Answer to RQ1a for FUEGO Sample Clauses

Answer to RQ1a for FUEGO. ‌ In this subsection we answer RQ1a for FUEGO. We repeat RQ1a below. • RQ1a: Can thread-level parallelization deliver a comparable performance and scalability for many-core machines compared to multi-core machines for parallel MCTS? Using FUEGO, which uses thread-level parallelization for implementing the Tree Parallelization algorithm, we have found in Subsection 3.1.3 that we cannot reach the same performance on the Xeon Phi as on the Xeon CPU. The maximum performance in terms of PPS for Tree Parallelization on the Xeon CPU is around three times more than the one on the Xeon Phi (see Figure 3.5). Moreover, the scalability of the Tree Parallelization algorithm in terms of PPS is better on the Xeon CPU (for up to 32 threads) than the Xeon Phi (for up to 240 threads). Our experiments show that the performance of the algorithm drops after 128 threads on the Xeon Phi (see Figure 3.5). For the performance in terms of PW, the Xeon CPU shows a steadily decreasing PW (see Figure 3.6a), as expected, where the Xeon Phi shows a hump-like shape (see Figure 3.6b). Hence, our answer to RQ1a reads: with thread-level parallelization we cannot reach the same performance of a multi-core machine on a many-core machine.