| |
As every serious sim
fan knows, new software is proving too demanding for our hardware. About two years
ago some of the principles used for accelerators on 3d workstations found their
way into the PC marketplace, and the world has changed forever! Finally we have
a way to deal with depth, high object detail and special effects (fog and transparency,
reflections, dynamic lighting, etc.) while increasing the frame rates to playable levels.
The beauty of the accelerator solution is that we can
attain high frame rates while freeing the CPU for other tasks: calculating the
figures needed to maintain a dynamic battlefield, adding real wind and weather
effects, increasing the accuracy of the flight model, and extending the physics
modelling to damage effects, weapons, and even secondary damage effects.
New Boards and Chipsets
Initially only
two accelerators met the challenge: the Verite R1000 by Rendition and the Voodoo
chipset made by 3dfx. Shortly afterwards nVidia
entered the picture with its Riva chipst, and then new
releases from ATI and others expanded the picture significantly.
The 3d wars have only really begun! ATI
has released their Rage Pro chip, which has built in support for DVD and
for Intels Accelerated Graphics Port. There
are two boards based on the new chipset, the XPERT@Play
and XPERT@Work. The former board is maximized for gamers and runs close to the
speed of the original Voodoo chipset but with more features. For example, the
XPERT@Play will do 16.7 million colors at 1280x764 with its 8 meg of SGRam. Early
benchmarks under WINBENCH '98 placed the XPERT@Play in its 8 meg version at 20%
FASTER than the Riva128 chipset (Direct 3d), but a new driver release made these
boards much closer in D3d speed.
Not content to stop there, ATI is now working on a twin texel solution (as nVidia's TNT and 3dfx V2 chipset). This new chip could easily place ATI up there with the other
twin texture processing boards, which have a significant power advantage over other designs.
The Velocity 128 3d used the
Riva 128 chipset, nVidia's second generation
chip. The Riva 128, with a 128 bit pipeline and 3.5 million transistors, used
100MHz SGRam and was rated at 20 billion operations per second or 100 million
pixels/sec. Like the Canopus Pure 3d the board also
had an output for your TV. However, the Riva is in the class of the Voodoo
Rush chipset and does standard 2d also, with built in Vesa 2.0. The V128 will
even coexist in peace with a Voodoo based board.
More recently nVidia released their TNT (for twin textel) technology. This chip more than doubled the power of the Riva 128 design, and the first release of a board using the chip was STBs Velocity 4400. This may prove to be the hottest D3d solution this year.
Rendition is not
out of the picture either! In the winter of 1998 Rendition released their 2nd generation chipset,
the 2200 and its variant the 2100. With an integrated 230MHz RAMDAC (170MHz
on the 2100), improved Z-buffer, support for AGP and up to 16MB of SGRAM, hardware
MPEG-2, video input and output, and at least double the performance of the V1000.
One of the early incarnations of this chipset on the
Diamond Stealth II S220 was as fast as 3dfx in D3d at 640x480, and 40% faster
at 800x600. There are no OpenGL drivers yet, however. 2d performance is better
than the V1000, but not as good as other 2d/3d solutions. Still, for the dollar
the card was a good buy at around $119.
Recently Rendition announced their third generation chipset, the Redline.
Its likely that this new chip will compete with the Riva TNT when released in the winter of 1998.
Click to continue
. . .
|
|
In late February 3dfx released their VOODOO
2 chipset, which more than doubled the speed of their first offering with
dual processors and a larger texture cache. This chipset is designed for both
AGP and standard PCI.
The new board has two texture processing units instead
of one. The first is used to render triangles while the second texture-maps
them. V2 will be able to simultaneously apply two textures to each triangle
for single-pass, single-cycle rendering of effects such as trilinear filtering,
lighting, spotlights, and image details. Unlike V1, V2 features triangle setup
in hardware thus further offloading the CPU. These features increase rendering
speed, but just as important they enable game developers to enrich objects with
much more complexity, detail, and depth.
The bigger surprise was that consumers could install TWO
PCI Voodoo 2 boards in a single PC. The chipsets will automatically detect each
other and begin operating in a mode 3Dfx calls "Scanline Interleave" or SLI
for short.
In the meantime if you aren't ready
to go to Voodoo2, consider a 3dfx board with 4 meg of texture cache. Why? When
a game has to draw a texture on a polygon, and that specific texture is not
present in the texture memory, it must throw away some textures that are currently
stored and load the new texture from main ram or disk and apply it to the polygon.
This results in a slowdown of the game.
As of this update (September 16/98) few have had their hands on the new Banshee boards.
However, early results show that Banshee is about 20% faster than the first release V2 boards. Banshee is a 16 meg V2 board with faster memory technology and a 100MHz clock compared to the 90MHz clock of the first release V2 boards. Banshee, like TNT, is both a 2d and 3d solution in one chipset.
But while Banshee is faster than V2, it isn't quite the speed of TNT in D3d. Furthermore, where TNT does very well at higher resolutions (800x600 and beyond) Banshee does not. And TNT is a full 2x AGP part, where the first release of Banshee is an AGP part only in name, lacking any AGP features at all. (The second release in early 1999 will be a true AGP part).
NECs PowerVR2 is a great improvement
over the first PowerVR chip but lacks some features that the top 3rd generation
chipsets incorporate. Rendition, 3dfx, and nVidia all have full feature sets.
Lacking features like fog effects in hardware means that the chip will be slower
than its rating when these effects are enabled. However, reports from users
are indicating that the board is comparable in speed to 3dfx and looks just
as good. I will finally be able to do hands on testing in the next week or so.
Recent
new entries into the 3d arena include the Matrox
MGA G200 chipset. This chipset brings Matrox back into the main stream of
3d acceleration while maintaining their lead in 2d speed.
3d Terms, Definitions and Benchmarks
What is alpha blending? What is depth cueing? How does MMX work?
Is the AGP a new economic measure? I thought you'd never ask!
For more information, go to...
3d Benchmark Comparison
3d Terms
Direct3d
AGP and MMX Explained
In late February 3dfx released their VOODOO
2 chipset, which more than doubled the speed of their first offering with
dual processors and a larger texture cache. This chipset is designed for both
AGP and standard PCI.
The new board has two texture processing units instead
of one. The first is used to render triangles while the second texture-maps
them. V2 will be able to simultaneously apply two textures to each triangle
for single-pass, single-cycle rendering of effects such as trilinear filtering,
lighting, spotlights, and image details. Unlike V1, V2 features triangle setup
in hardware thus further offloading the CPU. These features increase rendering
speed, but just as important they enable game developers to enrich objects with
much more complexity, detail, and depth.
What is the Accelerated
Graphics Port specification? What difference does it make? What is MMX and do
you need it? For more information see:
MMX
AGP
|
|
