Skip to content

A GPU-vs-CPU performance benchmark: (OmniSci [MapD] Core DB / cuDF GPU DataFrame) vs (Pandas DataFrame / Postgres / PDAL)


Notifications You must be signed in to change notification settings


Repository files navigation

Supercharging Analytics with GPUs

A performance benchmark


This repo contains multiple performance tests and their results regarding how GPUs accelerate several analytics operations; we use these two types of datasets:

  • Financial data (per-minute price for 63 ETFs)

  • LiDAR data (aerial LiDAR scan of Montreal, Canada)

Component versions evaluated

Conda Environments

  • requirements_cpu_node.txt : env where Pandas/Postgres tests were run;

  • requirements_gpu_node.txt: env where cuDF/MapD Core DB tests were run;

  • requirements_cpu_node_LiDAR-PDAL.txt : env where PDAL tests on the LiDAR data were run.

Node hardware specs

  • CPU node/worker : 8 vCPUs, ~60GB RAM

  • GPU node/worker: V100 GPU, 8 vCPUs, ~60GB RAM

Dataset Specs

Financial time-series:

  • Covering the last 20 years for 63 ETF symbols

  • Average per-minute price (available when traded)

  • 50 million records

  • 3.5 GB CSV file size

  • 6 GB in-memory size (RAM, Pandas DF)

  • 5 GB in-memory size (GPU memory, cuDF)

Geospatial-LiDAR (Montreal LiDAR aerial scan) - stats of one tile:

  • 18,306,827 points

  • 82 MB Laz

  • 1.6 GB CSV

  • 681 MB in the mapD DB

Notes regarding datasets

  • Financial data cannot be redistributed due to licensing issues, but a sample is provided to give you an idea regarding its format;

  • LiDAR data can be freely downloaded from the website of City of Montreal (link in the code), but if you think your use might entail too many downloads, it's a good idea to contact them to inquire about alternative options (e.g., re-hosting those files in your own infrastructure).


If you use all or part of the code in this repository, we suggest that you include the following notice with your document, code or product:

This code/product is partly or fully based on the code which was originally run on the UniAnalytica platform ( and is published by PatternedScience Inc. at and licensed under the terms of Apache License 2.0; a copy of the license is available in the GitHub repository.

Feel free to adapt the "This code/product is partly or fully based on" part to your situation and use. If you need some modifications to the above text/license to accommodate better your use, please contact PatternedScience Inc.

Copyright © 2019 PatternedScience Inc.


A GPU-vs-CPU performance benchmark: (OmniSci [MapD] Core DB / cuDF GPU DataFrame) vs (Pandas DataFrame / Postgres / PDAL)







No packages published