You just dropped two grand—maybe three, if you were desperate enough to buy from a scalper—on a brand-new NVIDIA flagship. You unboxed it, marveled at the sheer weight of the Blackwell silicon, and slotted it into your rig. Then, you installed the latest drivers. Suddenly, the screen went black. No signal. Not after a reboot, not after swapping cables, and not even after clearing CMOS.
The card is dead. Or at least, that’s what it looks like.
For a few weeks now, the enthusiast community has been on fire over rtx 5090 5090d bricked issues that seem to turn the world’s most powerful GPUs into expensive paperweights the moment they touch modern software. It’s not just an isolated manufacturing defect; it’s a weird, systemic collision between cutting-edge PCIe Gen 5 standards and early-stage Blackwell firmware.
Honestly, it's a mess. If you're currently staring at a dark monitor or debating whether to hit "express install" on that new driver, here is the ground-level reality of what’s actually failing.
The Driver Death Trap
Most of these bricking reports follow a hauntingly similar script. Everything works fine in the BIOS. You might even boot into Windows using the basic Microsoft display adapter driver. But the second the official NVIDIA GeForce driver finishes its installation, the card "vanishes."
Users on Chinese forums like Chiphell and Bilibili—where the RTX 5090D (the China-specific variant) is most common—were the first to sound the alarm. They reported that after the driver handshake, the GPU is no longer detectable. Not in Device Manager. Not in the BIOS "Board Explorer." Nowhere.
Why is this happening? It’s likely because the 50-series is NVIDIA’s first "true" PCIe 5.0 consumer architecture. In the BIOS or under basic drivers, the card often runs at Gen 4 or even Gen 3 speeds just to keep things stable. When the full driver kicks in, it tries to negotiate the full Gen 5 bandwidth. If your motherboard, your riser cable, or the GPU’s own signal integrity isn't 100% perfect, the handshake fails so hard that the card enters a protected "zombie" state.
Why the RTX 5090D is Getting Hit Harder
The RTX 5090D is the "Dragon" edition, a cut-down version meant to satisfy US export bans to China. While it has fewer cores, it shares the same PCB and power delivery as the global 5090. Interestingly, the bricking reports surged among 5090D owners first.
👉 See also: Star Cores Galactic Zone Grounds: Why We Are Still Guessing About the Milky Way's Engine Room
Some tech experts, including Roman "der8auer" Hartung, have hinted that board partners were incredibly rushed for this launch. We're talking about manufacturers having less than a week to finalize VBIOS and drivers before shipping. When you're dealing with the tighter signal tolerances of PCIe 5.0, that's a recipe for disaster.
The 5090D seems to be a victim of "first-batch" syndrome. Many of the early cards from brands like Colorful and Manli used firmware that didn't play nice with the shared PCIe lanes on modern motherboards. If you have an M.2 SSD stealing lanes from your top PCIe slot, the 5090D basically has a mid-life crisis and gives up.
The $1,000 Reset Bug and Hardware Flaws
It’s not just standard gaming setups. Even professional users and virtualization geeks are getting burned. CloudRift, a GPU provider, actually offered a $1,000 bounty because their RTX 5090s were bricking—or "soft-locking"—whenever they tried to reset a Virtual Machine.
Basically, the Blackwell chip fails to handle a "Function Level Reset" (FLR). The card gets stuck in a loop, starts throwing "unknown header" errors, and requires a full physical power cycle to wake up. For a card marketed as a professional-grade titan, having to flip the PSU switch every time you restart a VM is a joke.
Is it actually "bricked" though?
In the strictest sense, "bricked" means the hardware is permanently destroyed. In about 90% of these cases, the card isn't actually dead—it's just invisible.
There have been confirmed reports of some cards suffering from IC burn damage (the "real" bricking), but most users are finding that their cards "revive" if they move them to an older PCIe Gen 4 motherboard. This proves the issue is a signaling conflict. The card is so confused by the Gen 5 noise that it stops talking to the system entirely.
How to Save Your 5090 (or Prevent the Brick)
If you have a card that has already gone dark, don't panic and file for an RMA just yet. There’s a specific sequence of steps that has saved dozens of cards from the "paperweight" pile.
1. The Gen 4 Downgrade
The most successful fix is forcing your motherboard to ignore PCIe 5.0.
- Shut down the PC.
- If you can't see the BIOS, pull the 5090 out and use your CPU's integrated graphics.
- Go into your BIOS settings (usually under Advanced > PCIe Configuration).
- Change the slot speed from "Auto" or "Gen 5" to "Gen 4."
- Save, exit, and plug the 5090 back in.
Most of the time, the card will suddenly reappear.
2. Watch the 12V-2x6 Connector
We're also seeing a repeat of the 40-series melting issues, but with a twist. The new 12V-2x6 connector is supposed to be safer, but the 5090 pulls so much current (up to 575W stock) that any loose connection is a fire hazard. Some "bricking" is actually just the card's internal sensors detecting a massive voltage drop at the connector and shutting down to prevent a fire.
3. Avoid "Beta" VBIOS
There are a lot of unverified VBIOS files floating around TechPowerUp and various Discord servers claiming to "unlock" the 5090D or fix stability. Avoid them. Flashing a 5090D with a global 5090 VBIOS is the fastest way to turn a "soft-brick" into a "hard-brick" that even NVIDIA support won't help you with.
The Actionable Reality
NVIDIA hasn't issued a massive recall, and they probably won't. They’ll likely iterate on the drivers and hope the motherboard manufacturers release BIOS updates to clean up the PCIe 5.0 signaling.
If you are buying a 5090 or 5090D right now, do not use a PCIe riser cable unless it is specifically certified for Gen 5 (and even then, maybe don't). Stick the card directly into the motherboard. Before you even install the card, update your motherboard’s BIOS to the absolute latest version.
Most importantly, if you’re currently running a 5090 without issues, turn off automatic driver updates. Let the brave souls on Reddit be the crash test dummies for the next few months. If your card works today, keep it on that driver until the community confirms the next one doesn't come with a "black screen of death" side effect.
Next Steps for Owners:
Check your current PCIe link speed in GPU-Z. If it's fluctuating or showing errors while under load, go into your BIOS and manually lock it to Gen 4. You’ll lose maybe 1% of performance, but you'll gain a card that actually turns on tomorrow.