Optimizing oneDAL and oneDNN for ARM CPUs to Accelerate AI Workloads

Learn about Fujitsu's efforts to optimize AI workloads on Arm CPUs in this 31-minute technical presentation. Explore the optimization strategies for oneDNN and oneDAL libraries, focusing on the development of FUJITSU-MONAKA, a high-performance, energy-efficient CPU for data centers. Discover how SVE implementation, OpenBLAS kernel enhancements, and batch-reduced GEMM-based JIT kernels for MatMul and Convolution primitives contribute to improved performance across multiple AI algorithms. Gain insights into the expansion of JIT kernel support for various ISAs, including A64FX and Graviton3E CPUs, developed as part of Japan's NEDO-funded next-generation green data centre initiative.