I have been troubleshooting a roughly one-year-old PNY GeForce RTX 5070 Ti OC for about two months. When the card is under sustained heavy load, all three connected displays lose signal at the same time. Audio, Discord, networking, and the rest of Windows continue running, but the video never returns and I have to hard-reset the computer.
The issue appears in heavily modded or demanding games such as Minecraft with intensive shaders and Dragon Ball Sparking Zero at maximum settings. It does not occur in lighter games like Valorant, Overwatch, or Xenoverse 2, even during long sessions.
I have ruled out temperatures, voltage fluctuation, power-limit behavior, PCIe link instability, the 12V-2x6 cables, driver corruption, RAM and EXPO instability, and my original motherboard and PSU. The same failure occurs on a separate Ryzen 5 5600X/B550 system with a different EVGA 750W PSU and a fresh Windows installation.
The most repeatable trigger is OCCT's 3D Adaptive test. Once the load ramps past roughly 90–95%, the card fails every time while drawing around 250–260W. It never reaches its advertised power limit. Comparable reviews show healthy cards sustaining approximately 270–320W or more under similar workloads. Event Viewer shows no WHEA errors or display-driver events; only Kernel-Power Event 41 with BugcheckCode 0 after the forced restart.
I sent the card to PNY, where it reportedly passed more than 20 hours of testing using Steel Nomad, Superposition, and Valley across three 4K displays. However, those are fixed benchmark workloads rather than the variable-load pattern that reliably triggers my failure. The card also came back with a bent mounting bracket and a new fan noise that were not present before shipping. It still fails in exactly the same way at home.
Does this sound like a genuine hardware fault, possibly one that only appears during rapid load changes or a particular power-delivery condition, or is there another test or angle I should investigate before pursuing another warranty claim?
2 Answers
This strongly resembles a hardware-level GPU fault rather than a normal driver crash. The fact that every display dies together, the system remains partly responsive, the problem reproduces on a completely different platform, and OCCT triggers it consistently at a repeatable load point is compelling. A marginal VRM component, memory subsystem, GPU core, or board-level power signal could fail only under a particular transient or variable-load condition while passing conventional benchmarks.
The lower-than-expected power draw does not rule out a fault. The card may be crashing before it can pull more power, or a protection circuit may be cutting part of the board off. I would document the exact OCCT settings, video of the failure, monitoring logs, and the card's condition after the first RMA, then request another warranty evaluation or replacement. It may also help to ask whether PNY can run a variable-load test instead of relying only on fixed benchmarks.
You can sometimes restart the Windows graphics driver with Win+Ctrl+Shift+B. If it works, the screens should blink and return without rebooting, although it will not repair an underlying hardware problem.
I already tried that during the failure, but it does not recover the displays. A hard power cycle is still the only way to get the system usable again.