Ephrem Wu#
Ephrem is a Sr. Fellow at AMD. He has collaborated with many software and hardware teams to shape and enhance machine-learning accelerators. Since 2018, he has focused on improving Transformer efficiency, including GEMM scheduling, numerical algorithms, and speculative decoding. Ephrem joined AMD from Xilinx, where he spearheaded the development of the first 2.5D FPGA with analog transceivers and architected memory and compute units that enable design tools to automate the creation of low-latency systolic-array neural networks.
Posts by Ephrem Wu
