Benchmarks, with their conditions attached.
A performance number without its sample, genome build, reference set and pipeline version is not a benchmark — it is a claim. This page will carry all four for every figure, and is empty until the first measured run.
01 · Results
Pending first measured run.
| Metric | Value | Sample | Build | Reference set | Pipeline | State |
|---|---|---|---|---|---|---|
| 30× WGS wall clock | — | NA12878 | GRCh38 | — | — | Pending |
| SNV F1 score | — | NA12878 | GRCh38 | — | — | Pending |
| Indel F1 score | — | NA12878 | GRCh38 | — | — | Pending |
| Cost per genome | — | — | — | — | — | Pending |
02 · Methodology
What will be measured, and how.
Wall clock is measured end to end — from the start of analysis to delivered results — not as the sum of accelerated stages. Reporting only the fast parts is how a 30× component becomes a misleading 3× product claim.
Accuracy is measured as precision, recall and F1 against Genome in a Bottle truth sets, reported separately for single-nucleotide variants and for insertions and deletions, because their error profiles differ materially.
Cost per genome is all-in: compute at the instance pricing actually paid, storage amortised, data transfer, and any bitstream licence amortised per genome. A cost figure that omits the licence is not a cost figure.
The CPU baseline is stated alongside every speedup, with the instance type named, so the comparison can be checked rather than assumed.
03 · Why this page is empty
Because the pipeline it would measure does not exist yet.
The FPGA execution plane is engineering Build Phase 3 and has not started. We could populate this table from a CPU reference implementation, but that would not represent the shipping product and would be misleading in the flattering direction. “Up to 30×” is AWS's general figure for FPGA acceleration; “under six hours” is our design target. Neither is a GenoAccel measurement and neither will be presented as one.