A simple GUI tool to benchmark your hardware's FLOPS (Floating Point Operations Per Second) performance using popular neural network models.
Features
- Simple Mode: Quick ResNet50 benchmark (offline, no downloads)
- Advanced Mode: Test multiple models (ResNet50, EfficientNet-B0, ViT-B/16)
- Real-time progress tracking and detailed performance metrics
- Cross-platform GUI using tkinter (built into Python)
- Offline capable - works without internet connection
Quick Start
-
Install dependencies:
pip install -r requirements.txt
-
Run the application:
-
Click "Run Benchmark" for instant results, or enable "Advanced Mode" for more options.
Requirements
- Python 3.8 or higher
- ~2GB RAM for model loading
- Works with CPU and compatible GPUs
Performance Metrics
The tool provides:
- Runtime TFLOPS - Actual throughput on your hardware
- Hardware utilization - Efficiency percentage
- Throughput - Samples processed per second
- Timing - Per-sample and per-batch measurements
Models Tested
- ResNet50: Traditional CNN (available offline)
- EfficientNet-B0: Modern efficient architecture
- ViT-B/16: Vision Transformer (attention-based)
Notes
- Simple mode uses ResNet50 with random weights (no downloads required)
- Advanced mode can download pretrained weights for more comprehensive testing
- Results can be saved as text files or CSV for further analysis
Perfect for: Hardware evaluation, GPU benchmarking, AI workload testing