4 sourced reports, with what to check before you buy. Each one links to the page it was reported on.
Costly or dangerous
RTX 5090 12+4-pin power connector melting — repair centre receives seven bricked cards in two days
A single repair shop received seven bricked Nvidia cards in two days, mostly RTX 5090s across multiple brands including the Founders Edition, all with burnt 12+4-pin interfaces, some melted into the socket so the cable had to be snipped off.
What to check. Before buying, budget for a PSU/cable with a fully seated 12V-2x6 connector and check the 12V readout in GPU-Z; the repairman notes that if the 12+4-pin voltage drops well below 12 V (to ~11.3–11.4 V) the connector is likely failing and should be checked by hand for heat. This is the highest-cost failure mode — a melted connector can brick the card and damage the PSU.
RTX 5090 12V-2x6 connector current imbalance — two wires carry 22 A+, connector reaches 150 °C
Der8auer measured on a 5090 Founders Edition that two of the six 12V wires carried over 22 A while the rest sat around 2 A, and the PSU-side connector reached 150 °C, against a per-pin design of ~9.5 A.
What to check. The 12V-2x6 connector does not balance current across its six wires, so a single high-resistance pin can carry most of the load and overheat. This is the mechanism behind the melting reports; a shopper cannot see it without thermal/current instrumentation, so it is a real hidden risk on any 5090.
RTX 5090 high idle power draw (17–44 W) and temperature-dependent idle behaviour
Owners report idle draw between 17 and 44 W, and the draw rises with temperature over ~30 minutes as the cooler warms up; the Igor's Lab review of the MSI SUPRIM SOC measured 29–34 W idle and called it a possible driver problem.
What to check. Idle power draw is unusually high and temperature-dependent, which matters for a quiet/low-power build and for electricity cost in the EU. It is not a catastrophic failure but a hidden operating-cost and noise consideration; some owners reduce it with undervolting or forcing fans on.
RTX 5090 missing ROPs — ~0.5% of cards affected, up to ~4% performance loss
NVIDIA confirmed a rare production anomaly where ~0.5% of RTX 5090/5090D and 5070 Ti cards ship with one fewer ROP than specified, costing up to ~4% raster performance; NVIDIA says affected consumers can contact the board maker for a replacement.
What to check. A buyer cannot see this without checking GPU-Z (ROP count should be 176 on the 5090). If the card is affected, it is a performance defect that is RMA-able, so checking ROPs on arrival is a cheap way to catch it.
These reports come from the pages linked under each item: owner forums, repair guides and workshops. They are not a diagnosis of any one unit, and a careful check of the one you are buying still applies.
Text from the Kriko Graphics cards catalog, version 0.1.2, under the CC-BY-SA-4.0 license.
