本文探讨了CUDA编程模型的局限性，如数据类型限制、显式编程需求及跨CPU-GPU的数据处理挑战等，并介绍了ModernGPU项目，这是一个旨在促进GPU计算创新思维的库、算法指南和最佳实践集合。
http://on-demand.gputechconf.com/gtc/2013/webinar/opencv-gtc-express-shalini-gupta.pdf
PPT作者：Shalini Gupta, Shervin Emami, Frank Brill -NVIDIA
Modern GPU：http://nvlabs.github.io/moderngpu/
Modern GPU is code and commentary intended to promote new and productive ways of thinking about GPU computing.
This project is a library, an algorithms book, a tutorial, and a best-practices guide.
CUDA Disadvantages
Only 250 functions
Limited data types
GPU: 8-bit & 32-bit grayscale
CPU: +16-bit (HDR) & 32-bit color, ROI
Explicitly program for CUDA
Handle data transfers between CPU and GPU
Only on NVIDIA GPU
Some serial operations not sped up, e.g., Canny()
CUDA has startup delay
CUDA Start Up Delay
First CUDA call initializes CUDA module
Typical first call – CPU to GPU transfer (~2000ms and 1ms after that)
Affects single frame applications, videos OK
扫一扫
801
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
