Start with the kernel that actually ran
How artifact identity, target fingerprints, and paired timings keep GPU optimization work attached to reality.
Read notePhoenix Compute
A compact control surface for optimized GPU mining, JIT kernel exploration, and production-grade performance validation.
Engineering notes
How artifact identity, target fingerprints, and paired timings keep GPU optimization work attached to reality.
Read noteThe evidence ladder Phoenix uses to separate a promising kernel result from a deployable optimization.
Read notePhoenix focuses on correctness-gated GPU optimization, real workload validation, and clean integration boundaries between miner runtime and optimization tooling.