深度神经网络硬件基准测试现状及发展趋势

Deep neural network hardware benchmark status and development trend

基准测试作为一种客观的评价方式,在计算机体系架构的发展中扮演着重要的角色,有效地推动着面向不同方向的硬件和软件设计的演进。针对当前深度神经网络处理器的应用场景、发展现状和与通用处理器基准测试差异性进行了分析,给出基准测试参考架构设计,并对深度神经网络处理器基准测试的重要性与所面临的挑战进行了分析。

As an objective evaluation method, benchmarking plays an important role in the development of computer architecture, and effectively promoting the evolution of hardware and software design for different directions. In this paper, the application scenarios, development status and differences between general processor benchmarks of the deep neural network(DNN) processor were analyzed. The reference test architecture design of DNN processor was proposed. And the importance and challenges of DNN processor benchmarking were analyzed.