Problem
nemoclaw onboard fails on Jetson Orin Nano Super (8GB, JetPack 6.x, kernel 5.15.148-tegra) with two issues:
-
GPU not detected: nvidia-smi --query-gpu=memory.total returns [N/A] on Jetson (unified memory), so detectGpu() returns null and preflight reports "No GPU detected".
-
k3s panic on startup: The OpenShell gateway container ships iptables v1.8.10 in nf_tables mode, but the Tegra kernel lacks nft_chain_filter and related modules. The k3s network policy controller panics immediately:
iptables v1.8.10 (nf_tables): RULE_INSERT failed (No such file or directory): rule in chain FORWARD
panic: ...network_policy_controller.go:412
-
Default model too large: The default Ollama model nemotron-3-nano:30b does not fit in 8GB unified memory.
Expected behavior
nemoclaw onboard should detect the Jetson GPU, start the gateway successfully, and default to a model that fits in the available memory.
Environment
- Device: NVIDIA Jetson Orin Nano Super Developer Kit
- Memory: 8GB unified (CPU/GPU shared)
- OS: Ubuntu 22.04, kernel 5.15.148-tegra
- JetPack: 6.x
- NemoClaw: latest (npm)
- nvidia-smi output:
Orin (nvgpu), memory.total: [N/A]
Problem
nemoclaw onboardfails on Jetson Orin Nano Super (8GB, JetPack 6.x, kernel 5.15.148-tegra) with two issues:GPU not detected:
nvidia-smi --query-gpu=memory.totalreturns[N/A]on Jetson (unified memory), sodetectGpu()returnsnulland preflight reports "No GPU detected".k3s panic on startup: The OpenShell gateway container ships iptables v1.8.10 in nf_tables mode, but the Tegra kernel lacks
nft_chain_filterand related modules. The k3s network policy controller panics immediately:Default model too large: The default Ollama model
nemotron-3-nano:30bdoes not fit in 8GB unified memory.Expected behavior
nemoclaw onboardshould detect the Jetson GPU, start the gateway successfully, and default to a model that fits in the available memory.Environment
Orin (nvgpu), memory.total:[N/A]