“Asynchronous BVH Reconstruction on CPU-GPU Hybrid Architecture” by Kim, Kim, Lee and Han
Conference:
Type:
Entry Number: 14
Title:
- Asynchronous BVH Reconstruction on CPU-GPU Hybrid Architecture
Presenter(s)/Author(s):
Abstract:
We present a CPU-GPU hybrid architecture with an extended asyn- chronous bounding volume hierarchy(BVH) construction scheme [Ize et al. 2007] for ray tracing dynamic scenes. Because the perfor- mance of a ray tracer greatly depends on the acceleration data struc- ture, it is necessary to efficiently update the AS of dynamic scenes. To achieve this, Asynchronous BVH construction simultaneously performs both BVH reconstruction and bounding volume(BV) re- fitting. However, it may degrades the overall performance because it exploits the CPU to simultaneously perform both the BVH con- struction and rendering. To solve this problem, we improve this multi-core-CPU-based scheme by utilizing a CPU-GPU hybrid ar- chitecture. This hybrid approach is implemented based on [Aila and Laine 2009]. It improves the performance by 254~303% compared to a key-frame renderer implemented on NVIDIA OptiX.
Additional Information:
The experimental results is used as a comparison data in [Nah et al. 2014]. The detailed experimental results are included in our technical re- port(http://msl.yonsei.ac.kr/TR/msl2014-01.pdf).
References:
- Aila, T., and Laine, S. 2009. Understanding the efficiency of ray traversal on gpus. In Proceedings of the Conference on High Performance Graphics 2009, ACM, New York, NY, USA, HPG ’09, 145–149.
- Ize, T., Wald, I., and Parker, S. G. 2007. Asynchronous bvh construction for ray tracing dynamic scenes on parallel multi-core architectures. In Proceedings of the 7th Eurographics Conference on Parallel Graphics and Visualization, Eurographics Association, Aire-la-Ville, Switzerland, Switzerland, EG PGV’07, 101–108.
- Nah, J.-H., Kim, J.-W., Park, J., Lee, W.-J., Park, J.-S., Jung, S.-Y., Park, W.-C., Manocha, D., and Han, T.-D. 2014. HART: A hybrid architecture for ray tracing animated scenes. submitted to IEEE TVCG.
