Common Issues¶

Failure modes that show up on unlocked CMP 170HX hosts. Get the card under 80 °C before you debug software. Each section is symptom → likely cause → first fix.

Card not seen / wrong PCI ID¶

Symptoms: Empty lspci NVIDIA line, or device IDs that look like an A100 (20b0 / 20b2 / 20b5 / 20f1 class) when you expected a 170HX.

Likely causes: Bad seat, no EPS power, host slot dead, you bought the wrong board, Secure Boot / early driver bind confusion.

First checks:

Reseat the card and the 8-pin CPU/EPS-style adapter.
Confirm PSU 12 V budget.
lspci -nn | grep -i nvidia on a stock driver. Expect 10de:20c2 (8 GB) or 10de:2082 (10 GB), subsystems 1585 / 1557.
Try another slot. Clear CMOS if the board POSTs without the card but hangs with it.

Memory still 8 GB / 10 GB after unlock¶

Symptoms: Installer exited 0, reboot done, nvidia-smi still shows 8192 or 10240 MiB.

Likely causes: Wrong module winning the load order; Secure Boot still on; warm reboot only; wrong --profile; driver not 610.43.02/03.

First checks:

modinfo -n nvidia
mokutil --sb-state
nvidia-smi

On Ubuntu-class hosts, DKMS under updates/dkms/ often beats updates/cmpunlocker/. Point depmod / module priority at the cmpunlocker updates path (ServeTheHome unlock thread and cmpunlocker issues cover the exact override). Cold power off, then boot. Confirm --profile=8gb vs 10gb matches the SKU.

Gen stuck at 1 after unlock¶

Symptoms: Memory unlocked, but LnkSta still 2.5 GT/s.

Likely causes: Build without Gen2 patches; host slot forced Gen1; ASPM / slot quirks; width still x4 (expected without caps).

First checks: Confirm your cmpunlocker commit actually includes Gen2 (README / changelog). Leave host PCIe gen on Auto. Width staying at x4 without the capacitor mod is normal. Gen and width are independent.

DKMS / module priority fighting cmpunlocker¶

Symptoms: Unlock works once, then a kernel or nvidia DKMS rebuild restores stock behavior; modinfo -n nvidia leaves the cmpunlocker directory.

Likely causes: DKMS rebuild overwrote or outranked patched modules.

First checks: After every kernel or nvidia package update, re-run the unlock install or re-assert the updates/cmpunlocker priority. Snapshot the rootfs before upgrades. Pin nvidia packages when you can.

Thermal abort / runaway¶

Symptoms: Sudden power-off, clocks collapse, temps climb through the 70s °C without plateau under load.

Likely causes: Passive sink in still air; dry waterblock; power limit raised to ~300 W before cooling is proven.

First checks: Aim high-static-pressure fans through the fins, or fill the loop. Lab hard stop near ~80 °C. Dry-run without coolant: power off within ~5 minutes. See Air Cooling / Water Cooling.

Waterblock short from missing pads¶

Symptoms: Card dead after waterblock install; shorts on 12 V / ground; host refuses to POST with the card.

Likely causes: Empty IC footprints left bare under block pillars; backplate installed without the saved PCIe bracket spacing.

First checks: Do not keep powering a suspected short. Tear down, inspect pads, cover every empty footprint the pillars can reach, verify bracket spacing. Procedure detail: Watercooling Installation and Water Cooling.

Driver upgrade broke unlock¶

Symptoms: Worked on 610.43.02/03; after an nvidia package bump, modules fail to build or unlock no longer applies.

Likely causes: Unsupported driver major/minor; patched modules rebuilt against the wrong tree.

First checks: Roll back to 610.43.02 or 610.43.03, reinstall cmpunlocker, cold reboot. Cache those packages locally. Read the current cmpunlocker README before chasing newer drivers.