Data checked 2026-08-13. Data may be stale - beyond the review cycle.
Review cycles are documented on the Methodology page. Methodology
rockchip-linux/rknn-toolkit2
Software Tool Tier A edge_ai BSD-3-Clause
RKNN-Toolkit2 is Rockchip's official toolchain for converting and deploying models to RK series NPUs such as RK3588, the main path for RK-based companion computers.
Engineering Snapshot
Use cases & tasks 6
- Best suited for
-
- Edge AI deployment on Rockchip RK3588/RK3566/RK3568/RK3576 based companion computers
- Model conversion from ONNX/TensorFlow/PyTorch to RKNN format
- Quantization and optimization for Rockchip NPU inference
- Primary tasks
-
- Convert trained models to RKNN format for Rockchip NPU
- Quantize models (INT8/INT16) for efficient inference
- Deploy inference pipelines on RK3588/RK3566/RK3568/RK3576 boards
Stack & ecosystem
- Resource type
- Software Tool
- Ecosystem
- edge_ai
License & compliance
- License
-
BSD-3-Clause(Inferred)
Lifecycle & freshness Show
- Maintenance
- Active
- Latest version
- Not recorded
- Last activity
- 2026-08-10
- Last checked
- 2026-08-13
- Verification
- Official confirmed
What It Solves
Engineers need a Rockchip-supported toolchain to convert, quantize, and deploy neural network models onto RK3588/RK3566/RK3568/RK3576 NPUs for edge AI on companion computers.
Primary use cases
- Convert trained models to RKNN format for Rockchip NPU
- Quantize models (INT8/INT16) for efficient inference
- Deploy inference pipelines on RK3588/RK3566/RK3568/RK3576 boards
Secondary use cases
- Performance profiling and optimization on Rockchip NPU
- Integration with custom C++/Python applications on Linux
When to Use
Consider when
- Target hardware is Rockchip RK3588, RK3566, RK3568, or RK3576
- License compatibility with BSD-3-Clause is acceptable
- Model architectures are supported by RKNN-Toolkit2 (check release notes)
- Quantization accuracy trade-offs are evaluated for the specific model
Verify before adopting
- Model conversion success for your specific architecture and opset
- Quantization accuracy drop on representative validation set
- Inference latency and throughput on target RK NPU
- Compatibility with host OS (Linux) and Python/C++ API versions
Start Here
Adoption Checklist
- Needs verification Model conversion success for your specific architecture and opset
- Needs verification Quantization accuracy drop on representative validation set
- Needs verification Inference latency and throughput on target RK NPU
- Needs verification Compatibility with host OS (Linux) and Python/C++ API versions
Each check stays "needs verification" until an official source confirms it; unconfirmed items are never marked verified.
Known Limitations & Unknowns
Known limitations
- Only supports Rockchip RKNPU targets (RK3588, RK3566, RK3568, RK3576)
- Model operator coverage limited to RKNN supported ops
- Quantization may require calibration data and tuning
- Toolchain runs on Linux host (x86_64/aarch64); Windows/macOS not officially supported
Not publicly verified
- Exact model support matrix per RKNN version
- Performance benchmarks for specific models on each RK NPU
- Long-term maintenance roadmap and release cadence
Alternatives & Related Tools
Related tools
- onnx-runtime — Integrates with
How is it used?
Start from the recorded entry points below, then validate against the technical checklist.
- Official site https://github.com/rockchip-linux/rknn-toolkit2
- Repository https://github.com/rockchip-linux/rknn-toolkit2
- Documentation https://github.com/rockchip-linux/rknn-toolkit2
Technical checklist
- OK License identified Recorded: BSD-3-Clause
- OK Maintenance signal Active
- OK Verification status Official confirmed
- OK Source evidence attached 1 source record(s)
- NEEDS REVIEW Latest version recorded Not recorded
Official Links
Metadata & Governance
| License | BSD-3-Clause — Inferred |
|---|---|
| Commercial model | Unknown |
| Maintenance status | Active |
| Verification status | Official confirmed — Confirmed via the official repository API responses in SourceRefs below. |
| Latest version | Not recorded |
| Latest release | Not recorded |
| Last activity | 2026-08-10 |
| Last checked | 2026-08-13 |
| First seen | Not recorded |
Dataset facts
Facts above come from the official dataset card only; unconfirmed fields stay unknown.
Related Resources & Dependencies
- onnx-runtime — integrates with (confirmed)
Recent Activity
New resource: rockchip-linux/rknn-toolkit2
New repository resource added by the sprint promote pipeline.
Related Knowledge
Guides
- Deploying AI Inference on UAV Edge Compute — A repeatable workflow for choosing and validating an inference runtime for an onboard UAV computer.
Collections
- Edge AI Perception Starter — A starting stack for prototyping aerial perception models before edge deployment work.