A computing cluster and a computing cluster configuration method

Abstract

The invention provides a computing cluster for training a deep neural network and a computing cluster configuration method. Wherein the number of the single-layer networks contained in the deep neuralnetwork to be trained is N, and N is a natural number; wherein each single-layer network comprises a plurality of function nodes; wherein any one single-layer network is represented as the ith layerof network; I is less than or equal to N; wherein the computing cluster comprises sub-computing clusters in one-to-one correspondence with the N layers of single-layer networks; wherein the sub-computing cluster corresponding to the i layer of single-layer network is the i sub-computing cluster; wherein the computing nodes in the i sub-computing cluster are in one-to-one correspondence with the functional nodes in the i layer of single-layer network; the computing node comprises at least one basic computing module; wherein each basic computing module comprises m acceleration cards which are in point-to-point full interconnection, and each acceleration card comprises n artificial intelligence chips which are in point-to-point full interconnection; M is a naturalnumber not less than 2; wherein n is a natural number; and any basic computing module and the adjacent basic computing module are interconnected through chip-level point-to-point connection.

Description

technical field

[0001] The invention relates to the field of computers, in particular to a computing cluster and a configuration method for the computing cluster. Background technique

[0002] Deep neural networks are a popular topic in the field of machine learning in recent years and have achieved advanced results on many tasks.

[0003] Deep neural networks need to be trained before they can be put to use. Training a neural network model requires a lot of memory and computing resources. Therefore, the memory and computing resources that a single computer can provide are far from enough. Contents of the invention

[0004] In view of this, the present invention provides a computing cluster and a computing cluster configuration method for training a deep neural network.

[0005] In order to achieve the above-mentioned purpose of the invention, the present invention provides the following technical solutions:

[0006] A computing cluster for training deep neural network...

Images