EFFICIENT PARALLEL ALGORITHM FOR SUPER HIGH SPEED COMPUTER GRAPHICS RENDERING USING GPGPU

DOI Web Site Web Site 1 Citations 15 References

Bibliographic Information

Other Title
  • GPGPUを活用した超高速CGレンダリングのための効率的並列化アルゴリズム
  • GPGPU オ カツヨウ シタ チョウコウソク CG レンダリング ノ タメ ノ コウリツテキ ヘイレツカ アルゴリズム

Search this article

Abstract

<p> Computer Graphics are widely used as a powerful tool for design study at architectural design field. To visualize planning buildings by Computer Graphics at designing stage before its construction helps to enhance design quality and shorten designing term. On the other hand, Computer Graphics technologies become increasingly sophisticated and computation time to generate images goes on increasing accordingly. It is not uncommon to take tens of minutes to generate one image. Increase of rendering time prevents frequent use of Computer Graphics. Instant image generation is necessary to utilize Computer Graphics at design stage and dramatic improvement of rendering speed is highly required by architectural designer.</p><p> Calculation speed capability of CPU, that is heart of personal computer, currently faces limit through previous drastic progress and major progress of CPU speed can no longer be expected. In such condition of computer GPU (Graphic Processing Unit) recently attracts attention. GPU is less flexible than CPU on application but has thousands of cores for parallel computing. If we can take advantage of the characteristics of GPU, there is the big potential of high performance calculation to overtake CPU capability.</p><p> We pursue developing a practical super high speed Computer Graphics rendering system whose calculation speed is more than hundreds times faster than ordinal rendering systems by CPU calculation, and going to provide the system through internet for designers to use high end Computer Graphics images quickly and easily at design study. This paper proposes an efficient parallel calculation method to extract high speed characteristics of GPU, and judges effect of the proposal method by implementing the method to rendering software and timing its calculation speed to generate images to confirm its practicality.</p><p> GPU, namely parallel processing, has advantage at same calculation huge times, but has weak point at branched operation. It is basically difficult for Computer Graphics rendering software, that is constructed by huge branched operations for various rendering functions, to get benefit of high speed performance by GPU. Therefore we propose a hybrid rendering method of CPU calculation and GPU calculation. We extract small process of heavy load calculation from total rendering process, and apply GPU calculation to the heavy load process only. By calculating the rest using CPU huge branched operations by GPU is avoided and we can get benefit of GPU high speed calculation.</p><p> Furthermore, for efficient calculation using multi GPUs we propose equal load distribution method to get rid of variation of calculation load at each GPU and finish calculation simultaneously at all GPUs.</p><p> We implemented these proposed method for GPU calculation to CPU version of global illumination rendering software based on path tracing method. We timed rendering calculation of several test images by the GPU version of the software with a server system consisted of 7 GPUs, and got result of faster hundreds times to thousands times than CPU version of the same software.</p>

Journal

Citations (1)*help

See more

References(15)*help

See more

Related Projects

See more

Keywords

Details 詳細情報について

Report a problem

Back to top