The GeForce RTX 4090 is hands down the best graphics card that money can buy for the time being. Unfortunately, while the Ada flagship graphics card is fine, the included 4×8-pin to 1×16-pin power adapters reportedly pose a fire hazard, according to multiple user feedback.
Previously, two GeForce RTX 4090 owners have reported experiencing 16-pin power adapter meltdowns. Both owned custom models, with the first owner rocking a Gigabyte GeForce RTX 4090 Gaming OC and the latter user with an Asus TUF Gaming GeForce RTX 4090 OC Edition. The third report comes from a Facebook user (via Hassan Mujtaba (opens in new tab)), who coincidentally has an Asus TUF Gaming GeForce RTX 4090. According to the owner’s recount of the facts, he was benchmarking when he smelled the smoke from the power adapter. Fortunately, he could pull the connector out before it did any damage to the graphics card.
Nvidia has launched an investigation into the matter, and with good reason, since meltdowns are starting to become widespread. The user feedback shows that the problem may reside with the design of the 12VHPWR power connector. Bending the cables too close to the connector seemingly causes some terminals to loosen up, leading to uneven mating. In addition, it unbalances the load across the other terminals. PCI-SIG documented the potential thermal variance issue long before the 12VHPWR power connector debuted on the GeForce RTX 4090. Therefore, it certainly comes as a shock that the problem is still present on the finished product.
According to the PCI-SIG slides, which Seasonic shared on its Bilibili account (opens in new tab), the standards body stated that multiple suppliers and designs failed the tests. Cables, even if they have low cycles, didn’t fail as long as they weren’t bent. PCI-SIG tested the cable bending problem at approximately 30mm from connectors to the graphics card and power supply. However, even at that distance, PCI-SIG observed similar meltdowns. Exceeding 40 connection cycles on the connector also cause the problem. Do note that PCI-SIG tested on a “benchtop/lab” environment, which we assume is an open-air test bench. It’s not the same setting as a conventional PC case, but the results are what is essential here.
PCI-SIG’s tests revealed that the power connector was overheating at the mating point. The problem affected both rows of pins. PCI-SIG performed the tests at an ambient temperature of 26 degrees Celsius and took readings from the hot spots at 2.5 hours. The end of the cable connected to the power supply showed readings between 51.3 to 52 whereas the end to the graphics card peaked at 150.2. PCI-SIG observed melting between 10 to 30 hours. The unbalanced current resulted from the resistance variation between the different pins. The standards body noted that bending led to high resistance in the other pins, causing the current to transfer to the lowest resistance.
When excessively bending the cable, the loss of mating contact can happen to 6-pin or 8-pin PCIe power connectors. It’s nothing new. However, the problem seems more prevalent with the 12VHPWR power connector so users should pay extra attention to the installation. Custom cable manufacturers, such as CableMod, recommend a minimum distance of 35mm from the connector before bending the cable.
German publication Hardwareluxx (opens in new tab) has gotten its hands on power adapters from Nvidia, be quiet!, and Asus. Unfortunately, the adapters have different designs. Out of the three reported meltdowns, one was from Gigabyte, and the other two were from Asus. Without more cases, it’ll be hard to pinpoint if a specific OEM is at fault, though. In addition, three Hardwareluxx forum users have reported problems with the be quiet! adapter. One user reportedly attached a cable tie at a 40mm distance from the connector, but one of the pins still came out easily. Be quiet! has replaced the user’s adapter and will investigate the issue.
GeForce RTX 4090 owners should probably recheck their graphics card’s installation and avoid bending the cable for the 16-pin power adapter where possible. Hopefully, Nvidia will get to the bottom of the issue quickly and provide consumers with a solution.