At present, the campus in-depth learning platform is mainly divided into two technical fields, one is a container solution, and the other is a virtualization method. Considering the consideration of both teaching and scientific research, the choice of these two solutions is different. It is recommended to use a hybrid GPU platform or a multi container shared GPU model for the container approach. The virtualization solution can use V100S products and split different computing resources according to different needs.

At present, the commonly used software for finite element calculation is mainly commercial software, such as Matlab, Comsol, etc. Most of the calculation modules in Matlab support GPU acceleration, and the GPU type is not limited. At the same time, it supports parallel calculation. Therefore, we recommend using an 8GPU computing server. If the amount of computing is large, we can use a GPU cluster.

This field involves HPC high-performance computing, involves double precision computing, and has certain requirements for video memory capacity. V100S is recommended here, and four card machines are recommended for server selection because of the communication problem between GPUs. The 4GPU machines will not have the problem of bus speed reduction when doing end-to-end GPU interaction.