Performance modeling on DaVinci AI core

doi:10.1016/j.jpdc.2023.01.008

Performance modeling on DaVinci AI core

2023-05-01

4

PRE

AI

Y

Yifeng Tang *

DOI：10.1016/j.jpdc.2023.01.008

原文链接

原文求助

摘要

En 中文

The extensive use of Deep Neural Networks (DNNs) encourages people to design domain-specific hard-ware called Artificial Intelligence (AI) processors. The novel hardware makes optimizations challenging without a proper performance model that reveals working details and performance implications. This pa-per presents a performance model, Verrocchio, for Huawei DaVinci AI Core, which predicts the execution time of real-world DaVinci kernels. We propose specially-crafted micro-benchmarks to identify con-tention source, runtime behaviors, and bandwidth sharing, which significantly determine performance. Since DaVinci Core adopts a binary semaphore mechanism for synchronization, Verrocchio views each instruction as a discrete event and manages its execution time based on the programming logic. For eval-uation, Verrocchio achieves average error rates of 2.62% and 2.30% in sample kernels for single-core and double-core execution. We demonstrate an optimizing process of matrix multiplications with Verrocchio, achieving speedups of 1.70x for operators and 1.53x for applications and error rates of 5.06% and 5.25%.(c) 2023 Elsevier Inc. All rights reserved.

Keyword:

AI processors

Benchmarking

Performance modeling

Performance modeling on DaVinci AI core

摘要

期刊

机构

引用论文