benchHUB is a comprehensive cross-platform benchmarking tool that measures system performance across multiple domains. It features local benchmark execution, a REST API backend, web dashboard, and online leaderboard with score submission.

• 6 Benchmark Categories: CPU (prime calculation, parallel processing), GPU (tensor operations, ML training), Memory (bandwidth testing), Disk I/O (read/write operations), Machine Learning (dataset creation, model training), and Plotting (scatter plots, animations, large image rendering)
• Configurable Intensity Profiles: Light, Standard, and Heavy benchmark modes for different testing needs
• Cross-Platform GPU Support: CUDA (NVIDIA) and MPS (Apple Silicon) with graceful fallback
• Reference Index Scoring: Normalized scoring system combining CPU, GPU, and Memory performance
• Online Leaderboard: Submit and compare results with other users anonymously
• Streamlit Dashboard: Interactive web interface for visualizing and comparing benchmark results
• REST API: FastAPI backend for result submission and leaderboard management
• Modular Architecture: Extensible codebase for adding custom benchmarks

The main leaderboard showing ranked systems with medal indicators, reference index scores, and detailed hardware specifications.

Navigation sidebar and additional pages including download instructions and methodology documentation.

1. Clone the repository:

git clone https://github.com/Tennisee-data/benchHUB.git
cd benchHUB

2. Install the dependencies: Make sure you have Python 3.8+ installed. Create a virtual environment and install required packages:

python -m venv venv
source venv/bin/activate  # On Windows: venv\\Scripts\\activate
pip install -r requirements.txt

benchHUB uses three configurable intensity profiles to accommodate different hardware capabilities and time constraints. You can modify these in benchHUB/config/config.py:

• N_RUNS: 3 benchmark iterations
• DISK_FILE_SIZE: 25MB for I/O testing
• CPU_PRIME_LIMIT: Calculate primes up to 20,000
• MEMORY_ARRAY_SIZE: 100MB array operations
• GPU_MATRIX_SHAPE: 4096×4096 matrix operations
• ML_SAMPLES/FEATURES: 5,000 samples with 10 features

• N_RUNS: 3 benchmark iterations
• DISK_FILE_SIZE: 50MB for I/O testing
• CPU_PRIME_LIMIT: Calculate primes up to 50,000
• MEMORY_ARRAY_SIZE: 250MB array operations
• GPU_MATRIX_SHAPE: 8192×8192 matrix operations
• ML_SAMPLES/FEATURES: 10,000 samples with 20 features

• N_RUNS: 5 benchmark iterations
• DISK_FILE_SIZE: 100MB for I/O testing
• CPU_PRIME_LIMIT: Calculate primes up to 100,000
• MEMORY_ARRAY_SIZE: 500MB array operations
• GPU_MATRIX_SHAPE: 10000×10000 matrix operations
• ML_SAMPLES/FEATURES: 20,000 samples with 50 features

1. Run benchHUB (from the root benchHUB directory):

python -m benchHUB.main

It executes benchmarks defined in the project and saves results to the results/ directory in JSON format.

2. Visualize Results (from the root benchHUB directory): To launch the Streamlit dashboard for visualizing benchmark results:

streamlit run dashboard.py

This will parse the JSON files in the results/ directory and provide an interactive UI to analyse and compare metrics.

3. Adding Your Own Tests: To add a custom test, follow these steps:

• Create a new Python script in the sub directory benchHUB/ (I know the child should not have the parent's name but, it was such a great name...), e.g., custom_bench.py.
• Define your test logic, ensuring the output matches the format expected by the JSON parser.
• Import and integrate the test in main.py to include it in the benchmarking suite.

benchHUB includes a Reference Index scoring system that normalizes benchmark results into comparable scores:

• Composite Score: Combines CPU (40%), GPU (40%), and Memory (20%) performance
• Inverse Timing: Lower execution times result in higher scores
• Score Capping: Final scores are capped under 1000 for better user experience
• GPU Fallback: Systems without GPU support gracefully return 0 for GPU scores
• Anonymous Leaderboard: Submit results to compare with other users worldwide

The scoring system enables fair comparison across different hardware configurations while maintaining realistic score ranges.

For administrators managing the online leaderboard database, see the comprehensive guide:

📖 DATABASE_MANAGEMENT.md

This guide covers:

• Database cleanup procedures
• Score validation and maintenance
• Production database administration
• Troubleshooting common issues