Docs · Omnis platform

Platform, pipeline, and proof docs

Use this docs shell for CLI workflows, architecture, benchmark contracts, and public operating guidance across the Omnis platform.

CLIArchitectureBenchmarksDeployment

Viewing

GPU support matrix

Need another page?

Search the docs

Jump to buyers, verification, demos, self-hosting, or adapters without opening the full docs tree first.

Minimum supported compute capability (CC): 7.0 (Volta) and newer.
Optimized targets (“fast path”):
- Ampere: CC 8.0, 8.6
- Ada/Lovelace: CC 8.9
- Hopper: CC 9.0 (DPX available)
DPX acceleration: Optional, Hopper+ only. Controlled by OGN_ENABLE_DPX_FASTPATH (CMake) and OGN_ENABLE_DPX (env, default on).
CPU fallback: Planned default for tests/CI via OGN_ENABLE_CPU_FALLBACK (CMake) and OGN_FORCE_BASELINE (env, forthcoming). Intended for environments without a GPU.
CUDA toolkit: 12.x recommended; tested primarily on 12.4–12.6 images.
Build architectures (fat binary default): 70;75;80;86;89;90. Override with -DCMAKE_CUDA_ARCHITECTURES= to slim builds if desired.

For production pipelines, prefer Ampere or newer. Hopper unlocks DPX fast paths; pre-Hopper devices run the baseline kernels.