A case study of the B200 pipeline model A note on methodology: Everything in this article is based on my analysis of microbenchmarks executed directly on B200 silicon. Nvidia does not publish i...

When building a dataflow analysis framework in a compiler, you quickly encounter an array of mathematical terms: semilattices, semirings, Kleene algebras, Galois connections, and Knaster-Tarski fix...

Training is gradient descent, and gradient descent needs gradients, so every ML framework has an automatic differentiation engine at its core. What is less obvious is that the representation a fram...

If you train or serve models, you depend on MLIR whether or not you have ever written a line of it. XLA lowers through it, Triton is built on it, Mojo is MLIR-native, and Torch-MLIR, IREE, and ONNX...

Python is a productive orchestration language for inference serving, but it is the wrong tool for the critical path of token generation. Large-model inference is bound by memory bandwidth and by pe...

VLIW (Very Long Instruction Word) spent most of the last two decades as a specialist’s tool. After the commercial struggles of general-purpose VLIW designs, with Intel’s Itanium being the canonical...

A step-by-step field guide to building vLLM from source on Ubuntu 26.04, covering Python 3.14 compatibility, CUDA driver issues, and toolchain pitfalls.

How vLLM's op‑level IR reconciles the tension between a compiler target and hand‑tuned kernel dispatch, enabling graph‑level fusion while supporting multiple back‑ends.

If you have a massive compute architecture—whether it’s a modern wide-SIMD vector engine, a Tensor Core array, or a custom deep learning accelerator like a Systolic Array—you face one fundamental p...

In a previous post, we explored the monumental software stack required to run a simple “Hello, World!” program on a modern operating system. But what happens when we apply these concepts to a heter...