More evidence of poor quality control in NVIDIA GeForce RTX 30 series cards emerges
NVIDIA's high-end Ampere graphics cards can run hot, especially when it and its partners have a lapse in quality control (QC). Last month, NVIDIA left a finger glove in an RTX 3090, which caused its VRAM to reach 110 °C, exceeding Micron's safe operating temperature by over 15 °C. Now, two other malfunctioning cards have shown up online with VRAM cooling issues.
It turns out that both had misaligned or non-existent VRAM thermal pads, an essential part to prevent a card's memory from overheating. In the case of one Founders Edition card, NVIDIA had applied a thermal pad so poorly that it spread onto the GPU die. Unsurprisingly, this manufacturing defect was causing the GPU to overheat, too.
As the photo below shows, ASUS had failed to include a thermal pad over one set of VRAM chips, leaving them pressed against the heat spreader. In both cases, replacing the thermal pads fixed the overheating issues, but this is not something that someone should need to do on any graphics cards, let alone an RTX 3080 or an RTX 3080 Ti.
Ultimately, we would only recommend disassembling your graphics card if you already have experience in doing so. NVIDIA and its partners do not necessarily publish service guides, although there are third-party disassembly videos available. Nonetheless, reaching the VRAM on the GPU side of the mainboard can take some time, as Gamers Nexus has shown with the RTX 3090 Founders Edition.