A number of molecular dynamics, astrophysics, and similar simulation applications are scalable enough to use 8 GPUs per node. Especially when multiple nodes ar

A number of molecular dynamics, astrophysics, and similar simulation applications are scalable enough to use 8 GPUs per node. Especially when multiple nodes are involved, these scientific applications are in general more mature compared to deep-learning frameworks, which have just started supporting multiple nodes.