HP Professional AP500 Graphics: The Workstation Difference - Page 9

There are three AGP specifications: AGP 1X, AGP 2X and AGP 4X.

Page 9 highlights

Graphics: The Workstation Difference 9 Y screen coordinate Z value Red Green Blue Texture U coordinate Texture V coordinate Texture W (perspective) value Alpha When used in a triangle strip (common for high performance graphics), the PCI bus can handle roughly 10 million of the simple triangles per second. Using the full OpenGL interface, the PCI bus can handle 3-4 M triangles in the real world. Today's graphics accelerators can display 3-5 M triangles per second, which is a reasonable match for the PCI bus. Further, few applications can produce more than 0.5-1.0 M triangles per second to feed into the graphics system. Thus, the PCI bus is seldom a bottleneck for graphics on today's systems. The PCI bus supports multiple devices on a single bus. All of the devices share the bandwidth of the PCI bus. Several types of peripherals are beginning to approach or exceed the bandwidth of the PCI bus with a single device. These include graphics accelerators, network interfaces (such as Gigabit Ethernet) and storage controllers (Ultra2 SCSI and Fibre Channel). With multiple high performance devices, PCI bandwidth becomes a bottleneck. This is addressed in two ways: The first is to provide multiple PCI buses in a single system. This is done with the Professional Workstation SP100 and XP1000, which both offer two fully independent PCI buses, thus doubling the PCI bandwidth of the system. In addition, the XP1000 incorporates a 64 bit PCI bus, which is fully compatible with 32 bit PCI devices and, when used with 64 bit PCI devices, doubles the available bandwidth. The second approach is to use special purpose buses. The leading example of this is the use of AGP for graphics. AGP The AGP bus is the newest interface for graphics, and has received considerable publicity. In many ways, the AGP bus is an evolutionary growth of the PCI bus. In fact, the AGP specification notes that AGP is based on the 66 MHz PCI bus specification and adds a set of extensions to that specification. AGP differs from PCI in two major ways. First is performance. The AGP bus allows up to 1 GB/s of bandwidth (with AGP4X). Second, AGP is a point to point bus. This means that only two devices can be on the AGP -- the system controller (usually the same chip that functions as memory controller) and the graphics device. There are three AGP specifications: AGP 1X, AGP 2X and AGP 4X. They differ in performance, with AGP 1X providing 266 MB/s of bandwidth, AGP 2X providing 512 MB/s bandwidth and AGP 4X providing 1024 MB/s. AGP 1x is essentially an extension of the PCI 66 MHz interface. AGP 2X and 4X retain the 66 MHz bus of AGP 1X, but allow multiple data transfers on a single bus clock cycle. AGP 4X allows four data transfers on each clock cycle. 0054-0499-A

  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11

Graphics: The Workstation Difference
9
0054-0499-A
Y screen coordinate
Z value
Red
Green
Blue
Alpha
Texture U coordinate
Texture V coordinate
Texture W (perspective) value
When used in a triangle strip (common for high performance graphics), the PCI bus can handle
roughly 10 million of the simple triangles per second.
Using the full OpenGL interface, the PCI
bus can handle 3-4 M triangles in the real world.
Today's graphics accelerators can display 3-5 M triangles per second, which is a reasonable
match for the PCI bus.
Further, few applications can produce more than 0.5-1.0 M triangles per
second to feed into the graphics system.
Thus, the PCI bus is seldom a bottleneck for graphics on
today's systems.
The PCI bus supports multiple devices on a single bus.
All of the devices share the bandwidth of
the PCI bus.
Several types of peripherals are beginning to approach or exceed the bandwidth of
the PCI bus with a single device.
These include graphics accelerators, network interfaces (such
as Gigabit Ethernet) and storage controllers (Ultra2 SCSI and Fibre Channel).
With multiple high
performance devices, PCI bandwidth becomes a bottleneck.
This is addressed in two ways:
The first is to provide multiple PCI buses in a single system.
This
is done with the Professional Workstation SP100 and XP1000, which both offer two fully
independent PCI buses, thus doubling the PCI bandwidth of the system.
In addition, the XP1000
incorporates a 64 bit PCI bus, which is fully compatible with 32 bit PCI devices and, when used
with 64 bit PCI devices, doubles the available bandwidth.
The second approach is to use special purpose buses.
The leading example of this is the use of
AGP for graphics.
AGP
The AGP bus is the newest interface for graphics, and has received considerable publicity.
In
many ways, the AGP bus is an evolutionary growth of the PCI bus.
In fact, the AGP
specification notes that AGP is based on the 66 MHz PCI bus specification and adds a set of
extensions to that specification.
AGP differs from PCI in two major ways.
First is performance.
The AGP bus allows up to 1
GB/s of bandwidth (with AGP4X).
Second, AGP is a point to point bus.
This means that only
two devices can be on the AGP -- the system controller (usually the same chip that functions as
memory controller) and the graphics device.
There are three AGP specifications: AGP 1X, AGP 2X and AGP 4X.
They differ in
performance, with AGP 1X providing 266 MB/s of bandwidth, AGP 2X providing 512 MB/s
bandwidth and AGP 4X providing 1024 MB/s.
AGP 1x is essentially an extension of the PCI 66
MHz interface.
AGP 2X and 4X retain the 66 MHz bus of AGP 1X, but allow multiple data
transfers on a single bus clock cycle.
AGP 4X allows four data transfers on each clock cycle.