Xilinx FPGAs Accelerate the Development of the Evans & Sutherland Ensemble Image Generator
“Without the density and speed of the Virtex parts, the Vid chip probably would have been implemented in a gate array, causing a much longer development cycle for the Ensemble boards. The engineering team would have had less opportunity to roll in new and enhanced features during and after the main design cycle,” said John Snow, PC Simulation Engineering Manager. Since the early 1970s, E&S has produced industry leading computer image generators for commercial and military pilot training. These special purpose graphics computers generate realistic and detailed out-the-window and sensor simulations in real-time for training people to operate a wide range of vehicles including aircraft, ships, trains, tanks, and automobiles. In April 1998, a small engineering team began working on the product
definition of the company’s first PC-based image generator for the simulation
market using graphics accelerator boards based on the company’s latest
generation of REALimage(TM) technology. Other image generators made
by the company use custom ASICs designed specifically for the demanding
task of real time image generation. While based on a standard graphics
chip, these new PCI form factor graphics boards have been specifically
tailored for the simulation market and contain simulation-specific features
not generally available on standard PC-based graphics accelerator boards.
This new image generator system, Ensemble(TM), is software compatible with
the company’s current high-end Harmony(TM) image generator.
Standard REALimage-based graphics accelerator boards have a gate array called the Pixel Converter 2000 that fits in the same place that the Vid chip occupies on the Ensemble boards. The Vid chip implements most of what is in the Pixel Converter 2000 and adds to that all the simulation-specific post-processing features. The engineering team took the Verilog code for the Pixel Converter 2000 ASIC and, without modification, compiled the code using the Xilinx and Altera synthesis tool chains. The initial Xilinx target was a member of the XC4000 family. The results showed that the Pixel Converter code, which is implemented in a 50K gate, almost filled up the largest Xilinx and Altera parts available at the time and ran at about half the required frequency. Note that this was without modifying the code in any way to make it map more efficiently to the FPGA architecture, but the results left the team uneasy about the prospects of implementing the Vid chip in a FPGA. “At this time, Xilinx provided E&S with preliminary information about their new Virtex family. As soon as Xilinx delivered an early release of the tool chain with Virtex support, the Pixel Converter code was recompiled and targeted for Virtex. This time the results showed that the Pixel Converter code occupied about 50% of a XCV300 and ran at about 75% of the required speed. These results led us to believe that, with some work, the Vid chip could be implemented in a Virtex part and meet the system timing requirements. The team targeted the Vid chip to run at 80 MHz, which would support a 1280x1024-resolution screen refreshed at 85 Hz,” said Snow. The Quartet board, which is actually a two-board set with four REALimage 3000 chips, has the highest pixel fill performance of any PC-based graphics accelerator card currently produced by E&S. It can process 400 million textured, shaded, and antialiased pixels per second. The Quartet boards are designed to use Virtex XCV400 through XCV1000 chips for the Vid and four Spartan XCS40XL “Filter” chips. The Filter chips implement the antialiasing filter and reduce the pin count to the Vid chip. Detailed design of the Quartet board and the Vid chip began in August of 1998. The first prototype Quartet boards were powered on for the first time in early November 1998. A prototype Ensemble system was demonstrated three weeks later at a trade show for the simulation industry. “Using Virtex to implement the Vid chip allowed the Ensemble engineering team to pursue this aggressive design. The team was able to commit to pin outs of the Vid and Filter chips early in order to begin the circuit board layout while continuing to roll features into the design of the FPGAs. The availability of large amounts of block SelectRAM in the Vid chip also allowed us to implement many of our sophisticated post-processing algorithms,” concluded Snow. |
More
Information
Products
|