[N1x Linux Laptop][Onboard] Preflight rejects genuine N1x hardware — PCI identity check hardcoded to a single GPU device ID

[hulynn]

Description

NemoClaw's N1x platform-identity check (used both by the installer's "Deferred N1x preview" offer and by the onboarding wizard's preflight) hardcodes a single accepted GPU PCI device ID (10de:2e2a). Genuine N1x/RTX Spark hardware whose GPU reports a different PCI device ID is rejected outright: the installer silently skips the Express/preview offer, and the standard onboarding wizard hard-fails at preflight with no way to proceed. There is no flag or override to bypass this check.

Related to #8574, whose design comment (2026-08-12) intentionally scoped N1x GPU identification to exactly PCI device 10de:2e2a — the only device ID known at design time. This issue reports that the scope needs to expand to cover additional N1x/RTX Spark GPU device IDs as newer hardware batches ship, otherwise those units cannot onboard NemoClaw at all.

Platform scope: Reproduced on N1x/RTX Spark only. Confirmed on two physical N1x units from the identical FastOS build (N1x FASTOS 1.25.0, dated 2026-08-20T12:12:02Z, BUILD_TYPE=developer): Device A (GPU PCI device 10de:2e2a) passes this check and proceeds further into onboarding; Device B (GPU PCI device 10de:2e03) is rejected outright at preflight, even though nvidia-smi correctly identifies Device B's GPU by its full product name ("RTX Spark N1X"), unlike Device A which only reports a generic placeholder name.

Regression: No — never worked for this hardware. The check was intentionally scoped to a single PCI device ID before this second hardware batch existed; there is no version where this specific hardware has onboarded successfully.

Environment

Device A (passes preflight):      N1x / RTX Spark, GPU PCI device 10de:2e2a, nvidia-smi reports "NVIDIA JMJWOA-Generic-GPU"
Device B (rejected at preflight): N1x / RTX Spark, GPU PCI device 10de:2e03, nvidia-smi reports "NVIDIA RTX Spark N1X (6144-core Blackwell RTX GPU)"
OS:            N1x FASTOS 1.25.0 (both devices; DATE=2026-08-20T12:12:02+00:00, BUILD_TYPE=developer, PRODUCT_NAME=RTX Spark)
Architecture:  aarch64 (both devices)
Node.js:       v22.23.2
npm:           10.9.8
Docker:        Docker version 29.2.1, build a5c7197
OpenShell CLI: openshell 0.0.106
NemoClaw:      nemoclaw v0.0.114
OpenClaw:      N/A (onboard not completed on Device B; Device A also has not reached OpenClaw agent setup — blocked further along by a separate, unrelated vLLM preflight issue)

Steps to Reproduce

1. On an N1x/RTX Spark host whose GPU reports PCI device ID 10de:2e03 (check the VGA controller entry under lspci -d 10de:), run:

   curl -fsSL https://www.nvidia.com/nemoclaw.sh -o nemoclaw.sh && NEMOCLAW_INSTALL_TAG=v0.0.114 bash nemoclaw.sh

2. Observe steps [1/3] Node.js and [2/3] NemoClaw CLI complete normally, but the installer never prints [INFO] Detected N1x. and never offers the "Deferred N1x preview" prompt — it silently falls through to the standard interactive onboarding wizard instead.
3. At the "Select your agent" prompt, accept the default (1, OpenClaw).
4. Observe onboarding step [1/8] Preflight checks immediately hard-fails with no way to proceed.

Expected Result

The installer should recognize any genuine N1x/RTX Spark unit — identified by the trusted /etc/fastos-release marker plus an NVIDIA GPU on an Arm64 host — and either offer the "Deferred N1x preview", or (if declined) let the standard wizard's preflight pass the platform-identity check, regardless of which specific GPU PCI device ID that hardware batch reports.

Actual Result

[1/8] Preflight checks
  ──────────────────────────────────────────────────
  ✗ N1x requires the trusted FastOS marker, NVIDIA PCI identity, Arm64 host, and available NVIDIA GPU.

  Onboarding did not finish. Resume from the step that failed with:
    nemoclaw onboard --resume --name {sandbox}
  Completed steps are skipped; pass --fresh instead to start over.
[ERROR] Onboarding did not complete successfully.

Onboarding cannot proceed at all on this hardware — there is no workaround or flag to bypass the check.

Logs

Not captured — no separate log file is produced before this point in onboarding; the terminal transcript above is the complete failure output.

[sandl99]

Maintainer decision and root cause

Decision: Accept

Reason and placement: Accept as a core NemoClaw bug. N1x platform identity must not depend on one GPU PCI device ID when the trusted FastOS marker, Linux Arm64 host, NVIDIA vendor, display-class PCI device, and available NVIDIA GPU establish the accepted boundary.

Accountable maintainer: @sandl99

Root cause: PR #8903 encoded the only device known at the time, 10de:2e2a, as an exact match in both the installer and TypeScript identity detectors. The newer hardware batch reports 10de:2e03, so the installer misses the Deferred preview. Onboarding still finds the trusted FastOS marker, keeps the host in the N1x qualification path, and blocks generic Linux admission as host.platform.n1x_unqualified.

Accepted scope: Remove the GPU device-ID equality from both detectors. Preserve the trusted, bounded, root-owned, non-symlink FastOS marker; exact NAME="N1x FASTOS"; Linux Arm64; NVIDIA vendor; display-class PCI device; available NVIDIA GPU; fail-closed unreadable evidence; and explicit Deferred managed-vLLM intent. Standard onboarding after declining the preview remains outside this fix.

Validation plan:

• Add TypeScript source coverage for an N1x fixture whose GPU reports 10de:2e03.
• Add installer integration coverage for the same PCI identity.
• Preserve negative coverage for non-NVIDIA vendors, non-display PCI classes, untrusted markers, wrong architectures, unavailable GPUs, and unreadable evidence.
• Run the focused N1x identity, platform-qualification, and installer suites.
• Require physical 10de:2e03 N1x evidence that the installer detects N1x and onboarding passes platform identity before publication makes a broader operational claim.

[JoyceChenNV]

Following Device C has the similar issue:
Device C (rejected at preflight): N1x / RTX Spark, GPU PCI device 10de:2e02, nvidia-smi reports "NVIDIA Graphics Device"

[sandl99]

Physical N1x validation update

Tested revision: a3c5098 (installer banner: v0.0.117-42-ga3c509861)

Result: PASS for installer detection and onboarding platform-identity admission.

Observed on a physical N1x device:

• The installer printed [INFO] Detected N1x.
• The Deferred N1x preview prompt appeared and was accepted.
• All [1/8] Preflight checks passed.
• Preflight detected NVIDIA Graphics Device, 63076 MB.
• Onboarding advanced to gateway startup and inference-provider configuration.
• The managed gateway missed its health deadline, then the standalone fallback became healthy.
• The vLLM image and 23.5 GB model download completed. vLLM startup was still in progress when the captured transcript ended.

This run proves the two behaviors requested by the accepted validation plan for revision a3c5098: installer detection and platform-identity admission.

It is not yet current-revision evidence or a complete Express E2E result. At the time of this update, PR #10099 points to d1dcc8d. The captured transcript also does not include explicit lspci output confirming PCI ID 10de:2e03. A rerun on the latest PR commit will record the full commit SHA, FastOS and Arm64 identity, PCI identity, and the complete result.