***Gamma: 3D Labs' New Geometry Engine
by John Latta and David Lohse
3D Labs this week unveiled GLINT Gamma, an AGP-compliant second-
generation 3D geometry pipeline. Gamma will accelerate the complete 3D
geometry pipeline in a single chip by providing 1 GFLOP of floating-
point power dedicated to 3D processing, resulting in reported maximum
throughputs of 3.3 m meshed polygons/sec with shading, transforms,
lighting and backface culling enabled. The GLINT Gamma is expected to
support 100% of the geometry and lighting functionality of OpenGL 1.1.
GLINT Gamma-enabled boards are expected to start shipping during the
second half of 1997.
WAVE recently interviewed Neil Trevett, the VP of Marketing for 3D
Labs, on his views on the importance of Gamma.
-->WAVE: You have described that there needs to be an education process
on the role of GLINT Gamma - what is that process?
-->Neil Trevett: Many 3D applications on the PC are performance-limited
by the geometry pipeline running in software on the host. We believe
offloading the complete geometry and lighting pipeline into the GLINT
Gamma silicon is going to have a more significant impact on application
performance than anything we do to speed up pixel fill-rates. We are
working with the 3D board vendor community and the 3D software
development community to make them aware of what today's performance
bottleneck actually is, how to fix it with GLINT Gamma, and what that
is going to mean in terms of application-level performance on the PC.
-->W: Why cannot microprocessors fulfill the role of GLINT Gamma with
the increasing performance of the Intel processors?
-->NT: Although Intel's processors are getting faster, they continue to
be a long way behind what is needed to extract the best performance
from today's 3D rasterization chips. Because we use hardwired, non-
programmable logic to implement GLINT Gamma, it can easily provide both
more performance and better price/performance than any general-purpose
programmable CPU. Even if host CPUs could approach the performance
level of GLINT Gamma-type devices in the future you will still benefit
by off loading geometry processing into cost-effective hardware -
freeing the CPU to process the application.
-->W: What makes GLINT Gamma a significantly different product from
Delta?
-->NT: GLINT Delta is a 100 MFlop processor that offloads setup
processing from the host, to boost achievable geometry performance on
Pentium Pro-class machines from about 300K to 1 million polygons per
second - provided that 100% of CPU cycles are applied to the geometry
pipeline. GLINT Gamma is pin-compatible with GLINT Delta, but is a
1,000 MFlop processor that implements the complete 3D transform,
lighting and setup pipeline to boost geometry performance to 3.3
million polygons per second. The difference between GLINT Delta and
GLINT Gamma is not only the increased geometry performance, but also
that the increased performance is achieved with no host load - freeing
the host to run the application.
-->W: What vertical markets is Gamma targeted to?
-->NT: GLINT Gamma will speed-up any 3D application using any standard
API - such as OpenGL, Direct3D, Heidi or QuickDraw 3D. However GLINT
Gamma will have the most benefit for applications that use many
polygons and so are heavily geometry-bound. CAD uses many small
polygons and is one application area that will benefit greatly from
GLINT Gamma. GLINT Gamma will be more expensive than GLINT Delta and
so it will be used on the higher-end GLINT boards for the professional
markets such as CAD and animation authoring.
-->W: How will the ViewPerf benchmarks improve with Gamma?
-->NT: Today's Viewperf figures for a GLINT Delta-enabled board are
around the mid-twenties for CDRS-03 on high performance hosts. We
expect that GLINT Gamma will boost CDRS-03 performance to the mid-
forties. This figure will be achievable even on less expensive hosts -
because the CPU is now doing nothing but composing the lists of
polygons to be rendered.
-->W: Why is AGP important to Gamma?
-->NT: Even with hardware setup processing such as GLINT Delta, which
eliminates the need to pass slope information across the backplane to
the graphics subsystem, the limit in performance for 33 MHz PCI is
about 1.5M polygons per second. GLINT Gamma will be the first
professional AGP device, and will use AGP to increase the performance
of 3D on the PC - rather than reducing cost. With AGP we will be able
to realize the full 3.3M polygons/sec performance of GLINT Gamma. We
have also found that AGP's sideband addressing to be essential for
efficient processing of 3D vertex array data structures by GLINT
Gamma's intelligent DMA controller.
www.3dlabs.com
Wave Issue 9706 3/28/97 Article 5-01