Nvidia recently an Editors Day at its campus in Santa Clara, CA, at which the company unveiled its new graphics hardware, the GTX 200 GPU family, and offered up hardcore gaming, visual computing, and compute functions as examples of the core uses this new technology serves.
The event began with a demo of the upcoming Photoshop CS4, which will include 3D object manipulation and editing. The demo included 2D decals being applied to 3D models as textures. While the tech wasn't demanding enough to trumpet the quality of Nvidia's GPUs, it did show that gaming isn't going to be the be-all-end-all purpose of its processors from here on out -- a thread that was taken up and run with throughout the day.
After the short presentation, David Kirk, Nvidia's chief technical officer, came out to discuss the company's direction, particularly CUDA, technology which allows non-graphics applications to be coded in C and run on its GPUs.
"If you want to run some other program you don't want to hide your program inside the graphics API... CUDA allows you to do that... it runs outside the graphics pipeline," noted Kirk, adding that there have been over 60,000 downloads of the free CUDA SDK so far, worldwide.
Throughout the day, Nvidia execs highlighted how CUDA, which is compatible with many of the company's currently-released GPUs, has enabled independent developers to make major performance gains on applications; for example, the National Center for Atmospheric Research has reported a 20% speedup in its weather forecasting code, with only 1% of that code ported to CUDA.
Kirk moved on to discuss how the power of the next generation of Nvidia GPUs will allow for the expected improved graphics performance -- using new techniques as well as raw horsepower. "We want to find a bridge between these worlds of the traditional graphics pipeline and this highly parallel massive compute we build as part of the GPU... this leads us to what I would call 'programmable graphics', which is the next step in graphics computing."
In Kirk's view, the goal here is hybridization between rasterized graphics and real-time ray tracing for certain objects that would benefit -- such as cars. "I think the best opportunity for us is to pick and choose among all of these techniques... the [improvement] of graphics hardware is going to allow for massive parallel processing... mix and match these techniques to produce new effects."
When asked if these techniques could be applied to the PlayStation 3's Nvidia-supplied RSX processor as well as the next generation of Nvidia GPUs, Kirk replied that "anything's possible" and posited that "the RSX would do the rasterization and the cell could be used to calculate the ray tracing," also noting that it's up to Sony to support the technique in its development tools.
At this point, Nvidia's Jason Paul officially unveiled the GeForce GTX 200 series, the company's new line of GPUs, which will go on sale this month. The GeForce GTX 280 will be available on June 17 for $649; the GTX 260 on June 26 for $399.
With 1.4 billion transistors and 240 processing cores operating at 933 gigaflops of power, it's a formidable GPU series -- or as Paul put it, "This is the largest and most complex GPU in the world," describing it as "a new archtecture that allows us to go beyond gaming."
Important to its performance, according to VP of technical marketing Tony Tomasi, is its ability to switch processing "thousands" of threads with no cost or overhead, delivering high efficiency without having to "tightly pack or couple" these threads, as with other architectures.
An impressive demo of how this can be used for non-gaming applications -- which may be relevant to developers -- was performed by Sam Blackman, CEO of Elemental Technologies. Its BadaBOOM encoding software allows video conversion to popular formats to be performed at a much faster rate than conventional programs.
Blackman demoed on a PC with the GTX 280 GPU, encoding an HD movie at 1400 x 1080 resolution using the H.264 codec at the rate of 47 frames per second -- comparing to 5 frames per second using the standard CPU-driven codec in Adobe Premier Pro.
Tomasi discussed the thinking that went into the design of the GTX 200 series. "We did a lot of tuning... to get a lot closer to the theoretical peak" -- with performance measured at 93.1% vs. theoretical limits of the hardware. According to Tomasi, Nvidia "[tries] to optimally balance our architecture not just for today's applications but what we think is coming."
Stanford professor Vijay Pande, director of the [email protected]
project, came out to explain the vast improvements in processing power the latest Nvidia tech offers to the project -- offering over five times the performance of the PS3's [email protected]
Manju Hedge, Nvidia's VP of PhysX Solutions, talked up the GTX 280, too, saying that porting the PhysX code to CUDA had offered vast performance gains compared to the days when PhysX supplied its own hardware.
Of course, the primary purpose of Nvidia's GPUs is still PC gaming. Dominic Guay, technical director of Far Cry 2
for Ubisoft Montreal, delivered a demo of the game on a GTX 280-equipped PC. Using the company's new multi-purpose game engine known as Dunia, the game depicts Africa realistically. "We wanted to move away from the goal of having beautiful screenshots... that's an old goal... we wanted dynamic beauty," said Duay. "A big part of the beauty of nature is the dynamism of nature. The way trees shift in the wind."
The demo was impressive, of course, and was followed up by demos from developers such as Gas Powered Games with Space Siege
and GRIN with Bionic Commando
, among several others -- including an intriguing demo of Korean online world Nurien
, which showed the value of top-notch graphics in a casual context.
Gearbox Software's technical director Corrinne Yu delivered an impassioned presentation about how Nvidia's GPUs offer compelling new programming challenges for games -- that will result in the best results yet seen in the industry.
The day ended with Jen-Hsun Huang, CEO and co-founder of the company, delivering his vision for the future and answering the press' questions. His talk boreded on obsession with parallel computing -- using the new GPU for more tasks than just delivering graphics processing in a gaming context.
A question was asked about the integration of compute functions in DirectX 10, but Huang doesn't mind, pointing out that Nvidia remains cross-platform with CUDA, which he sees as a distinct advantage. In summation of the company's current direction, he offered, "We believe in creating an architecture that's general purpose, but it's very focused in what it does."