Pick a workflow, plug in how many samples/complexes/edges/genomes you have, and get the AWS spot cost plus runtime. Numbers are from 12 verified Clusterra benchmarks — not vendor estimates.
VRAM minimums for 30+ biotech tools — real production numbers, not READMEs.
Live spot vs on-demand pricing for T4, A10G, L4, L40S, A100, H100 in us-east-1.
The 12 source runs these numbers come from — with public datasets and methodology.
Each cost in this calculator comes from a single verified Clusterra benchmark run: same hardware, same container digest, same public reference dataset. The blog posts on /benchmarks/ show the input data, commands, and output checksums for each.
The calculator scales linearly with your sample count, which is accurate for workflows that are embarrassingly parallel (AlphaDIA cohort, Boltz-2 co-fold, AlphaFold-Multimer batch). For workflows with a campaign-level fixed cost (FEP networks with shared protein prep, RELION refinement with shared reconstruction), there’s a small per-campaign overhead the calculator approximates as zero — in practice it’s under 2% of total.
Most vendor “cost per X” numbers quote on-demand pricing or a vendor markup. Biotech compute on AWS runs on Spot — checkpoint-restart on eviction handles the volatility — and Spot is typically 50–80% off on-demand. Quoting on-demand pricing inflates the bill by 2–5×.
Live spot prices for biotech instances are on the spot price page. These calculator numbers used those prices at the time of each benchmark run.
Wall-clock here is the time for one sample/complex/edge to finish on the recommended hardware. If you have 100 of them and the cluster autoscales (which it does on Clusterra), they finish in the same wall-clock time — not 100× longer. This is the difference between a sequential queue and an elastic cluster, and it’s why a 50-sample AlphaDIA cohort finishes in the same 13 minutes as a single sample.
For an end-to-end quote including all of the above, the easiest path is book a pilot call.