In Your Face. In a Good Way
Visualizing the relationships in large data sets keeps getting better.
“What you see is what you get” can always stand improvement. This is especially true for industrial designers and engineers inspecting the glut of data associated with creating, evaluating, and simulating products and production processes. The Next Big Thing in data synthesis is really an old thing: applying new hardware and software to data visualization.
It’s in the cards
While computer microprocessors are getting more and more powerful, they can still be a bottleneck, especially when processing and displaying massive datasets. The typical solution has been to throw more central processing units (CPUs) at the problem, hence parallel processing, blade computing, and lately, on the computer motherboard itself, multicore microprocesors. Another solution is to offload some of the grunt work—data processing—to some other chip. The prime candidate for this is the graphics-oriented microprocessor on the graphics card: the graphics processing unit (GPU).
Giving GPUs more work requires more powerful graphics cards. Consider the Quadro FX 5800, from Nvidia Corporation (www.nvidia.com). It offers up to 240 CUDA parallel cores and has 4 GB of graphics memory (frame buffer). The graphics card, which retails for about $3,500, can generate billions—as opposed to millions—of color variations. Its memory bandwidth runs up to 102 GB per second. Moreover, the card provides 64-bit floating point capabilities in shading, filtering, texturing, and blending. Rendered images are stunning, realistic, responsive, and sans “jaggies.” (However, the maximum power consumption is 189 watts.)
Back up a bit. CUDA—Nvidia’s Compute Unified Device Architecture —is a software approach for parallel processing on the multicore processors found on Nvidia’s graphics cards. In short, CUDA is helping make high-performance computing ubiquitous. CUDA is available on three brands of Nvidia graphics cards: Tesla, for high-performance computing; Quadro, for professional graphics workstations; and GeForce, for conventional users. For example, according to Nvidia, each of the 128-thread processors in a GeForce 8-series GPU can manage 96 concurrent threads, for a maximum of 12,288 threads. Each of those processors has its own floating-point unit, set of registers, and shared local memory. That’s a lot of parallel processing power. It’s also great for electronic games, which is the test market for visualization wherever massive data sets are involved, including product design, the design of large assemblies, finite element analysis, computational fluid dynamics, and simulations of everything from products to entire production lines.
A new algorithm
Ray tracing is a graphics rendering technique for generating physically correct simulations of lighting. The technique involves tracing the individual paths of individual light sources in a scene. This approach accounts for the natural scatter of light rays in several different directions (called “incoherent rays”). It’s these secondary reflections that give an image its photorealism. It’s also these secondary reflections that require memory access to many disparate parts of a scene. Ray tracing is time-consuming and computationally intensive. Generating photorealistic images can take hours.
San Francisco-based Caustic Graphics is the latest company to come up with a new algorithm for speeding up ray tracing. In April, 2009, the company introduced CausticRT, which for $4,000 consists of the CausticOne card and the OpenGL-based CausticGL programming API. CausticOne is a PCI Express ray tracing accelerator card with 2-GB RAM. It acts as a coprocessor that complements the existing GPU or CPU in a computer. Together, CausticOne and CausticGL organize the incoherent rays into a data flow that better utilizes a desktop computer’s GPU/CPU to produce photorealistic images up to 20 times faster than current CPU-based ray tracers, according to company officials.
CausticRT does not require the CausticOne card. The ray tracing software can take advantage of any multicore computer running on Apple Mac OS X (Leopard) and Microsoft Windows XP. (Linux is forthcoming.) The current CausticRT version supports over 100 million uninstanced vertices with no paging required. The ray tracer can render scenes with any number of lights.
Coming soon
Seeing 3D displays without goggles (called “autostereoscopic 3D display”) is one of the Holy Grails in visualization. In November, 2008, NEC LCD Technologies showed a 12.1-inch, glasses-free, horizontal, double-density pixel (HDDP) 3D LCD display (SVGA; 800 x 600 pixels). The HDDP structure consists of horizontally striped RGB color sub-pixels. The 3D effect comes from alternately displaying pixel-by-pixel the images for the right eye, then the left eye. Industrial design and surgery displays should be available in a year.
WOWvx 3D technology from Philips Electronics involves covering a flatscreen display with transparent lenticular sheet for the glasses-free 3D effect. The lenticular sheet offsets the pixels in the LCD image plane. A person sees in one eye the LCD pixels directly under the lenticular sheet; in the other eye, the person sees the offset pixels. The LCD provides high-definition displays—3840 x 2160 pixels, which is basically four times the pixel count of current HDTV resolution. The high-speed display of this huge number of pixels results in multiple images at slightly different angular views presented fast enough to trick the brain into seeing 3D.
And now for something really different. From Japan’s National Institute of Information and Communications Technology comes a box: gCubik. Each side of this box is a 3.5-in. touch screen display (for now, VGA; 640 x 480 pixels). By looking at these screens, people see a 3D object “inside” the box. Touching the screens lets people “move” the object “inside” the box. Speakers and 6-axis acceleration sensors make the object inside the box interactive.
