4 ms·
In my text, I said "cannot easily". When you consider the general search problem, it seems so easy to parallel. But to adapt the search process to thousands of
by liuliu 18y ago
In my text, I said "cannot easily". When you consider the general search problem, it seems so easy to parallel. But to adapt the search process to thousands of threads, the cost to avoid repeat search is high because you have to maintain a list of taboo area. By carefully design, you may avoid that problem. But here, I am talking about extra-time cost and not intend to say that it is impossible.
- Retric 18y agoModern game programmers still only need to split the AI over 8 or so cores to take advantage of "Multi processor" systems. Unless you are running the AI off the GPU. Getting that to work well can be as simple as separating a complex evaluation of each node across a different CPU. It's not going to be 4x as fast on 4 CPU's but there would still be a significant boost. Or with multiple AI's just give each of them a core and move on. Now if your talking about 10,000 node clusters it becomes a much harder problem but efficiently doing most things on a huge cluster is hard.
- liuliu 18y agoYes, you get my point. The graphics can take advantage of GPU, but the AI merely can.