Compute Requirements
What kind of computer you need to run the Voyant API alongside the Carbon 30.
The Carbon 30 does its optical sensing and detection onboard and streams data to your host over Ethernet. The host assembles that data into the point cloud and runs your application on top.
Your requirements depend on the host’s role:
- A headless streaming/recording node — the pipeline via the Python, C++, or ROS bindings, or direct CLI recording to
.binwithvoyant_logger_binary— is light and runs comfortably on hardware as small as a Raspberry Pi 5. - Running the Voyant Visualizer raises the bar — its live 3D rendering wants a GPU plus more CPU and RAM. That, not the pipeline, is why laptops and workstations should be specced higher.
How much CPU it needs
The pipeline is lightweight. On our Raspberry Pi 5 reference, the full live client plus recording uses about 50% of one CPU core in a nominal configuration and a representative scene — leaving most of the four-core machine free for your own application. Load scales with frame rate and scene density, so size for the densest scene you expect and leave some margin.
Choose hardware by per-core speed, not core count. The point-cloud processing runs primarily on a single core, so a device with fast cores outperforms one with many slower cores. It does not need dedicated cores — it shares whatever is available with your application.
Power mode
Run the host in performance mode. You may see occasional drops in balanced mode, and likely will in a power-saving / battery-saver mode — a lowered CPU clock starves the single-core pipeline.
- Windows: Settings → System → Power & battery → Power mode: Best Performance.
- Linux: pick the Performance profile in your power settings (headless:
powerprofilesctl set performance).
Networking
The sensor delivers its stream as many small UDP packets. On any Linux host, if you see dropped data, raising the kernel UDP receive-buffer limit (net.core.rmem_max) eliminates burst packet loss under load — a one-time step, and the most effective fix. See Raising the UDP receive buffer (with the macOS and Windows notes).
The Visualizer is the heavy tool
The Voyant Visualizer is by far the most demanding piece of the toolkit: on top of the point-cloud pipeline it renders live 3D point clouds, which needs a GPU.
On a host without a supported GPU (e.g. a Raspberry Pi 5), the Visualizer can fall back to software rendering: launch with LIBGL_ALWAYS_SOFTWARE=1 and set point quality to Square. It works, but expect heavy CPU use (~3 cores on a Pi 5), so a GPU machine is strongly preferred — see the Visualizer guide.
For headless or resource-constrained deployments, run the client and recorder tools, or your own application through the Python or C++ bindings. These are substantially lighter than the Visualizer. When you need visualization from an edge deployment, stream or replay the data to the Visualizer running on a separate workstation.
Recommended hosts
| Host | Guidance |
|---|---|
| x86 — Intel Core i7 / Core Ultra 7-class or better, 16 GB RAM | Recommended for laptops, mini-PCs, and workstations. Comfortable headroom for the pipeline, the Visualizer, and your application. |
| Modern mainstream x86 (Core i5 / Ultra 5, U-series) | Runs the full pipeline fine in a performance power mode (see above); balanced or power-saving modes risk drops on this tier. |
| Raspberry Pi 5 | Recommended — our reference board. Runs the headless live client and recording pipeline with headroom to spare. |
| NVIDIA Jetson Orin (Orin Nano / Orin NX / AGX Orin) | Recommended. Same per-core class as a Pi 5 for the pipeline; the extra cores and integrated GPU are headroom for your workloads, not the pipeline. |
| NVIDIA Jetson Xavier | Workable. Validate against your frame rate and scene density before deploying. |
| Older / low-power boards (Jetson Nano, TX2, Raspberry Pi 4 and earlier) | Their weaker single-core performance may not keep up with the point-cloud pipeline. Contact us before deploying. |
Questions about a specific target platform? Reach out at: support@voyantphotonics.com