Benchmark
Native C11 consensus-polish kernel
Simulated
- Result
- 60 / 60 decodes byte-identical to the NumPy reference; median speed-up 2.46× (bootstrap 95 % 2.41-2.51) (speed-up ×)
- Classification
- Software-generated DNA strands passed through a software channel model. Not a laboratory result.
- Methodology
- Correctness first: 50 golden hashes, 100,000 randomised reads, hypothesis tests, thread invariance, ASan/UBSan. Then a paired timing ratio per read file; 10,000-resample bootstrap.
- Dataset
- 60 decodes on seeds 94000-94009 across six cells (D13-F1 at coverage 3, 5, 10; both profiles).
- Configuration
- V8 decoder (NumPy polish) and V9 decoder (native vnx_cl_edit_costs) on the same read file per decode.
- Environment
- Development host: shared VPS, Intel Xeon Gold 6240 @ 2.60 GHz, 4 cores / 8 threads, Ubuntu 24.04, Python 3.12, CPU only. Host load 0.87-7.39 during the benchmark; the ratio is paired per read file.
- Limitations
- Speed-up range 1.06-2.98×. Timings on a shared host. The reads are simulated.
- Version
- V9
- Source
- v9.0.0/experiments/v9/consensus/results/benchmark-d13f1.jsonl
- Reproduce
experiments/v9/reproduce.sh decoder· how to reproduce