Matrox Millennium G200
Is the MGA-200 another in a fairly long list of graphics chips released before it's time? Well, if you talk with anyone directly involved with marketing I'm sure that the answer would be no, but from the volumes of mail I've received since the chip was made generally available back in July, I wonder. I first tested the Millennium G200 back in July when I was asked to build a Pentium II system for a client and put the card through it's paces on an FIC VA-503+ (w/ CD version of MVP3 chipset) and at the time I recall being quite impressed. Sure, there were a few problems but I really didn't get a chance to work with it since it had to go out when the system I was building was done. Financial constraints being what they were, I was unable to get ahold of another until just a week or so ago and, as I'm sure that the majority of G200/Super7 users know, the newest drivers address 90% of problems experienced by stalwart users of the platform.
Matrox, with it's twenty years of expertise in the graphics hardware field had a banner
year in 1997, winning over 300 awards for their fine products and yet was not at all
content to rest on their laurels. Since it's announcement at the beginning of May it was
clear that Matrox was focusing a great deal of time, energy and money into developing the
MGA-200. And the chip itself sounded extremely promising. Basing the MGA-200 on a 128-bit
DualBus Architecture,employing two independent 64-bit buses that operate in tandem within
the graphics engine, effectively doubling the raw performance of almost every operation.
The 128-bit DualBus architecture also uses Dual Command Pipelining which allows read and
write phases of two consecutive commands to be overlapped and executed concurrently. The
DualBus Architecture delivers unquestionably fast 2D performance even at high resolution
and color depth. The 64-bit granularity of the 128-bit DualBus means high refreshment for
small bitmaps and fonts, generating greater performance than with a traditional 128-bit
The MGA-200 also uses a process called Vibrant Color Quality Rendering (VCQ). Using 32-bpp color accuracy throughout the rendering pipeline, VCQ is capable of rendering displayed images in 32-bpp color from source texture maps of up to 32-bpp color. The VCQ architecture also provides high quality analog color output to avoid washed out images.
One thing above all else is for sure with the MGA-200: the chip delivers outstandingly sharp contrast and highly saturated deep color that is, for all intents and purposes unmatched in image quality by any of the current nex-gen chips, especially D3D games, image editing, CAD and animation.
Within the MGA-G200, Symmetric Rendering Architecture (SRA), treats AGP memory almost as if it were local framebuffer memory. It is able to draw to, render to and read from AGP memory using the high bandwidth of the AGP 2X bus. The MGA-G200 also leverages the SRA to implement a hierarchical texturing system (HTS) for 3D rendering. With HTS, the MGA-G200 can store surfaces - either textures, bitmaps or video streams - across three levels depending on the frequency of use or the size of the texture. The first level is a large on-chip cache, the second level is the up to 16MB of local framebuffer memory and the third level is the AGP memory. - The Symmetric Rendering Architecture is used as a key component of 3D, 2D and video operations achieving a high level of performance in all application areas. The MGA-200's Symmetric Rendering Architecture provides excellent image quality without compromising performance. Applications that use features such as high-resolution triple buffering, extra bitmaps and texture storage especially benefit from SRA. Examples of such applications are again; D3D games, 3D design, video capture and video decompression.
The MGA-200 also is capable of performing Per Pixel Trilinear Filtering, an advanced texture filtering algorithm which provides image quality much higher than bilinear filtering and must be calculated on a per pixel basis. The trilinear filtering algorithm checks texel scale factors (the ratio between the size of the rendered texture versus the size of the source texture) on a per pixel basis and uses the scale factor to index into two adjacent MIP maps. It then takes four texels from each MIP, bilinearly filtering each set of four pixels and then averages the results based on scale factor.
The highest resolution source texture maps available are always used by the MGA-200 for rendering unless the application specifically provides lower resolution MIP maps to be used in conjunction with high resolution textures. This means that the fast 3D engine of the MGA-G200 can handle higher levels of performance without needing to downscale textures. This keeps the rendered image crisp and clean and reduces eyestrain created by blurry images.
The MGA-200 can also handle full scene anti-aliasing - a technique where all low resolution aliasing artifacts are removed keeping the rendered image smooth and realistic.
Additional MGA-200 Features
Sounds pretty hot right? But with so much depending on MGA-200's DualBus Architecture's correct use of AGP x2 mode, its little wonder that most of the complaints I received were regarding the user's ability to get the card to function in AGP x2 mode = with bus mastering difficulties running a close second.
With the release of upgraded overclocking applications like MOC (Matrox Overclock), the ability to get your Millennium or Mystique G200 into AGP x2 mode is a thing of the past but when all is said and done, I doubt that you'll see much difference in performance. You may however see the image quality take a hit if you run certain applications while forcing the G200 into AGP x2 mode. But here, see for yourself the differences generated in FutureMark's 3DMark 99 benchmark application, in AGP x1 and AGP x2 modes:
|G200 AGP x1 vs x2 Testing - 3DMark 99 Benchmark|
G200 AGP x1
G200 AGP x2
|3DMark Result (3DMarks)||1093.34||1093.07|
| CPU Geometry
|Rasterizer Score (3DRasterMarks)||761.58||785.3|
|Game 1 - Race - FPS||20.72||20.68|
|Game 2 - First Person - FPS||15.73||15.75|
|Fill Rate - MTexels/s||77.34||77.34|
|Fill Rate With Multi-Texturing - MTexels/s||77.84||77.83|
|2MB Texture Rendering Speed - FPS||122.89||123.06|
|4MB Texture Rendering Speed - FPS||87.52||93.57|
|8MB Texture Rendering Speed - FPS||44.39||53.91|
|16MB Texture Rendering Speed - FPS||29.95||37.94|
|32MB Texture Rendering Speed - FPS||22.46||27.86|
|Point Sample Texture Filtering Speed||108.72%||108.73%|
|Bilinear Texture Filtering Speed||100%||100%|
|Trilinear Texture Filtering Speed||78.24%||78.26%|
|Anisotropic Texture Filtering Speed||Not Capable||Not Capable|
|6 Pixel/individual - KPolygons/s||530.59||528.31|
|6 Pixel/strips - KPolygons/s||656.2||655.63|
|25 Pixel/individual - KPolygons/s||495.44||493.57|
|25 Pixel/strips - KPolygons/s||611.73||611.35|
|50 Pixel/individual - KPolygons/s||480.91||481.03|
|50 Pixel/strips - KPolygons/s||532.33||531.09|
|250 Pixel/individual - KPolygons/s||181.24||181.19|
|250 Pixel/strips - KPolygons/s||186.5||186.54|
|1000 Pixel/individual - KPolygons/s||58.26||58.16|
|1000 Pixel/strips - KPolygons/s||61.05||61.03|
|Processor Type||AMD K6-2||AMD K6-2|
|Processor Speed||350 MHz||350 MHz|
|L1 Cache size||64 KB||64 KB|
|L2 Cache size||512KB||512KB|
|Physical Memory||128 MB||128 MB|