Deep Reinforcement Learning on SX-Aurora

wellvill Co.,Ltd. has evaluated their deep reinforcement learning algorithm on SX-Aurora’s Vector Engine. The result is very impressive. SX-Aurora outperforms GPU(P100) system about two times.

Their algorithm is based on “dueling double DQN” and solves 3D bin packing problems such as packing multiple products in boxes in a logistics center, loading packages into a truck, etc.

3D bin packing

Since they are using TensorFlow, it can run on VE with TensorFlow for VE without special modification.

result

This graph shows elapsed time for 90000 steps. SX-Aurora(1VE) is 1.9 times faster than GPU P100 server (2:54 vs 5:26). We do not know the reason of this big difference since we have not looked deeply. But we think this result is important in the viewpoint that this evaluation was done by a real user, not a specialist of performance tuning. Our partner is thinking that this speedup helps their activities. We hope that SX-Aurora contributes to improve workflow of algorithm developers.


See also