End-to-end speedup vs CPU
HG002 30x WGS on A100 40G
12×
Private pilot baseline measured against a tuned CPU pipeline. Full breakdown requires pilot access.
Product · Omnis Genome Nexus
GPU-native genomics compute architecture.
Receipt-first. Deterministic. Reproducible.
Job Spec → GPU fabric → receipt → optional proof bundle
Most genomics platforms accelerate individual tools. OGN treats genomics as a compute fabric: inputs, references, alignment state, variant evidence, logs, artifacts, and receipts are coordinated through one execution model.
OGN Core Kit is the open-source contract layer. OGN Engine is proprietary. OGN Cloud/Gateway is the hosted control plane.
You define success
Core Kit contractEngine executionCloud artifactsReceipt verification
Public evaluators can validate, mock-run, and verify receipts without private engine access.
Evidence
Benchmarks and artifacts you can audit — speed, accuracy, determinism, and proof bundle deliverables.
VerifiedExamine proof bundle
End-to-end speedup vs CPU
HG002 30x WGS on A100 40G
12×
Private pilot baseline measured against a tuned CPU pipeline. Full breakdown requires pilot access.
Proof bundle deliverable
Hashes + container digests
Optional
Receipt is the minimum artifact. Proof bundles are emitted only when requested and registered.
Rerun determinism
Pinned inputs + containers
Tracked
If a rerun differs, it is visible in the determinism + verification report.
HG002 SNP F1
GIAB HG002 harness
0.9991
Private pilot baseline; benchmark pack separates public evidence from private run folders.
Proof, not promises
Built for teams that require proof
Designed for teams burned by pipeline drift, brittle glue, and opaque outputs. OGN makes receipts the minimum trust artifact and keeps proof bundles optional in the run contract.
Failure mode (anonymized)
Same reads. Same reference. Different VCF. The team lost weeks reconciling drift that turned out to be unpinned containers and nondeterministic stages. OGN is built so drift is caught before artifacts ship.
Proof bundle
- This is what you get: VCF, metrics, hashes, container digests, and a reproducible run script.
- A determinism + verification report shows whether reruns match and where drift would come from.
Determinism
- Pinned containers + regression harnesses keep HG002 stable run-to-run.
- Verification gates cross-check variants and metrics before artifacts ship.
Performance
- GPU-native kernels + streaming reduce turnaround time without giving up traceability.
- Benchmark packs include cost breakdowns, configs, and container digests.
Start with the public contract
Validate the Job Spec, mock-run without private engine access, then request a private pilot when you need engine-backed runs.
Part of the Omnis platform
Omnis Genome Nexus runs on the same GPU rail as Omnis Helix and the VeriBiota verification layer.
Benchmarks and proof artifacts
We benchmark directly against leading workflow managers and GPU stacks. Full HG002 reports ship with configuration details, container digests, verification outputs, and cost breakdowns.
Used internally to catch drift before it hits production, and shipped as part of private pilot packets.
You get full benchmark packs, not cherry-picked screenshots. Every run includes raw metrics, container digests, command lines, and determinism checks. Most pipelines fail once you require reruns to match and artifacts to be auditable.
Proof bundle deliverable
Hashes + container digests
Optional
Receipt is the minimum artifact. Proof bundles are emitted only when requested and registered.
Rerun determinism
Pinned inputs + containers
Tracked
If a rerun differs, it is visible in the determinism + verification report.
HG002 SNP F1
GIAB HG002 harness
0.9991
Private pilot baseline; benchmark pack separates public evidence from private run folders.
Benchmarks last updated Dec 7, 2025.
Runtime comparison against declared baselines
HG002 30x WGS on A100 40G
Use cases
Built for teams where runtime, trust, and programmability are non-negotiable.
High throughput clinical calling
Run 30× genomes end-to-end with deterministic artifacts and predictable turnaround.
Explore High throughput clinical calling →
Variant → candidate ranking
Triage candidates on one deterministic GPU rail.
Explore Variant → candidate ranking →
Platforms embedding genomics
Add GPU genomics to LLM agents and workflow products with one API. Power automated reviewers, fast filters, and custom pipelines.
Explore Platforms embedding genomics →
Developer corner
Command line and API first. Tie Omnis Genome Nexus into your existing workflow managers or call the GPU fleet directly via gRPC. Example: submit a 30× genome, stream metrics, and export VCF and summaries in one shot.
OGN gateway
$ ogn submit \ --reads s3://datasets/hg002/R1.fastq.gz \ --reads s3://datasets/hg002/R2.fastq.gz \ --ref s3://refs/grch38_noalt.fa \ --pipeline wgs30
Deterministic container builds, VeriBiota verification, and streaming artifacts.
Run the public contract, then request engine access
Validate Job Spec v1, mock-run without private engine access, verify receipt.json locally, and move to a private OGN pilot when you need proprietary execution.