generation Fermi graphics processors to make a run at the Top 500 Supercomputers list next year. The Oak Ridge "Jaguar" supercomputer is expected to deliver 20 Petaflops of compute power, an order of magnitude better than the current top supercomputer in the world, according to Nichols.

Nvidia wows the crowd with Iray, a turnkey rendering solution for 3-D graphics application developers built by its subsidiary Mental Images. Here, Iray is used to stream a hosted, photorealistic mock-up of an office space over the Web as an example of a cloud computing application that architects and designers could use to show clients their work. Iray creates "beautiful images" that can't be distinguished from photographs, said Jen-Hsun Huang, while Nvidia graphics horsepower allows new image views to be rendered in mere seconds.

Nvidia sells its "co-processing" message with an illustration of time efficiencies that would be gained by taking a processing job and using the CPU to muscle through the serial code portion while the GPU tackles the portions that lend themselves to parallel programming. "It's always been 'A' or 'B,' " said Nvidia CEO Jen-Hsun Huang, referring to the programming of code for either central processors or graphics processors. "But why can't we have 'A' and 'B?' "

GPU-driven applications such as this simulation of water breaching a levee from Johns Hopkins University point to the real-world, life-saving benefits of parallel programming for graphics processors, according to Nvidia's Jen-Hsun Huang. While Nvidia is popularly known for its acceleration of visual eye candy in video games and the like, during his GTC keynote, Jen-Hsun took great pride in Nvidia's work with researchers at Johns Hopkins and companies such as TechniScan, a maker of a GPU-based early detection system for breast cancer.