A Computational Model for Tensor Core Units
A Computational Model for Tensor Core Units
复制标题
DOI：
10.1145/3350755.3400252
复制
发表时间：
2019-08
期刊：
影响因子：
--
通讯作者：
Francesco Silvestri;Flavio Vella
中科院分区：
文献类型：
--
作者：
Francesco Silvestri;Flavio Vella
为了满足深度神经网络的有效训练和推理需求，已经引入了大量特定于领域的架构，例如Google Tensor Processing Units和NVIDIA Tensor Cores。这些架构的一个共同特征是设计用于有效地计算给定小尺寸的密集矩阵乘积。为了拓宽利用这些系统的算法类，我们提出了一个计算模型，名为TCU模型，它捕获的能力，本机乘小矩阵。然后，我们使用TCU模型设计快速算法的几个问题，包括密集和稀疏矩阵乘法和离散傅立叶变换。最后，我们强调TCU模型和外部存储器模型之间的关系。
To respond to the need for efficient training and inference of deep neural networks, a plethora of domain-specific architectures have been introduced, such as Google Tensor Processing Units and NVIDIA Tensor Cores. A common feature of these architectures is the design for efficiently computing a dense matrix product of a given small size. In order to broaden the class of algorithms that exploit these systems, we propose a computational model, named the TCU model, that captures the ability to natively multiply small matrices. We then use the TCU model for designing fast algorithms for several problems, including dense and sparse matrix multiplication and the Discrete Fourier Transform. We finally highlight a relation between the TCU model and the external memory model.