system-burner

Hardware burn-in stress tests using real ML workloads on MNIST.

Scripts

File	Description
01_tsne_cpu.py	t-SNE 2D on full MNIST pixels (n_jobs=-1)
02_pca_cpu.py	PCA 2D on full MNIST pixels
03_umap_cpu.py	UMAP 2D on full MNIST pixels (n_jobs=-1)
04_matmul_cpu.py	4096×4096 matrix multiply loop (BLAS multi-core)
05_matmul_gpu.py	8192×8192 matrix multiply loop (CUDA)
06_mnist_mlp_gpu.py	MLP 784→768→256→10 training loop (CUDA)
07_mnist_cnn_gpu.py	CNN training loop (CUDA)
08_mnist_vit_gpu.py	Vision Transformer training loop (CUDA)
09_xgboost_cpu.py	XGBoost 100 trees loop, all cores
10_xgboost_gpu.py	XGBoost 100 trees loop, CUDA
11_xgboost_gridsearch_cpu.py	XGBoost GridSearchCV (n_jobs=-1 outer, single-core inner)
12_xgboost_gridsearch_gpu.py	XGBoost GridSearchCV CUDA outer n_jobs=-1

Usage

python 01_tsne_cpu.py      # CPU test
python 05_matmul_gpu.py    # GPU test

All scripts run as infinite loops. Press Ctrl-C to stop.

Requirements

pip install -e .

GPU scripts require CUDA-enabled PyTorch and a compatible NVIDIA GPU.