Performance analysis of a 240 thread tournament level MCTS Go program on the Intel Xeon Phi

摘要: In 2013 Intel introduced the Xeon Phi, a new parallel co-processor board. The Phi is cache-coherent many-core shared memory architecture claiming CPU-like versatility, programmability, high performance, and power efficiency. first published micro-benchmark studies indicate that many of Intel's claims appear to be true. current paper study on complex artificial intelligence application. It contains an open source MCTS application for playing tournament quality Go (an oriental board game). We report speedup figures up 240 threads real machine, allowing direct comparison previous simulation studies. After substantial amount work, we observed performance scales well 32 threads, largely confirming results this program, although surprisingly deteriorates between threads. Furthermore, (1) unexpected anomalies CPU small problem sizes numbers (2) sensitive scheduling choices. Achieving good programs not straightforward; it requires deep understanding search patterns, scheduling, (3) its cores caches. practice, less straightforward program than originally envisioned by Intel.