CPU and GPU palyginimas vykdant sablonu atitikties algoritma
Image processing, computer vision or other complicated optical information processing algorithms require large resources. It is often desired to execute algorithms in real time. It is hard to fulfill such requirements with single CPU processor. NVidia proposed CUDA technology enables programmer to u...
Gespeichert in:
| Veröffentlicht in: | Science future of Lithuania Jg. 6; H. 2; S. 129 |
|---|---|
| Hauptverfasser: | , |
| Format: | Journal Article |
| Sprache: | Englisch |
| Veröffentlicht: |
Vilnius Gediminas Technical University
01.04.2014
|
| Schlagworte: | |
| ISSN: | 2029-2341 |
| Online-Zugang: | Volltext |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Zusammenfassung: | Image processing, computer vision or other complicated optical information processing algorithms require large resources. It is often desired to execute algorithms in real time. It is hard to fulfill such requirements with single CPU processor. NVidia proposed CUDA technology enables programmer to use the GPU resources in the computer. Current research was made with Intel Pentium Dual-Core T4500 2.3 GHz processor with 4 GB RAM DDR3 (CPU I), NVidia GeForce GT320M CUDA compliable graphics card (GPU I) and Intel Core I5-2500K 3.3 GHz processor with 4 GB RAM DDR3 (CPU II), NVidia GeForce GTX 560 CUDA compatible graphic card (GPU II). Additional libraries as OpenCV 2.1 and OpenCV 2.4.0 CUDA compliable were used for the testing. Main test were made with standard function MatchTemplate from the OpenCV libraries. The algorithm uses a main image and a template. An influence of these factors was tested. Main image and template have been resized and the algorithm computing time and performance in Gtpix/s have been measured. According to the information obtained from the research GPU computing using the hardware mentioned earlier is till 24 times faster when it is processing a big amount of information. When the images are small the performance of CPU and GPU are not significantly different. The choice of the template size makes influence on calculating with CPU. Difference in the computing time between the GPUs can be explained by the number of cores which they have. Keywords: image processing, GPGPU, template matching, CUDA. Vaizdu apdorojimas, kompiuterine rega ir kiti sudetingi algoritmai, apdorojantys optine informacija, naudoja didelius skaiciavimo isteklius. Daznai siuos algoritmus reikia realizuoti realiuoju laiku. Si uzdavini isspresti naudojant tik vieno CPU (angl. Central processing unit) pajegumus yra sudetinga. nVidia pasiulyta CUDA (angl. Compute unified device architecture) technologija leidzia panaudoti GPU (angl. Graphic processing unit) isteklius. Tyrimui atlikti buvo pasirinkti du skirtingi CPU: Intel Pentium Dual-Core T4500 ir Intel Core I5 2500K, bei GPU: nVidia GeForce GT320M ir NVidia GeForce 560. Tyrime buvo panaudotos vaizdu apdorojimo bibliotekos: OpenCV 2.1 ir OpenCV 2.4. Tyrimui buvo pasirinktas sablonu atitikties algoritmas. Algoritmui realizuoti reikalingas analizuojamas vaizdas ir ieskomo objekto vaizdo sablonas. Tyrimo metu buvo keiciamas vaizdo ir sablono dydis bei stebima, kaip tai veikia algoritmo vykdymo trukme ir vykdomu operaciju skaiciu per sekunde. Is gautu rezultatu galima teigti, kad apdorojant dideli duomenu kieki GPU realizuoja algoritma iki 24 kartu greiciau nei tik CPU. Dirbant su nedideliu duomenu kiekiu, skirtumas tarp CPU ir GPU yra minimalus. Lyginant skaiciavimus dviejuose GPU, pastebeta, kad skaiciavimu sparta yra tiesiogiai proporcinga GPU turimu branduoliu kiekiui. Musu tyrimo atveju spartesniame GPU ju buvo 16 kartu daugiau, tad ir skaiciavimai vyko 16 kartu sparciau. Reiksminiai zodziai: vaizdu apdorojimas, bendrosios paskirties GPU, sablonu atitiktis, CUDA technologija. |
|---|---|
| ISSN: | 2029-2341 |
| DOI: | 10.3846/mla.2014.16 |