HP P Class 733/800/866MHz / 1GHz Supplementary Manual page 3

Geometry Engines
When comparing AGP 2X DMA to AGP 4X with Fast Writes, keep the following in
mind:
• AGP 4X with Fast Writes occupies the CPU while writing data to the graphics
device, preventing your application from doing other useful work. AGP 2X DMA
frees the host CPU while writing data to the graphics device, resulting in superior
application performance.
• Claims that 900MB/sec bandwidth is required for today's graphics applications is
simply wrong. In fact, 400MB/sec is sufficient to transmit over 14 million
triangles/sec to the graphics device. Furthermore, a 900MB/sec data transfer rate is
unattainable by real applications on systems with a 133MHz front side bus. AGP 4X
with Fast Writes does nothing to alleviate this bottleneck.
• AGP 2X DMA is a reliable data transfer mechanism that is known to produce excel-
lent application performance. AGP 4X with Fast Writes provides no performance
benefit over AGP 2X DMA, and its complexity compromises system reliability.
An additional DMA engine in the host interface chip uses 66MHz PCI protocol to
transfer data from the graphics device to main memory. This is especially useful for
reading the contents of the framebuffer, a critical operation for many Digital Content
Creation, Video Editing, and Visualization applications.
The host interface chip also supports fast hardware state switching for acceleration of
multiple concurrent rendering applications. Applications that use multiple OpenGL
rendering contexts will also benefit from this feature. An application that caches state
for different rendering scenarios in multiple OpenGL contexts will be able to rapidly
switch between them.
The geometry engines perform geometric transformations, lighting, model clipping,
and other vertex operations on incoming geometric data. This frees the host CPU,
leaving more processing power available for application work.
The geometry engines use floating point units based on hp PA-RISC processor tech-
nology to achieve maximum floating-point performance.
There are three geometry engines per geometry accelerator chip. The hp fx
single geometry accelerator chip containing three full geometry engines, while the hp
10
fx has two chips for a total of six full geometry engines.
Each hardware geometry engine supports a rich geometry feature set, including:
• Lighting and shading for up to eight separate OpenGL light sources
• All OpenGL primitive types
• Transformations
• View volume and model space clipping
• Material properties for accelerated rendering of lit surfaces
• Texture coordinate generation, useful in Scientific Visualization applications
• Environment mapping for fast realistic surface reflections
• Texture coordinate generation, useful in Scientific Visualization applications
• Environment mapping for fast realistic surface reflections
• Second generation hardware occlusion culling implementing faster rejection of
invisible geometry based on its bounding volume.
5
has a