Sitemap - 2026 - The Software Frontier
Inside Google’s TPU: How it works, and what it costs
A wonderful dive into how FP4 works
Why Cached Tokens Cost 10% and Vanish in 5 Minutes
How NVIDIA's Data Center Business Actually Works in 2026
Distilling in depth ROCm: How it Actually Works
Exploring how Triton actually compiles
DeepSeek V4-Flash: The Cost of Deciding What to Read
Invite your friends to read The Software Frontier
How Blackwell’s Tensor Memory Actually Works
How CUDA Binaries Actually Work
How the NVIDIA Compiler Moat Actually Works
Kimi K3: 2.8 Trillion Parameters, Four Bits at a Time
Exploring how MLIR works: the compiler rewiring the AI stack
How LLVM Works: The IR That Took Over Modern Computing
Decode Is Memory-Bound. Speculation Is the Arbitrage
We wrote the CUDA reference we could not find
The Kill Switch Was in the Mail
How three companies set the price of intelligence
The Vertical and the Loop: valuation, compute, and the Anthropic IPO
The Blackwell Migration Question
How Systems Really Fail, Part IV
The Llama 3.3 70B Benchmark Problem
How Systems Really Fail, Part III
How Systems Really Fail, Part II
How Systems Really Fail, Part I
We built the CUDA guide I wish I had three years ago
Mastering CUDA and High-Performance Computing, Part X
Mastering CUDA and High-Performance Computing, Part IX
Mastering CUDA and High-Performance Computing, Part VIII
Mastering CUDA and High-Performance Computing, Part VII
Mastering CUDA and High-Performance Computing, Part VI
Mastering CUDA and High-Performance Computing, Part V
Mastering CUDA and High-Performance Computing, Part IV
Mastering CUDA and High-Performance Computing, Part III
Mastering CUDA and High-Performance Computing, Part II
Mastering CUDA and High-Performance Computing, Part I
How Meta turned the Linux Kernel into a planet-scale Load Balancer. Part III
How Meta turned the Linux Kernel into a planet-scale Load Balancer. Part II
How Meta turned the Linux Kernel into a planet-scale Load Balancer. Part I