标题: GPU AI - Technology Preview [打印本页] 作者: jocover 时间: 2010-6-4 07:17 标题: GPU AI - Technology Preview This technology preview is a snapshot of some internal research we have been working on and talking about at various conferences for the past couple years. The level of interest in GPU-accelerated AI has continued to grow, so we are making this (unsupported) snapshot available for developers who would like to experiment with the technology.
The GPU accelerated path planning software provided in this technology preview is applicable to domains that include robotics, video games, synthetic environments (SE) and artificial intelligence in general. Systems that exploit multi-agent modeling for simulating the safe motion of virtual multi robots, non-player game characters and virtual humans navigating in a dynamic environment are just a few examples of applications that benefit from GPU-accelerated AI.
Please share your experience and feedback on this GPU AI technology in our developer forums.
Features supported in this release:
* Roadmap Construction: generates a graph of collision-free paths for a given game level or configuration space based on an input 3D mesh
* Path Searching: Finds optimal path for each agent from start to goal across the roadmap graph using either A* or Dijkstra search
* Multi-agent Collision Avoidance: Consults roadmap to intelligently guide each agent from start to goal while avoiding collisions with other dynamic agents as well as static and moving obstacles
Benefits of GPU AI:
* Faster than CPU AI: exploits GPU parallelism in every phase of the simulation, using 10s of thousands or even hundreds of thousands of GPU threads
* Offload the CPU: free up CPU resources for other work
* Flexible Roadmap Generation: roadmaps can be pre-computed offline or generated/updated during runtime
* Widely supported: works on 100s of millions of GPUs, anything with compute capability 1.1 or later
http://developer.nvidia.com/object/gpu-ai.html作者: akiragtx 时间: 2010-6-4 08:42
提示: 作者被禁止或删除 内容自动屏蔽作者: goldman948 时间: 2010-6-4 09:04
以后cpu的工作就是运行windows以及显卡驱动了在游戏方面作者: gz_easy 时间: 2010-6-4 10:44
我觉得现在的AI运算应该是逻辑转换比较多,CPU比较适合做,GPU可以大大加速可能需要浮点运算的数学模型,也许要看今后AI中运算类型决定由谁来做,不知道该不该叫做混合型AI运算。
此外文中提到"using 10s of thousands or even hundreds of thousands of GPU threads",但毕竟动用GPU资源,外加GPU PhysX开销,也许会不堪重负。作者: Edison 时间: 2010-6-4 12:10
如果是基于物理的 AI,GPU 还是比比较适合的。作者: goodayoo 时间: 2010-6-4 12:24
对于AI,还是CPU的强项,也是CPU最后的阵地,如果这块给GPU占领的话,CPU确实就不重要了,不过编程方法和环境才是GPU通用计算面临最大的问题,世界上99%的程序都是运行在CPU上的,如果改用GPU,问题很复杂。作者: gzpony 时间: 2010-6-4 13:13