The optimal image partition searching algorithm for the sort-first parallel rendering선정렬 병렬 렌더링을 위한 최적 이미지 분할 검색 알고리즘

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 437
  • Download : 0
According to recent commodity GPU improvements, applications for multiple GPU and Cluster Graphics have started to spread widely. Basically, the commodity based cluster graphics systems use the off-the-shelf components; but there has been little considerations within the pipeline to combining intermediate results between GPUs. When it comes to combining intermediate results for efficient interactive or realtime rendering, the rendering performance will probably drop without further architectural support or improvement. One option to bypass such communication is to process pixel areas with a single dedicated processor; this is one of the key characteristics of sort first architecture. For the commodity based cluster graphics systems which do not have the high bandwidth necessary to interchange intermediate results of GPUs, sort-first architecture has been widely selected as a parallel rendering architecture. Such processor isolation provides architectural advantages, but also incurs certain disadvantages. One advantage is to eliminate intermediate exchange results such as image space projected polygons or depth buffers. On the other hand, one disadvantage is the fact that such a system faces load imbalance because partition algorithm did not divide graphics primitives but image area. Another disadvantage arises because it has to divide image space for processors. Primitives happen to be projected on the pixel area of more than one processor and cause inefficiently redundant processes on more than one processor. Such overlapped primitives become a major reason for performance inefficiency. According to recent GPU improvements, the performance gap between CPUs and GPUs has increased. If individual primitives are selected as work units, a typical CPU could be unable to keep pace with the GPU processing speed. Therefore, a chunk or group of primitives is selected as a rendering unit by recent realtime rendering algorithms and applications. When relatively la...
Advisors
Wohn, Kwang-Yunresearcher원광연
Description
한국과학기술원 : 전산학과,
Publisher
한국과학기술원
Issue Date
2012
Identifier
568099/325007  / 020025299
Language
eng
Description

학위논문(박사) - 한국과학기술원 : 전산학과, 2012.2, [ vii, 69 p. ]

Keywords

Image partitioning; 클러스터 그래픽스; 병렬 렌더링; 선정렬 구조; 그래프 분할법; 이미지 분할법; Graph partitioning; Sort-first architecture; Parallel Rendering; Cluster Graphics

URI
http://hdl.handle.net/10203/197812
Link
http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=568099&flag=dissertation
Appears in Collection
CS-Theses_Ph.D.(박사논문)
Files in This Item
There are no files associated with this item.

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0