Parallel Graphic Processing
"When will be able to generate in real-time an animated film such as 1998's A Bug's Life
by
Pixar? The short answer is decades. Generating an average frame of A Bug's Life
takes three to four hours on a 330MHz Ultra SPARC, and requires about a gigabyte of
geometric data alone, as 1998. To reach a minimum real-time rate of 12 frame per
second, this means a speed-up of about 150,000 times is needed. Moore's Law give
an acceleration rate of 10 times every 5 years; this puts real-time A Bug's Life
at around the year 2024"
From the book "Real-Time Rendering" by T.Moeller and E.Haines.
But interactive ray-tracing of complex scenes is possible today. For example,
a parallel system with 60 CPUs, can be render a scene with 35 million spheres at
15 frame per second.
Our experience about parallel graphic processing deals with two problems; the first is
a classic parallel processing objective applied to graphic algorithms, that is
speeding-up the process. In other words, speeding-up their computational performance.
The second problem, more particular, consider that in the case of graphic-numerical
processing, whose output is represented by numerical results and drawings, the host
computer cannot be a remote graphics workstation with increase in the cost of
communication with the system.
So the research for efficient tchniques to make graphics calls by the parallel system
assumes considerable importance. In this situation we need introduce a new metrics
evaluation of the performances that keeps count of the graphic visualization phase.
For more detail obout our studies and results in this area see the references.
As a remark, we underline that a distributed graphic processing, is simply able
using the well know Xwindow System. In fact it allows us to write Client applications
that run on different computers and using the TCP/IP and Xprotocol can directly
draw on a display of a remote workstation that must have the X-Server software.
References
-
G.Casciola, R.Rossi, Valutazione di un sistema multiprocessore a
ipercubo: l'Intel iPSC/2, Dipartimento di
Matematica dell'Universita` di Bologna, (1990).
-
G.Casciola, R.Rossi, Un algoritmo parallelo di ray
tracing per ipercubo Intel iPSC/2: strategie adottate e valutazione,
Atti del convegno internazionale ICO GRAPHICS '91, (1991).
-
G.Casciola, S.De Santis, R.Quadalti, Algoritmi paralleli per il problema
NURBS Surface/Surface Intersection, Dipartimento di
Matematica, Universita` di Bologna, (1992).
-
G.Casciola, R.Rossi, Guida all'utilizzo e alle prestazioni
dell'ipercubo Intel iPSC/ 860, Progetto
Finalizzato Sistemi Informatici e Calcolo Parallelo, N.1/138, (1993).
-
G.Casciola, S.Morigi, Graphics in parallel computation for
rendering 3D modelled scenes, Parallel Computing, 21 (1995)
-
G.Casciola, S.Morigi, Un sistema distribuito per la modellazione solida
sculturata, Giornate di Studio sul Calcolo Parallelo, Lucca (1996)
-
Green, Stuart, Parallel processing for computer graphics,
Pitman: The Mit Press, (1991).