The Chipset

GLINT R3 Rasterization Processor

The Oxygen GVX1 features 2 processors: one for the rasterization process, and one for geometry. The GLINT R3 is the first of the 2 processors. It controls the rasterization process, which is the process of converting a polygonal 3D image, stored in the frame buffer, into the image onscreen that is built from pixels, which includes lighting and textures. The GLINT R3 texture unit has capability for single-pass multitexturing, which is simply applying multiple textures on to a polygon's surface to achieve the illusion of a realistic surface. This feature is nothing new to us in the gaming world, as the newer chipsets provide the similar ability to apply multiple textures per clock cycle.

The integrated RAMDAC that the GLINT R3 uses is a 300MHz one, allowing for speedy conversion and transfer of signals to our CRT monitor. The Oxygen GVX1 gives us a maximum resolution of 2048x1536x32 at 60Hz. Not bad! Additionally, there is a Digital Flat Panel connector which gives us the capability of connecting to a DFP, at a resolution of up to 1280x1024. A final feature is the ability to use up to 8 monitors in conjunction with more AGP and PCI Oxygen VX1 and GVX1 cards. Finally, the ability to use multiple AGP and PCI Oxygen GVX1 only, also for multi-head support, is planned to be out soon.

GLINT Gamma G1 Geometry Processor

The GLINT Gamma G1 Geometry processor provides a function that is not new to the higher-end market that the Oxygen GVX1 is actually intended to serve. It is a geometry processor, which acclerates transform and lighting operations in hardware. This takes tremendous load off of the CPU as far as 3D operations go.

You see, today's gaming cards rely on the CPU's floating point unit to calculate the algorithms necessary to position 3D objects. Because of this, we see that with newer video cards, the CPU is often the bottleneck for the video subsystem, and that by upgrading the CPU to a faster one, we get better 3D performance from our video cards. These calculations are aided by special additional instructions that can be added to software to optimize them for specific processors. For AMD, we have 3DNow!, and for Intel we have SSE. These specific instruction sets allow for SIMD, which stands for Single Instruction Multiple Data. What this means is applying a specific operation to multiple sets of data at the same time, which speeds up overall calculations.

Why are you telling me all this?

The implications of this, if it could be cost-effectively applied into the gaming arena, are huge. It would mean that running a slow CPU and running a fast CPU in conjunction with a 100% hardware transform and lighting video card would show very similar performance in 3D. The CPU almost becomes a non-factor in the video subsystem as all of the 3D calculations are done onboard the video card. The CPU could have effects on other parts of the application, however, but the video subsystems would perform equally. Nowadays, we know that this is not the case. A TNT2 on a PII 266 versus a PIII 500 would give different numbers for benchmarks!

The GLINT Gamma G1 also has the capability to display 16 simultaneous light sources. This, combined with its geometry processing, results in a total 4.75 million lit and transformed triangles a second. In contrast, the Permedia 3 chipset, also a 3dlabs product, does ~2 million triangles a second when used with a PIII 500. (Since the Permedia 3 is not hardware accelerated, transform and lighting operations are CPU dependent) To learn more about transformation and lighting, as well as a whole slew of other hot new video technologies, check out our Next Generation 3D Accelerators article!