I'm troubleshooting recurring crashes with a brand-new ASUS ROG Astral RTX 5090 OC. The system uses a Ryzen 7 9800X3D, ASUS ROG Strix B850-E Gaming WiFi, Corsair RM1000x SHIFT 1000W, 32 GB of DDR5-6000, and Windows 11 with NVIDIA driver 610.88.
The GPU can run games for hours, but when the problem appears, Escape from Tarkov often crashes within a couple of minutes of loading into a raid. I get a black screen or driver timeout, followed by a reboot or BSOD. Event Viewer repeatedly reports nvlddmkm Event ID 14 errors, including:
- PCIE P2PREQ, Uncorrectable SRAM Error
- PCIE REORDER, Uncorrectable SRAM Error
- An uncorrectable ECC error in the GPU's PCIe P2PREQ and REORDER units
These are followed by nvlddmkm Event ID 153 GPU reset errors and eventually VIDEO_TDR_FAILURE (0x116). WinDbg points to nvlddmkm.sys, with Arg3 0xC000009A.
So far I have reinstalled Windows, installed fresh AMD chipset and NVIDIA drivers, reseated and cleaned the GPU and PCIe slot, checked the power connector, moved an NVMe drive so the graphics slot runs at PCIe 5.0 x16, tested PCIe Gen4, reduced the memory speed from DDR5-7200 to 6000, and removed all GPU tuning software. HWiNFO generally shows no PCIe receiver, replay, LCRC, TLP, DLLP, or other corrected or uncorrected link errors. Temperatures and 12V readings also look normal.
Synthetic GPU and VRAM tests, including OCCT and FurMark, have passed, even though certain games can trigger the issue. The most unusual part is that a normal reboot often does not help, but shutting the PC down completely and switching the PSU off for a while usually restores stability for a couple of days. That makes me wonder about a stuck GPU firmware or GSP state, PCIe Gen5 compatibility, the power cable or PSU, the motherboard BIOS, or a faulty card.
The card has never had an unofficial BIOS flashed. I'm planning to try another official Corsair 600W cable, update to ASUS's newer VBIOS, test PCIe Gen4 again, try reduced GPU power and clock settings, and possibly replace the PSU with a higher-wattage unit. Has anyone encountered these specific P2PREQ or REORDER uncorrectable SRAM/ECC errors, and did a VBIOS update, Gen4 mode, cable or PSU replacement, motherboard BIOS change, or GPU replacement resolve them?
3 Answers
Because the card is nearly new and the errors point to uncorrectable faults inside PCIe-related GPU units, I’d test the physical and firmware path before assuming Windows is responsible. Try the newer official VBIOS, the other performance-mode switch position, and force the slot to PCIe Gen4. If Gen4 makes the crashes disappear, that strongly suggests a Gen5 link or signal-integrity issue involving the card, motherboard, slot, or power delivery. Also test with reduced power, core, and memory clocks; if that changes the behavior, document it for a warranty claim.
A 1000 W supply is within the usual recommendation, but a high-end RTX 5090 can draw around 600 W by itself and transient behavior matters. The rest of the system can push the supply closer to its limits, especially with a power-hungry CPU, pumps, and many fans. Try a different, properly seated 600 W cable and, if possible, test with a known-good higher-capacity PSU. Make sure the connector is fully latched and not sharply bent near the plug. A full PSU shutdown clearing the problem could indicate that power cycling is resetting a fault state, although it does not prove the PSU is defective.
I switched from the 90-degree Type 5 cable to the original Corsair cable, and the system has been stable for about two days so far. I’m also testing a 1500 W PSU as a precaution, but I’ll need more time before deciding whether the cable actually fixed it.
Keep several minidumps from the VIDEO_TDR_FAILURE crashes and compare them rather than relying on one analysis. They may not identify the exact hardware fault, but repeated dumps can confirm whether the same NVIDIA reset path is involved. Since Windows was freshly installed and multiple drivers were tried, software is becoming less likely. If the issue remains after the official VBIOS, current motherboard BIOS, alternate cable, Gen4 testing, and a known-good PSU, I would pursue an RMA for the graphics card. Repeated uncorrectable ECC/SRAM errors are not normal behavior.
The card was bought brand new about two and a half months ago, but the crashes only started around two weeks ago. I have not flashed anything except official BIOS versions. OCCT VRAM and GPU stress tests pass, so the fact that games trigger it while synthetic tests remain stable is what makes it difficult to isolate.