Administrator Guide

17 RAPIDS Scaling on Dell EMC PowerEdge Servers
6 Conclusion and Future Work
We have shown how Dell EMC PowerEdge servers with NVIDIA GPUs can be used to accelerate your data
science pipeline with RAPIDS. We have compared performance using both NVIDIA NVLINK & PCIE GPUs
using scale-up and scale-out server’s solutions using different storage configurations.
Main highlights:
o Using NYC-taxi 2014 dataset, server C4140-M 4V100-SXM2-16GB in Multi Node mode
(8x16=128GB) with RMM enable yielded the shortest total E2E time (~59 seconds), 58% faster than
server R940 4xV100PCle-32GB Single Node (~101 seconds)
o C4140-M 4xV100-SXM2-16GB with local data on SATA drive was 3% faster than remote data on
NFS
o System Profile Settings in “Performance” mode yield ~7%-9% in boost performance
The experiments run in this paper show the basic method to deploy RAPIDS with DASK on multiple
nodes. As alternative to automate the deployment for multi-node in production environments, the tests
can be conducted using a cluster resource manager such as SLURM, PBS, Kubernetes, Yarn among
others.