Uma arquitetura de motor de física para games 3D com processamento híbrido entre CPU e GPU e distribuição dinâmica de carga

• Autor(a) principal: Joselli, Mark Eirik Scortegagna
• Data de Publicação: 2007
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Federal Fluminense (RIUFF)
• Texto Completo: https://app.uff.br/riuff/handle/1/17891
• topic: Ciência da computação; Computação gráfica; Computação distribuída; Jogo tridimensional; Arquitetura de computador; Arquitetura para jogos digitais; Jogo em computador; Jogos digitais; Simulação física de corpos rígidos; Motor de física; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO::COMPUTABILIDADE E MODELOS DE COMPUTACAO

The realism in 3D games, virtual reality and simulations are trespassing by a fast evolution, requiring each time more power from the computer processing. The physics engines, necessary for giving physics reality, require a lot of this processing because of their characteristic of having high complex mathematics calculation. With the power increase of the programmable GPUs (Graphics Processing Unit) this calculation can be also processed in this graphics board. This work presents a new physics engines that has some of its calculations processed in both the CPU and GPU, using an architecture without precedent. This engine was called GDE (GPU Dynamics Engine). The processing of mathematical calculation on the GPU is highly optimized for a higher number of calculation because of the parallel structure of the GPU, allowing a better performance of the GDE in the CPU when there are few number of bodies and a better performance in the GPU for higher number of bodies. This work also presents a specific architecture for games, to be used with the GDE together with the framework GUFF (Games UFF). The GDE has implementations in both CPU and GPU. Because GPU has a better processing in some case and the application can share the processors with the system and others applications, it has been developed heuristics to automatic distribution of computation between CPU and GPU. This research has no known precedence