RTX 3080 no video output
common causes & diagnosis
Power delivery failure pathways
RTX 3080 no video output most commonly stems from collapsed voltage rails in the power delivery subsystem. The GA102 die requires 0.8–0.95V at VDD_CORE with minimal noise. Failures originate in the ISL6259 or TPS51125 PWM controllers, or in associated MOSFET stage collapse.
Key voltage rail checks
Before component-level diagnosis, establish baseline voltages with the card under minimal load (no CUDA, seated in a test rig):
| Rail / Test Point | Typical Voltage | Tolerance | Failure Indicator |
|---|---|---|---|
| VDD_CORE | 0.85V | ±0.05V | < 0.6V or floating |
| VDD_MEM | 1.35V | ±0.10V | < 1.2V or ripple > 50mV |
| VDD_PLL | 0.90V | ±0.08V | Unstable, drifts > 0.05V |
| VAUX_3V3 | 3.3V | ±0.2V | < 2.8V indicates PMIC fault |
| VAUX_5V | 5.0V | ±0.3V | < 4.5V or sag under load |
Measure with the PCIe power connectors fully seated. A missing 8-pin or 6-pin auxiliary connector results in immediate VDD_CORE starvation.
GDDR6X memory interface collapse
No video output paired with POST success and stable power rails often indicates VRAM controller or PHY failure. RTX 3080 uses 10 × Samsung K4Z80325BC-HC14 GDDR6X modules (8GB total). Memory controller sits in the GPU die; PHY terminators and training logic reside on the board.
Memory training failure diagnosis
When BIOS reports GPU present but driver hangs or no display appears, memory training has likely failed. Check VDDQ_MEM (supply for memory termination resistors) and VREF_MEM (reference voltage for read/write eyes).
- VDDQ_MEM: 0.675V (half of 1.35V VDD_MEM) — tolerance ±0.05V. Drift indicates capacitor leakage or resistor divider fault.
- VREF_MEM: Generated by TPS51911 or similar — typically 0.675V with ripple < 20mVpp.
- Memory ODT (On-Die Termination): Controlled via MAX5969 hot-swap controller on some models. Verify pull-up voltages on GPIO pins match schematic: typically 1.8V.
BIOS and NVVDD_AUX boot failure
RTX 3080 boots through a multi-stage power-up sequence controlled by firmware. If NVVDD_AUX collapses before GPU code execution, the card will not signal presence on PCIe even though the slot recognizes it.
Sequencing and PMIC faults
NVVDD_AUX (typically 1.8V, tolerance ±0.1V) supplies the IO pad ring, PMIC internal logic, and PCIe PHY. Collapse here is almost always a MP8756 or MP2888 PMIC failure:
- Output capacitor dry-joint or capacitance loss (>20% ESR rise).
- Feedback resistor divider drift (open in upper or lower resistor).
- Enable (EN) pin not driven high during power-good window.
- PMIC soft-start function stuck or disabled.
Measure PMIC feedback pins against GND with a 100mV DC-coupled probe. A clean 0.5 – 1.5V signal indicates the regulator is attempting to regulate. Flat rail or oscillation > 200kHz suggests IC failure or parasitic L-R instability.
VBIOS and EEPROM corruption
RTX 3080 VBIOS resides in an AT25M02 or W25Q64FV SPI flash. Corruption—especially in the power table or thermal management sections—causes instant GPU hangup during initialization.
Detection and recovery
If the card boots to BIOS POST but the driver hangs on load:
- EEPROM data lines stuck high or low (SPI bus shorted to GND or 3.3V).
- Power table entries pointing to invalid clock/voltage states (e.g., ASIC requesting 1.2V at base clock, causing immediate VRM collapse).
- Thermal throttle register corrupted, limiting GPU to 10% clocks.
Verify with external programmer. Most successful recoveries involve re-flashing a known good BIOS from NVIDIA's archive. However, if flash IC is mechanically damaged (tombstoned, cold joint), re-flashing will fail. Inspect U1450 (or equivalent EEPROM location) for lifted pads.
Systematic troubleshooting workflow
Follow this sequence to isolate the fault:
- Power rails: Measure VDD_CORE, VDD_MEM, VAUX_3V3 at idle. If any are < 80% nominal, proceed to PWM controller test.
- BIOS detection: Does the system BIOS see the GPU at all? Run
lspcior check BIOS GPU menu. If absent, PCIe PHY or PMIC failure. - Driver load: Linux:
dmesg | grep NVIDIA. Windows: Device Manager. Does the device enumerate? If not, power or VBIOS corruption. - Memory training: Boot with
nvidia-smior equivalent. If command hangs, memory interface failed. Measure VREF_MEM and VDDQ_MEM. - Scope analysis: Capture VDD_CORE ripple during transient load (0.5V step). Ripple > 100mVpp indicates ESR saturation or inductor saturation in the power stage.
- Thermal: Measure GPU die temp at idle (should be room temp ±5°C). If sensor is >60°C at idle, thermal shutdown is likely active—check for solder bridging on temp sensor (typically an NTC thermistor).