FOR IMMEDIATE RELEASE
Tokyo, July 2, 1997 -- Fujitsu Limited today announced that it has developed the world's first 3D geometry processor. Called the Fujitsu Pinolite (TM), Fujitsu's new chip can dramatically improve the performance of the 3D graphics performance of graphics accelerator cards for personal computers.
Jointly developed with Fujitsu Laboratories, Ltd., Fujitsu will start shipping samples of the new chip (part number MB86242) immediately. When the Pinolite chip is used with a 2D or 3D graphics controller LSI on a PCI add-in card, 3D graphics performance can be boosted to 750,000 polygons per second.
Pinolite enables an ordinary home PC to display the same high quality, interactive 3D graphics as the highest performance game machines. The new 3D chip performs as good or better geometry operation performance as the fastest PC microprocessors, such as a Pentium II-266MHz chip. Pinolite takes the heavy burden of 3D geometry operation away from the CPU to achieve a well balanced interactive 3D graphics environment.
Geometric transformation performs coordinate transformation to calculate the motion (rotation and parallel displacement) of the approximated polygon model, calculate the projection to the two dimensional plane, and calculate the effect of light sources to each vertex. These geometric transformations are usually performed in floating-point format to increase precision.
Usually on a personal computer, these geometric transformations are executed by a host CPU. But the floating point unit integrated within a host CPU is not actually optimized for these such data flow operations for graphics, there are cases where the floating point operation performance of a host CPU is the bottleneck of the entire system operation. Also, once a host CPU starts these geometric transformations, all other operations up to a host CPU could be suffer and may cause a significant system performance drop, because geometric transformations are so heavy and complicated for a general purpose CPU.
Concerning such problems, the Pionolite was developed as an optimized 3D geometry processor for PC graphics. The Pinolite takes over the heavy burden of geometric transformations from CPU to offer a well balanced system performance and system throughput improvement. The Pinolite contains two separated PCI master/target controllers (one for data transactions with a host CPU and the other for post geometry data transactions with a graphic controller) which work simultaneously all the time.
And the Pinolite processor core part executes all the geometric transformations simultaneously, while these two ways of data transactions take place. By using this simultaneous operation scheme, the geometric transformations do not suffer from the overhead of data transactions through PCI buses. The graphic controller interface is PCI, so the Pinolite can be used as a front end processor for regularly available graphics controllers chips and cards.
The host CPU performance has been significantly improved. The most recent leading edge CPU such as Pentium II-266 offers very high floating point operation performance, as much as 500,000 polygons/sec depending on the APIs. However, in order to achieve this level of performance, host CPU must concentrate almost exclusively on these geometric transformations. As a result of such concentration, not only all other tasks, but even the scenario layer process and collision detection of highly complicated polygon objects, sometime suffer. This makes a smooth and dynamic interactive 3D graphics operations very difficult. The Pinolite offers a constructive solution to cover this fundamental problem of a shortage of floating point execution power in PC systems, by complementing the capabilities of today's CPUs.
Fujitsu Limited, Public Relations
A unit forming 3D graphics. This unit is used to divide the surfaces of a three dimensional object into an approximate representation. Polygons are usually made up of a very large number triangles or squares joined together.
*2 Flat shading
A rendering method to determine brightness by the normal vector on a polygon and the position of the light source and to shade the entire surface of a polygon with the color of the brightness. This rendering method produces a clear difference in the colors of adjacent polygons, making their boundary lines visible, so it is unsuitable for rendering smooth surfaces.
*3 Gouraud shading
A rendering method to produce color gradual shading over the entire surface of a polygon is performed by determining brightness with the normal vector at each vertex of a polygon and the position of the light source, and performing linear interpolation between vertices.
The normal vector at each vertex can be determined by taking an average of the normal vectors of all the polygons having the common vertex. For a triangular polygon, the brightness at each vertex is determined by the normal vector obtained for each vertex and the position of the light source. Therefore, the brightness of pixels inside a triangle is determined by interpolation. This rendering method represents color gradual variations between adjacent polygons, so it is suitable for rendering smooth surfaces.