Search papers, labs, and topics across Lattice.
Peking University
1
0
3
LLM decoding on 3D-stacked memory shifts the bottleneck from memory bandwidth to compute, demanding a rethink of compute microarchitecture.
