TL;DR
The OrangePi 6 Plus is a larger, heatsink-equipped single-board computer packing a 12-core CIX SoC, up to 64GB LPDDR5, a powerful Immortalis GPU and a substantial NPU. Hardware and IO are ambitious, but Linux support is uneven: a Debian Bookworm image works, kernel choices create trade-offs, and advanced features require extra setup.
What happened
The reviewer tested the OrangePi 6 Plus, a full-featured ARM64 single-board computer that departs from credit-card sized boards by offering a larger footprint and an integrated heatsink. The board ships with a 12-core CIX CD8180/CD8160 SoC (tri-cluster Armv9.2 architecture), Immortalis-G720 MC10 GPU, and an on-board NPU rated at roughly 30 TOPS dedicated (about 45 TOPS system-wide). Storage and IO are extensive: two full-size M.2 2280 NVMe slots (PCIe 4.0 x4), dual 5GbE ports, multiple display outputs including HDMI 2.1 and DP1.4, dual USB-C PD power input, and a 40-pin GPIO header. A Debian Bookworm image with kernels 6.1 and 6.6 was available for testing; initial NVMe boot required a firmware update. Desktop use felt responsive and handled high-resolution displays and 4K video, but driver and kernel limitations create trade-offs for Vulkan, NPU support, and other subsystems.
Why it matters
- Delivers desktop-like responsiveness on ARM64 hardware, potentially broadening SBC use beyond embedded projects.
- High-bandwidth IO (dual M.2 NVMe, PCIe Gen4, dual 5GbE) enables storage- and network-heavy applications on a compact board.
- Software stack constraints—kernel version, drivers and NPU tooling—could limit out-of-the-box functionality for many users.
- Requires hands-on work (firmware updates, extra packages, compiling) to enable certain peripherals and apps, making it more suitable for experienced users.
Key facts
- SoC: CIX CD8180 / CD8160 (codename CIX P1) — 12 cores in a tri-cluster Armv9.2 layout (4× Cortex-A720 high-performance, 4× Cortex-A720 mainstream, 4× Cortex-A520 efficiency).
- GPU: Arm Immortalis-G720 MC10 with hardware ray tracing and support for Vulkan 1.3, OpenGL ES 3.2 and OpenCL 3.0.
- NPU: Dedicated ~30 TOPS (claimed) with system-wide AI up to ~45 TOPS; supports INT4/INT8/INT16/FP16/BF16/TF32 data types.
- Memory: LPDDR5 128-bit available in 16GB, 32GB or 64GB configurations; theoretical peak bandwidth up to 96 GB/s noted in specs.
- Storage/expansion: Two M.2 2280 slots (PCIe Gen4 x4), one microSD slot, dual 5GbE Ethernet, M.2 Key-E slot for Wi‑Fi/Bluetooth modules, and two 4-lane MIPI CSI camera interfaces.
- Video and ports: HDMI 2.1 (8K@60Hz), DisplayPort 1.4, two USB-C ports with DP Alt Mode, eDP, two USB 3.0 Type-A and two USB 2.0 Type-A.
- Power and form factor: Dual USB-C PD (20V/5A) 100W input; board measures 115mm × 100mm and ships with an integrated heatsink.
- Software: Debian Bookworm image provided with kernels 6.1 and 6.6; Vulkan drivers present but limited to an early 1.3 implementation and Mesa/Panfrost support is constrained by kernel version.
- NPU tooling: NeuralONE AI SDK is required to target the NPU; common frameworks like stock PyTorch/TensorFlow won’t run on it without conversion/SDK steps.
- Initial NVMe boot required a firmware update; Bluetooth audio required installing pipewire-related packages and manual pairing steps.
What to watch next
- Availability of an official Ubuntu 24.04 image and any ETA — source indicates plans shifted and there is currently no ETA.
- Progress on upstreaming/mainlining kernel patches so newer kernels can support Panfrost and NPU drivers without breaking HDMI or other hardware — this remains uncertain in the source.
- Improved NPU drivers and broader tooling that would let standard AI frameworks use the on-board NPU more directly — not confirmed in the source if/when this will arrive.
Quick glossary
- NPU (Neural Processing Unit): A processor designed to accelerate machine learning inference tasks, optimized for low-precision numeric formats common in AI workloads.
- LPDDR5: Low-Power Double Data Rate 5, a mobile-oriented DRAM standard offering higher bandwidth and lower power than prior LPDDR generations.
- M.2 2280 NVMe: A common form factor (22×80mm) for high-speed PCIe-attached solid-state drives using the NVMe protocol.
- Vulkan: A low-overhead, cross-platform graphics and compute API commonly used for modern GPU-accelerated rendering.
- Panfrost: An open-source Mesa driver for ARM Mali/Immortalis GPUs providing kernel-space and userspace components for graphics acceleration.
Reader FAQ
Does the OrangePi 6 Plus ship with a desktop-ready OS?
A Debian Bookworm image with GNOME is available and provides a responsive desktop experience; an Ubuntu 24.04 image was planned but currently has no ETA.
Can I use the NPU with standard PyTorch or TensorFlow?
No — the reviewer reports that stock PyTorch/TensorFlow won’t work out of the box; the NeuralONE AI SDK and model conversion are required to target the NPU.
Does the board boot from NVMe out of the box?
The board required a firmware update before it would successfully boot from an NVMe drive for the reviewer.
Are graphics drivers ready for modern Vulkan applications?
Vulkan is available but limited to an early 1.3 implementation; full Panfrost support requires a newer kernel, which conflicts with some board patches.
ORANGEPI 6 PLUS REVIEW: THE NEW FRONTIER FOR ARM64 SBC PERFORMANCE 2025-12-27 By ekianjo So after our previous reviews (that started mainly around RISC-V since we are really interested in…
Sources
- OrangePi 6 Plus Review: The New Frontier for ARM64 SBC Performance
- Orange Pi 6 Plus board has a 12-core CPU, 45 TOPS AI …
- Benchmarking the Orange Pi 6 Plus Single Board Computer
- Orange Pi 6 Plus ARM Development Board Full specifications
Related posts
- WIRED’s Picks: Best Netflix Shows to Watch Right Now (Dec 2025)
- Carnegie Mellon team builds camera that focuses on multiple depths
- Tuft & Needle Original Hybrid Mattress Review — Soft Choice for Guests