The Philosophy
Fridays with Faraday is dedicated to the craft of performance engineering. We believe that understanding systems at the deepest level—from register bits to memory hierarchies to distributed protocols—is essential for building truly efficient software.
Every post here aims to go beyond surface-level explanations. We measure, profile, and validate. We show the assembly, the flame graphs, the memory traces. If a technique claims 2x improvement, we demonstrate it with reproducible benchmarks.
Topics We Cover
- LLM Inference Optimization — vLLM internals, attention mechanisms, KV cache management, continuous batching, speculative decoding
- GPU Programming — CUDA kernels, memory coalescing, Habana Gaudi optimization, tensor parallelism
- Embedded Systems — Microcontroller optimization, register-level programming, power management, real-time constraints
- Performance Profiling — perf, eBPF, flame graphs, memory profiling, systematic performance analysis
- Distributed Systems — Inference serving, load balancing, multi-GPU coordination
The Name
Michael Faraday was a self-taught scientist who made fundamental contributions to electromagnetism through meticulous experimentation. He didn't have formal mathematical training, so he relied on physical intuition and careful observation.
That spirit guides this blog: understand through measurement, validate through experiment, explain through concrete examples. Theory is useful, but profiler output is truth.
About the Author
I'm a performance engineer with a background in embedded systems and large-scale inference. I've spent years optimizing code from 8-bit microcontrollers to multi-GPU clusters, and I believe the fundamentals are the same: understand your hardware, measure before optimizing, and question every assumption.