No description available.

Background. Modern game applications pressures hardware to its limits, and affects how graphics hardware and APIs are designed. In games, rendering geometry plays a vital role, and the implementation of optimization techniques, such as view frustum culling, is generally necessary to meet the quality expected by the customers. Failing to optimize a game application can potentially lead to higher system requirements or less quality in terms of visual effects and content. Many optimization techniques, and studies of the performance of such techniques exist. However, no research was found where the utilization of computational resources in the GPU, in state-of-the-art games, was analyzed. Objectives. The aim of this thesis was to investigate the potential problem of commercial game applications wasting computational resources. Specifically, the focus was set on the triangle data processed in the geometry stage of the graphics pipeline, and the amount of triangles discarded through clipping. Methods. The objectives were met by conducting a case study and an empirical data analysis of the amount triangles and entire draw calls that were discarded through clipping, as well as the vertex data size and the time spent on processing these triangles, in eight games. The data was collected using Triangelplockaren, a tool which collects the triangle data that reaches the rasterizer stage. This data was then analyzed and discussed through relational findings in the results. Results. The results produced consisted of 30 captures of benchmark and gameplay sessions. The average of each captured session was used to make observations and to draw conclusions. Conclusions. This study showed evidence of noteworthy amounts of data being processed in the GPU which is discarded through clipping later in the graphics pipeline. This was seen in all of the game applications included in this study. While it was impossible to draw conclusions regarding the direct impact on performance, it was safe to say that the performance relative to the geometry processed was significant in each of the analyzed cases, and in many cases extreme.

When developing applications in Direct3D today, developers can choose between using the fixed function pipeline and the programmable pipeline. The programmable pipeline is more flexible than the fixed function pipeline, but what is the price for high flexibility? Is high flexibility desired at any cost? How is the choice of pipeline affecting performance? The purpose of this master thesis is to evaluate the performance of the two pipelines. This will be achieved by developing a benchmark program, which measures performance when various graphical effects are tested. The results of the evaluation will hopefully help developers to decide which pipeline to use, in terms of performance. In the end we will see that the fixed function pipeline is faster than the programmable pipeline in all our tests.

This report documents implementation details, results, benchmarks and technical discussions for the work carried out within a master’s thesis at Linköping University. Within the master’s thesis, the field of software rendering is explored in the age of parallel computing. Using the Open Computing Language, a complete graphics pipeline was implemented for use on general processing units from different vendors. The pipeline is tile-based, fully-configurable and provides means of rendering visually compelling images in real-time. Yet, further optimizations for parallel architectures are needed as uneven work loads drastically decrease the overall performance of the pipeline.

One of the state-of-the-art highlights of deep learning in the past ten years is the introduction of generative adversarial networks (GANs), which had achieved great success in their ability to generate images comparable to real photos with minimum human intervention. These networks can generalise to a multitude of desired outputs, especially in image-to-image problems and image syntheses. This thesis proposes a computer graphics pipeline for 3D rendering by utilising generative adversarial networks (GANs). This thesis is motivated by regression models and convolutional neural networks (ConvNets) such as U-Net architectures, which can be directed to generate realistic global illumination effects, by using a semi-supervised GANs model (Pix2pix) that is comprised of PatchGAN and conditional GAN which is then accompanied by a U-Net structure. Pix2pix had been chosen for this thesis for its ability for training as well as the quality of the output images. It is also different from other forms of GANs by utilising colour labels, which enables further control and consistency of the geometries that comprises the output image. The series of experiments were carried out with laboratory created image sets, to pursue the possibility of which deep learning and generative adversarial networks can lend a hand to enhance the pipeline and speed up the 3D rendering process. First, ConvNet is applied in combination with Support Vector Machine (SVM) in order to pair 3D objects with their corresponding shadows, which can be applied in Augmenter Reality (AR) scenarios. Second, a GANs approach is presented to generate shadows for non-shadowed 3D models, which can also be beneficial in AR scenarios. Third, the possibility of generating high quality renders of image sequences from low polygon density 3D models using GANs. Finally, the possibility to enhance visual coherence of the output image sequences of GAN by utilising multi-colour labels. The results of the adopted GANs model were able to generate realistic outputs comparable to the lab generated 3D rendered ground-truth and control group output images with plausible scores on PSNR and SSIM similarity index metrices.

[pt] Este trabalho apresenta um novo sistema de renderização distribuída destinado ao uso em agrupamentos de PCs. É feita uma extensão à linha de produção gráfica convencional para uma linha de produção gráfica distribuída, que pelo uso de múltiplas linhas de execução permite paralelizar as operações feitas na CPU, na GPU e na rede que interliga os PCs do agrupamento. Este sistema serviu de base para a implementação e o teste de três arquiteturas para renderização distribuída: uma arquitetura com ordenação no início, uma arquitetura com ordenação no fim para renderização volumétrica e uma arquitetura híbrida que tenta combinar as vantagens da ordenação no início e da ordenação no fim. É apresentado um novo algoritmo de balanceamento de carga baseado nos tempos de renderização do quadro anterior. O algoritmo é de implementação muito simples e funciona bem tanto em aplicações com gargalo na geometria quanto em aplicações com gargalo na rasterização. Este trabalho também propõe uma estratégia de distribuição de trabalho entre os computadores de renderização do agrupamento que usa eficientemente os recursos gráficos disponíveis, melhorando assim o desempenho da renderização. Um novo algoritmo de partição paralela do modelo entre os computadores do agrupamento é proposto para a arquitetura híbrida. / [en] This work presents a new distributed rendering system destined for PC clusters. The conventional graphics pipeline is extended to a distributed pipeline that parallelizes the operations done on the CPU, the GPU and the network by using multiple threads. This system was the base for the implementation of three distributed rendering architectures: a sort-first architecture, a sort-last architecture for volume rendering, and a hybrid architecture that seeks to combine the advantages of both sort-first and sortlast architectures. A new load-balancing algorithm based on the rendering times of the previous frame is proposed. The algorithm is very simple to be implemented and works well for both geometry- and rasterization-bound models. A new strategy to assign tiles to rendering nodes is proposed which effectively uses the available graphics resources, thus improving rendering performance. A new parallel model partition algorithm is proposed for the hybrid architecture.