May 10, 2026
First Roofline Note
A placeholder post to validate the Astro, MDX, interactive component, and Netlify publishing flow.
Technical notes
Processor architecture, GPU performance, and LLM systems from first principles.
Notes, diagrams, experiments, and interactive tools for understanding how algorithms meet hardware.
Architecture
Microarchitecture, memory systems, datapaths, pipelines, and the constraints that shape real machines.
Performance
Roofline models, arithmetic intensity, GPU kernels, bandwidth ceilings, and bottleneck analysis.
LLM Systems
Inference performance, batching, KV cache behavior, quantization, and algorithm-hardware tradeoffs.
Catalogs
Accelerators
Training and inference chips, memory capacity, bandwidth, precision peaks, interconnects, and roofline plots.
Memory
HBM and DDR standards compared by bandwidth, organization, platform generation, and system role.
Interconnects
Scale-up and host I/O fabrics including NVLink, Infinity Fabric, PCIe, CXL, and emerging open fabrics.