Benchmarks, experiments and research notes on Rust, GPUs and machine learning. Every number comes with the script and the hardware that produced it.
- 01
Training a 100M transformer in pure Rust: RustingBrain vs PyTorch and TensorFlow
A Rust deep-learning library with NVRTC-compiled CUDA kernels against torch.compile and XLA on one RTX 3060. Where it wins, where it loses, and a benchmark bug I shipped.