Hardware-native performance evidence on AMD Zen

Systems research into evidence-backed performance work on AMD Zen — methods before marketing numbers.

Research question

Can an evidence-driven performance loop specialized to AMD Zen produce more trustworthy optimization decisions than generic benchmark folklore?

Hypothesis

Tight coupling between telemetry, microarchitecture awareness, and reproducible benchmarks reduces false performance wins.

Methodology

Kraftverk + Silicera research software: collect telemetry and specialization experiments on AMD targets. Publish methods in-repo; withhold unverified headline numbers from this site.

Results

Infrastructure for measurement exists. Numerically impressive claims are intentionally omitted until reproducible suites are ready to cite.

Limitations

AMD-exclusive direction; incomplete public benchmark corpus.