GPU Benchmark
pip install gpu-benchmark && gpu-benchmark
Run this command to benchmark your GPU and contribute to our global benchmark results
Loading benchmark data...
How the Benchmark Works
1Installation
A simple CLI tool that works on any CUDA-compatible NVIDIA GPU with Python 3.8+
2Run the Test
The benchmark runs for 5 minutes after loading the Stable Diffusion pipeline (default) or selected model.
What We Measure
Images Generated
Number of images your GPU can generate in 5 minutes (model dependent).
Max Heat (°C)
Maximum GPU temperature reached during the benchmark.
Avg Heat (°C)
Average GPU temperature during the benchmark.
Country
Your location (detected automatically).
Rank
Your result as a percentage of the best result in the entire database for the selected model: (your_score / best_overall_score_for_model) * 100
(num_images_generated / max_overall_score) * 100
GPU Model %
Your result as a percentage of the best result for your specific GPU model: (your_score / best_score_for_your_gpu_model) * 100
(num_images_generated / max_score_for_gpu_model) * 100
Power (W)
Average GPU power consumption during the benchmark.
Memory (GB)
Total available GPU VRAM.
Platform
The operating system or environment used for the benchmark.
Provider
The provider or host of the GPU (e.g., private, runpod, vast.ai, etc.).
Acceleration
The acceleration technology used (e.g., CUDA version).
PyTorch Version
The PyTorch version used for the benchmark.