In-depth Server Hardware Troubleshooting

Processor

• Can POST to BIOS (provides display) with motherboard, memory, and power supply

• Any visible scratches on any part of the CPU

• Thermal paste inside vent hole (this can be cleaned out)

• Does not allow system POST to BIOS

• Causes topology discrepancy with known-working parts (i.e. missing Memory DIMM slots when memory is clearly installed and tested in other systems)

Memory

• Can POST to BIOS (provides display) with motherboard, processor, and power supply

• Any visible scratches on any part of the Memory DIMM

• Does not allow system POST to BIOS

• (IF ECC/DDR5) ECC reports errors for installed DIMM

Power Supply (non-redundant)

• Can POST to BIOS (provides display) with motherboard, memory, and processor

• Any visible scratches on any part of the power supply

• Power cables worn out corners (i.e. someone jamming the cables in)

• Does not allow system POST to BIOS

• Voltage detector does not detect power while power supply is plugged into outlet and is switched ‘on’

Redundant Power Supply

(DO NOT TEST STANDALONE, ONLY IN COMPATIBLE BAREBONE)

• See MFR manual for ‘working’ LED activity (i.e. green LED/blinking)

• Allows barebone to power on (fans spinning)

• Any visible scratches on any part of the power supply (guiding ‘wings’ bent, scratch on large flat gold-contacts)

• See MFR manual for ‘failed’ LED activity (i.e. amber LED/static)

• Does not allow barebone to power on

Drive (HDD/SSD/U.2 SSD)

• Recognized in BIOS, OS, and SMART status Good/OK

• LIGHTLY USED

  • <1000 Power-on Hours

• Anything above can be categorized as ‘used’, but there are also other factors such as lifetime reads/writes (i.e. lots of writes would take a toll on SSD’s_

• Raw errors (SMART in cautious/bad)

Drive (NVMe M.2/PCIe SSD)

• Recognized in BIOS, OS, and SMART status Good/OK

• NVMe smart-log shows no:

  • no_err_log_entries

• LIGHTLY USED

  • <1000 Power-on Hours

• Could be critical temperatures logged, but that could be caused by the system it was installed in, not the drive’s fault

• Not recognized in BIOS or OS (make sure BIOS is set correctly to see NVMe devices; i.e. ‘legacy’ or ‘ops/bifurcation’ BIOS settings may need to be toggled)

• NVMe smart-log shows no_err_log_entries (check nvme error-log /dev/nvmeX for specifics)

Graphics Card

(non-computational/has display ports)

• Provides video display from each port (unless noted otherwise from MFR)

• Passes Furmark 24h

• Passes GPU Standalone Validation (10/5/3)

• Any major scratches on plastic casing or gold contact pins

• Missing emblems (I’d prefer them missing, but customer’s may see this as a ‘used’ product)

• Bent video display bezel/bracket (this can be replaced if we have the correct one)

• Does not provide video display

• Does not show up in lspci (Linux) or device manager (Windows)

• Falls off the BUS during GPU stress tests listed in ‘pass condition’

Graphics Card

(Computational/passive/no display ports; i.e. Tesla series)

• Passes GPU Standalone Validation (10/5/5)

• Can output nvidia-bug-report

• Any major scratches on plastic casing or gold contact pins

• Bent bezel/bracket (this can be replaced if we have the correct one)

• Does not show up in lspci (Linux)

Network Card (NIC)

• Provides IP via DHCP

• Negotiates network speeds as spec’ed (i.e. ethtool <port> | grep -i speed)

• This is all board usually, so bent bracket or missing LP (low-profile) or FL (full-height) bracket; full kits are usually sold with both

• Does not show up in lspci (Linux) or device manager (Windows)

• Does not provide, or drops, IP while system is under full load (for thermal)

Chassis

• Has all chassis screws and accessories

• All jumper pins attached and working

• Obvious scratches, bents, wear on chassis

• Missing fans

• Missing accessories

• Plate for mounting the motherboard is warped or missing stand-offs that prevents secure motherboard installation (motherboard touching aluminum/metal will short the board if improperly mounted)