SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Gold/Mining/Energy : Day trading in Canada -- Ignore unavailable to you. Want to Upgrade?


To: Chad Barrett who wrote (1815)12/14/1998 12:17:00 AM
From: Marc  Respond to of 4467
 
AGP is based on PCI, but is designed especially for the throughput
demands of 3-D graphics. Rather than using the PCI bus for graphics
data, AGP creates a dedicated point-to-point channel by which the
graphics controller can directly access main memory. This AGP channel
is 32 bits wide and runs at 66 MHz. This translates into a total
bandwidth of 266 MBps, as opposed to the PCI bandwidth of 133 MBps.
AGP also supports two optional faster modes, with throughputs of 533 MBps and 1.07 GBps.

AGP vs. PCI

Is there really that much difference between PCI and AGP? If so, is
the difference great enough that I should invest in a new graphics
card now? Well, in order to make that determination we should look at
the first question. As you can see above, AGP offers a greater data
transfer rate advantage when it comes to moving the geometry stream
from the processor to the graphics card. When handling large texture
databases, it's GART table allows the operating system to manage
textures in off screen memory as well as in system memory, and allows
the graphics card direct access in either location.

Before AGP came along, game developers had three options for managing
textures:

1.Limit the texture database to only that which would fit in
off-screen memory. Although this delivers outstanding frame rates,
memory size constraints limit the developers artistic creativity. This
means, with some of the older graphics cards out there, texture space
could be reduced to as little as one megabyte.

2.Use the operating system to manage textures in main memory, and make
the processor copy textures from main memory to graphics memory as
needed. This process can create a major bottleneck even with
relatively small textures.

3.Or, place the most frequently used textures in off screen memory
(like the first method above), and then reserve additional space in
system memory for the remainder of the texture database. In this case,
if the graphics chip needs a texture that is not in graphics memory
the accelerator must use PCI master mode (DMA) to transfer the
needed texture into texture swapping space in the graphics memory.
This seems to have been the preferred method for most game developers
as it and can be programmed under Direct X Immediate Mode and results
in acceptable performance.

Basically AGP is a hardware-modified approach to this last method.
Memory management is a little more flexible, and because of its faster
clock speed and bus pipelining, the data transfer rate is increased.

AGP deals with textures in two different ways. One, called "DMA Mode",
which operates similarly to the last method above, but the transfers
occur over the AGP bus rather than PCI. The other is called Execute
Mode. The other called "Execute Mode". allows the graphics chip to
access texture information in main memory without first copying it to
local graphics memory. In effect, bus throughput of both of these
modes is the same.

The folks at Intel are hyping Execute Mode primarily because it only
runs on AGP and, as we all know, they want users with existing systems
to upgrade their PCI Pentium systems in favor of more costly Pentium
II AGP systems. However, in the real world, game developers and users
should prefer DMA mode because it offers outstanding performance and
compatible with current PCI accelerators.

AGP's DMA mode provides excellent concurrency because the software can
determine which texture to swap before the texture is actually needed,
while the processor is still calculating the geometry. The graphics
card can then begin fetching the texture before it is needed to paint
pixels on the screen. In AGP Execute mode, the raster engine determines texture accesses at the final stage of the 3D pipeline. At
this point, the accelerator is useless without immediate access to
textures. In this way, AGP creates a nasty performance bottleneck.
Instead of having immediate access to textures in high bandwidth local
memory, the accelerator goes into Wait State while it reconciles
access to slower system memory across the AGP bus fracturing
concurrency.

On the other side of the performance equation is the CPU. When AGP
texturing is turned on, the graphics card takes control of the main
system memory bus in order to access texture data. When this happens,
the CPU access to main memory is blocked. If the load on the CPU is
not great, this isn't a problem. But if the CPU is engaged in a high
demand task (such as a game) it is likely to encounter roadblocks in
its attempt to access main memory. While not currently a problem this
could however affect future implementation of larger game programs
that use extremely large texture maps. The best solution to this
problem is having a larger local graphics memory configuration. More
memory on the card itself negates the necessity of accessing textures
from main memory. In conclusion, while AGP does have a lot going for
it, anyone who cares about 3D performance is advised to make your
upgrade by choosing a high performance graphics card with a generous
memory configuration - or at least the option to add a memory upgrade
later. Some cards based on the Intel 1740 chip for example only have
2MB of onboard memory. Here at MediaTek we think that purchasing one
of these cards may lead to a backward glance in the near future. We
encourage you to read as many graphics card reviews as you can before
making that purchase. The more you learn now the happier you will be
as this platform progresses.


Hope it helps!

super7.net