I have an NVIDIA RTX 5090 Founders Edition (Serial: 1791425038200) exhibiting an abnormal idle power draw of ~126-127W with fans running at 30% and zero active processes, compared to 7-10W on a healthy identical unit. The defect has been confirmed across extensive testing: the issue followed the card when swapped between PCIe slots on an AMD EPYC 9654 server (ruling out slot/riser causes), persisted after ruling out X server activity, nvidia-drm modeset, snd_hda_intel HDA audio binding, PCIe power management settings, and ECC errors, and most conclusively reproduced identically on a completely separate AMD AM5 consumer platform with a fresh driver installation — confirming this is a hardware defect independent of OS, platform, or configuration. Under full compute load via gpu-burn the card is severely power-throttled to ~585 MHz SM clock versus ~2940 MHz on the healthy unit, while simultaneously hitting its 450W power cap, resulting in approximately 5x lower performance despite near-ambient temperatures (30-33°C). Both cards run identical VBIOS (98.02.2E.00.03) and driver (580.126.09). I have detailed telemetry logs from both idle and load testing available if helpful. Support indicated I had to post this information here for them to be able to help me with an RMA. Any guidance on next steps for an RMA would be greatly appreciated.
Is anyone from the developer team able to comment at all on this.
Customer service is refusing to discuss an RMA since it’s a Linux system and no one has responded here.
I suggest you just re-test this on Windows and work with customer support from there. Forget about the Linux aspect to make things easier from a support POV.
I don’t think you will get any official support via this forum. This place is more for driver bugs feedback and such.
I appreciate you responding. That’s what I was worried I was gonna have to do.
Same issue here with my RTX 5090 FE. Idle power is around 60W on Ubuntu, while my other normal 5090 FE cards are only around 6W. Under training load, it reports full power, but stays cool with low fan speed and runs slower than the good cards.
Did NVIDIA ever approve RMA for your card, or did they ask you to reproduce it on Windows first?