Skip to content

Analysis, visualization, and machine learning on the top500 supercomputers

Notifications You must be signed in to change notification settings

tommygorham/top500

Repository files navigation

top500

Data Visualization & Analysis of the latest top500 dataset using Python and Jupyter notebook.

Latest Dataset: November 2024

Machine Learning

Correlation between Theoretical Peak Performance and Computer Architecture

Theoretical_Peak+Performance_Heatmap_of_Corr_Spring22

ML Predictions

Key Findings and Observations

  • Power (kw) and total cores show strong positive correlations indicating that power consumption and more cores generally lead to higher performance.

  • Cores per socket, Processor Speed (mhz) and CPU Type show relatively weaker positive correlations suggesting that cores per socket, CPU type, and processor clock speed are not significant drivers or predictors of supercomputer performance in modern systems.

Top500 Visualizations & Charts

Power vs Performance

Interactive Hover Plot

Non-interactive Plot
Power Vs Performance

CPU Share GPU Share

Heterogeneity

Interconnects

Trends

2023 -> June 2024

  • AMD CPU Share now 31%, up from 24% last year (2023)
  • 2% increase in CPU+GPU machines from 2023
  • 88.6% of GPUs still NVIDIA

2022 -> 2023

  • 28 new machines with AMD CPUs replace 28 machines with Intel CPUs
  • 17 new CPU + GPU Machines
  • 90% of GPUs are NVIDIA

About

Analysis, visualization, and machine learning on the top500 supercomputers

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published