HP Professional AP500 Graphics: The Workstation Difference - Page 8

Graphics: The Workstation Difference 8 both X-Windows and Windows NT. These two windowing systems are radically different in the way they work, but the same OpenGL interface is available on both for applications running in either environment. Of course, all 2D operations performed by an application are dependent on the windowing system, but the 3D OpenGL interface will be identical across the operating systems. AGP and PCI Two hardware interfaces are commonly used for graphics accelerators AGP and PCI. AGP is the newer bus; it is essentially a customization and extension of the PCI bus to support graphics. The AGP bus is up to 8 times as fast as the PCI bus, which offers clear advantages for graphics. However, as the analysis in the following sections shows, the PCI bus is adequate for today's 3D graphics workloads. This is supported by graphics benchmark results when are almost identical when AGP and PCI versions of the same graphics chip are tested in the same system. AGP does have several advantages, and will begin to show its performance edge as graphics and system performance continue to improve. PCI Bus The PCI bus is a true industry standard. Unlike proprietary buses commonly used on traditional UNIX workstations, the PCI bus is used by many vendors, on many platforms, supporting many different processors and operating systems. Millions of PCI based systems from hundreds of companies are shipped each year, supported by thousands of high performance peripherals. As a hardware specification, PCI has many advantages. It is a modern design, it runs at a very fast 33 MHz, and it is a full 32-bit interface. The base PCI bus delivers 133 MB/second of bandwidth, meeting the needs of today's systems. This performance is well matched to the needs of 3D graphics systems. A 3D vertex used in rendering will contain between 12 bytes and 40+ bytes of data. The simplest case would be Gouraud shading with compressed X and Y coordinates. The data structure might look like this: X coordinate Y coordinate Z value Red Green Blue This example is using common computer conventions, where each row is a 32-bit (4 byte) "word", and the series of rows represent continuous memory. A more complex case would support non-compressed coordinate values, shading, and perspective-corrected texture mapping. Non-compressed coordinate values eliminate the requirement for the CPU to perform data compression, thus reducing CPU load at the expense of additional bandwidth for data transfer. Texture mapping adds additional data to each vertex: U and V coordinates and, perhaps, perspective correction. Such a data structure might look like this: X screen coordinate 0054-0499-A