// learn
Field guides
Long explainers on how inference runs, how the silicon under it is built, and how I build with agents. Each subject has a weekly tracker that follows it.
- inference 2 guides Inference engines tracker →
- silicon 2 guides Inference silicon tracker →
- agents 1 guide Frontier models tracker →
Agents you can repeatThe agents I trust aren't the clever ones. They're the ones whose every decision a script could make has been taken away from them.Read the guide →
A clock target is a budget"Run at 2.6 GHz" sounds like one number, but it's a budget of about 385 picoseconds that four different teams have to share.Read the guide →
Where the GPU memory goesOn a shared GPU, the memory for every request is set aside before the first one arrives, and most serving surprises come from not knowing where it went.Read the guide →
How Inference Chips Are BuiltWhy GPUs are inefficient for LLM inference, how ASICs go from RTL to silicon, and what Cerebras, Fractile, Groq, Etched, and Tenstorrent are doing differently to overcome the memory wall.Read the guide →
Inference, End to EndPrefill and decode, why one decode is memory-bound, precision and the KV cache, speculative decoding, the memory hierarchy, and where the CPU re-enters the picture.Read the guide →
// trackers
Weekly trackers
An agent reads the release notes every week and writes up what moved, so I don’t keep twenty tabs open.
Inference engines vLLM, SGLang, TensorRT-LLM, LMDeploy, llama.cpp and the rest. Open tracker → Frontier models Closed and open frontier models, small on-device models and agent frameworks. Open tracker → Inference silicon GPUs, wafer-scale chips, dataflow LPUs, inference ASICs and RISC-V. Open tracker →