RTX 4090's Usable VRAM Drops by 1.5GB; RMA Rejected Three Times, Card Returned with Only New Thermal Paste

Summary

A user named Ok-Dimension-741 on a forum reported that their RTX 4090's GDDR6X VRAM developed uncorrectable errors, reducing usable memory from 24GB to about 22.5GB. After sending the card back for repair three times, the manufacturer only reapplied thermal paste and returned it, refusing to replace the VRAM or the card. Windows Event Viewer logged uncorrectable GDDR memory errors in three separate partitions, while OCCT's VRAM test showed 4.7 million errors within 15 seconds. nvidia-smi's row remapping diagnostic revealed 33 uncorrectable errors but zero remapped rows, with 192 banks remaining. GDDR6X has built-in ECC that can detect and correct single-bit errors, but when multiple bits flip beyond ECC capability due to physical degradation or manufacturing defects, uncorrectable errors occur and data cannot be repaired. This incident highlights potential issues in high-end GPU quality control and after-sales service, raising concerns among gamers about the VRAM reliability of the RTX 4090 and the RTX 40 series. For players, VRAM errors can lead to graphical artifacts, driver crashes, or even data corruption. For the industry, this may prompt manufacturers to strengthen memory testing and reevaluate repair policies. Neither Nvidia nor the board partner has issued an official statement yet, so the follow-up is worth watching.