AI Agent / Web3·EVM · Solana
Render Network
Decentralized GPU network originally built for OctaneRender / Cinema 4D rendering and now extended to AI/ML compute via the Compute Clients program. Artists submit jobs in RNDR credits; node operators run idle GPUs and earn RENDER. The Render API + Manager app let pipelines submit jobs and query status programmatically.
- 01decentralized GPU rendering
- 02Octane / Blender / Cinema4D pipelines
- 03AI/ML inference and fine-tuning compute
- 04studio rendering automation
- 05RENDER-paid compute marketplace
- # Install the Render Network Manager app from rendernetwork.com
- pnpm add axios
| Variable | Scope | Description |
|---|---|---|
| RENDER_API_KEY | Server | Local API key issued by the Render Network Manager app for programmatic job submission. |
| RENDER_API_URL | Server | Local Manager API base URL (defaults to http://localhost:<port> exposed by the Manager app). |
Use Render Network's local API (exposed by the Render Network Manager app) to submit and monitor jobs. Authenticate every request with the `RENDER_API_KEY` header from the Manager and call `POST /jobs` with a payload describing the scene (Octane ORBX, Blender .blend, or container-based AI workload), GPU tier, and priority. Poll `GET /jobs/{id}` for `status` (`Queued`, `Rendering`, `Completed`, `Failed`) and `GET /jobs/{id}/frames` for output URLs. Pricing is denominated in RNDR credits; convert to RENDER on the bridge (Solana <-> EVM) when topping up. For AI/ML workloads, use the Compute Clients program to package the model as a container with a known entrypoint — RNDR is rendering-first, so expect a curated, lower-throughput fleet for non-render compute compared to io.net or Akash.
- ⚑RENDER token bridges between Solana and EVM via Wormhole — keep accounting separate per chain or balances will appear off.
- ⚑Pricing is in RNDR credits indexed to USD, but RENDER token volatility still hits operators; expect periodic rate adjustments per tier.
- ⚑GPU tier availability (Tier 1 high-end) clusters around US/EU; APAC node coverage is thin and adds queue time.
- ⚑AI/ML compute is gated through Compute Clients onboarding — you cannot self-serve arbitrary CUDA containers like on io.net/Akash.
- ⚑Image rendering output (Octane) is the mature path; text/LLM inference and image-gen pipelines are newer and have fewer compatible nodes.
- ⚑Network congestion during major studio releases pushes queue times into hours on Tier 1; build retry-with-tier-fallback logic.
- ⚑Slashing for malformed output exists at the operator side, but consumer-side disputes go through manual review — bake QA frame checks into your pipeline.