Session

DescriptionThis session is about hardware accelerator system designs and optimizations for DNNs. As deep learning models become larger and larger, the quest to find the most energy efficient, and throughput optimized hardware platforms to perform model tasks is very important. Here in this session, many hardware platforms are covered. Hydra is a mixture-of-expert inference system using chiplets, SynGPU does its optimizations on GPU. We have 3 other papers that explore many-core architectures. Finally, paper SAGA focuses its optimizations on memory.