Nvidia Fermi Marks Company's Next-Generation GPU Architecture

Nvidia plans on releasing its next-generation graphics processing unit architecture, code-named Fermi, within the next few months. Fermi will incorporate more than three billion transistors and 512 CUDA cores, Nvidia says, which will accelerate its parallel-computing abilities and performance for applications involving ray tracing, physics, finite element analysis, high-precision scientific computing, sparse linear algebra, and search algorithms.

"It’s a brand-new architecture which is designed to be a computer first," Nvidia CEO Jen-Hsun Huang told a developer conference in San Jose, according to Reuters. "We call it a supercomputer with the soul of a GPU." Fermi will initially be incorporated into Nvidia’s GeForce, Quadro and Tesla graphics-processor lines.

Jeff Nichols, associate lab director for Computing and Computational Sciences at Oak Ridge National Laboratory, joined Huang onstage during the conference to announce that Nvidia’s new architecture will be utilized in his facility’s upcoming supercomputer, designed to research areas such as energy and climate change.  

According to Nvidia, one of Fermi’s key assets is its eightfold acceleration of peak double precision arithmetic, which is useful for the sort of supercomputing and high-performance applications that would run at the Oak Ridge facility. The architecture announcement also plays into Nvidia’s continued expansion from its traditional graphics-processing comfort zone into high-end computing for business and scientific endeavors.

In addition to CUDA, Fermi will support C++ and Microsoft’s Visual Studio development environment. Nvidia has not offered a definitive timetable for the architecture rollout. 

Nvidia’s market approach involves not only high-end workstations and supercomputers, but also the mobility market. Microsoft’s Zune HD portable media player uses Nvidia’s "Tegra" chip, based on an 800MHzARM 11 GPU and a Nvidia GeForce GPU paired with an image processor and high-definition video processor, to power its functionality with relatively low power drain.

Despite a recessionary environment driving down both PC sales and graphics chip shipments—or perhaps because of it—competition within the traditional graphics market has become increasingly intense. Earlier in 2009, Nvidia managed to take a few points of market-share away from industry stalwarts Intel and AMD, even as AMD’s ATI graphics business increased in the desktop segment.

Timed to coincide with March’s CeBIT 2009 conference in Europe, Nvidia announced a lineup of notebook GPUs for enthusiasts and power users, including GeForce GTX 280M and 260M GPUs, along with GeForce GTS 160M and 150M GPUs for high-performance notebooks. Nvidia insisted at the time that the GeForce GTX 200M and GeForce GTS 100M Series GPUs would offer 50 percent faster frame-rates than its previous-generation processors; in addition, those GPUs also support CUDA.

Analysts have predicted that the market for processors could spike in the third and fourth quarters of 2009, especially the graphics-processing segment, as the release of new operating systems such as Windows 7 drives a widespread tech refresh.

Jon Peddie Research stated in an April research note that, at least based on estimated graphics shipments and supplier market share earlier in 2009, the year could end on a positive note for Intel, Advanced Micro Devices and Nvidia. In the research firm’s calculations, Nvidia held a 31.1 percent domestic market share.