Rendering, the calculation and coloration of individual
pixels based on the geometry and surface of an object, has existed for
some time now. First results of the Raytracing method were already
described in the 1960's.
However, it became clear at an early stage that this process would place
substantial demands on the hard- and software employed. In particular, large
and complex constructions lead to a rapid deterioration in overall system
In the real world we are dealing with photons, which once emitted from
a light source are either reflected on, refracted off or absorbed by different
surfaces. As the geometry increases, the computational demands rise enormously
in order to accurately define the surfaces which contribute to the color
of an individual pixel.
The initial innovation lies in the algorithm used to execute a parameter
examination very quickly. Without such an algorithm, it is impossible
to use raytracing efficiently.
The algorithm has already been integrated into
software and assists this software in achieving a 10 to 100 fold increase
in performance in comparison to existing software.
At this juncture it is important to note that microprocessor performance
only doubles every 1,5 years. This means that only a 10 fold performance
increase is necessary to catapult the market lead of AVALON (Software-Version)
to five years. By corollary, it could similary be remarked that AVALON-software
(our prototype) will
always enjoy a substantial lead, since it will automatically benefit from further development.
Although the prototype-software is already very powerful the performance achieved
is not adequate to render images in realtime. The use of the word realtime
is in reference to the PAL television standard, which requires the generation
of 25 full frames per second in order to offer seamless motion to the viewer.
In order to use rendering functions for games, flight simulators, or
similar products, it is necessary to refresh the projected image every
40 milliseconds at a minimum, so that the observer can enjoy an uninterupted
flowing scene. This corresponds to the 25 frames per second realtime representation.
At the moment the software version of AVALON can calculate one frame
in a few seconds. By using the AVALON chip many processes which can only be
insufficiently implemented in software can easily be turned into parallel
tasks in hardware. This allows much more powerful computation elements
than todays most advanced microprocessors (i.e. the Pentium III or Athlon
– PC CPUs of the newest generation).
The chip design is a further innovation. It involves not simply porting
the software to a chip, but also dividing it into multiple small units
which may then be analysed, optimised and mapped onto logic circuits. Such
a complex rendering chip has not yet been produced. Estimates put the speed
of the processor at approx. 500GOps (500 billion operations per second).
This immense power is equivalent to roughly 300-500 parallel Pentium III
500 processors (using SIMD instructions) !
This is only possible due to the dedicated logic circuits used for the AVALON
graphics chip. Since it is known which computations can be completed
independently, it is possible to process these in parallel.
This is the stage
where architecture comes in. The architectual conception is key in this design.
Based on our concepts the architecture is feasible. We have proved it in a half
year technological study made by the Rheinisch Westfälische Technische
Hochschule Aachen (RWTH Aachen) [06/2000 - 12/2000].
Unnecessary to say that
this concepts were not easy to find. A lot of hurdles were to manage and some
issues are still waiting. But we are confident to solve them too.