Jetson Nano GPU Accessibility and Kernel Module Audit

Tegra T210 Driver Stack Status

The node900–node903 hosts are Jetson Nano 2GB dev kits powered by NVIDIA's Tegra T210 SoC. Each includes an integrated Maxwell architecture GPU, yet the GPU stack is completely inaccessible on all four devices. Comprehensive audit confirms:

• nvidia-smi: Command not found; no nvidia-utils package installed.
• OpenCL: No platforms detected via clinfo — libOpenCL.so either absent or unable to enumerate backends.
• CUDA runtime: No CUDA libraries available; no nvcc compiler; no cudnn stack.
• Kernel modules: The nvidia-drm and tegra-related kernel modules are not loaded. dkms or proprietary driver installation is needed to populate these.

Root Cause Analysis

The Jetson Nano platform requires NVIDIA's proprietary L4T (Linux for Tegra) driver stack — the standard open-source nouveau driver does not support Maxwell GPU features on T210. Remediation requires:

1. Installing the correct BSP (Board Support Package) matching the JetPack version for each device.
2. Loading nvidia-drm kernel module with modprobe after installation.
3. Rebooting to activate GPU hardware access paths.

Impact on Lab Infrastructure

The node90x cluster was provisioned as a low-cost GPU research platform for the lab. Without accessible GPUs, this resource is currently limited to general-purpose compute workloads (SSH access verified on node900 and node903). node901 remains offline (port 22 timeout) and node902 has been offline since 2026-08-27 pending physical power/link inspection.

Remediation Priorities

• P1: Bring node902 back online (physical inspection).
• P2: Install L4T BSP on node900 to validate GPU access path.
• P3: Replicate verified BSP configuration to remaining devices via ansible-test common role validation.

Next Steps

Today's work concludes the Sep 13–20 gap period. The lab infrastructure landscape has evolved significantly: Ansible testing has migrated from Molecule to native ansible-test, NFS keytab deployment remains a known gap, and the stargate controller role requires cleanup before collection release. The cold-storage replication topology is under active design with symlink-based cross-host visibility replacing bind mounts.