Cluster workstation and method for realizing same

Abstract

The invention discloses a cluster workstation. The workstation comprises a plurality of mainboards, IB (InfiniBand) daughter cards, a middle plate and an IB switching module, wherein each of the mainboards comprises an IB network card arranged thereon; the IB daughter cards are detachably connected with the mainboards; the IB network cards of the mainboards are connected with the middle plate through the IB daughter cards; and the IB switching module is connected with the middle plate. The invention also discloses a method for realizing the cluster workstation. The method comprises the following steps of: connecting the plurality of network cards in the plurality of mainboards to the middle plate through the IB daughter cards; and connecting the middle plate to the IB switching module. According to the cluster workstation provided by the invention, all single workstation nodes are subjected to resource integration, so that an integrally unified workstation is formed to provide data communication and image processing service for users. Through high-speed interconnected channels among all the nodes inside a cluster, the data intercommunication between a GPU (Graphics Processing Unit) of different nodes and a graphics card can be realized, and the cluster workstation integrates operating, graph and image processing and information interacting.

Description

technical field [0001] The invention relates to a cluster workstation and a method for realizing the cluster workstation. Background technique [0002] A workstation is a processing machine that integrates graphics and images, data calculations, and result delivery. The performance of workstations has always been the bottleneck of high-performance computing and large-scale computing, especially when there are large-scale graphics and image processing tasks. A key factor in the effectiveness of task processing. The graphics and image processing architecture of CPU (central processing unit) + GPU (graphics processing unit) has gradually spread. Using GPU resources to process graphics and auxiliary calculations can make up for the bottleneck of CPU single processing speed. GPU-assisted computing separates the CPU from multiple tasks, allowing the CPU and GPU to perform their duties, thereby improving performance. [0003] At present, high-speed graphics and image functions ca...

AI Technical Summary

Problems solved by technology

[0004] The current solution has the performance of a single node (that is, a single motherboard, a single workstation), and there is no corresponding workstation redundancy; at the same time, the scalability of the workstation is not strong, and the addition of nodes needs to be connected through Ethernet, but the speed of Ethernet is still high. The bottleneck of most large-scale high-performance graphics operations, at most it can satisfy some small-scale operations

The scalability is not strong, there is no corresponding redundancy, and the performance of node connections is limited

Images