RTX 5090 random black screen + 100% fans + hard lock — sometimes no POST until AC power is removed
retained text
I am looking for help diagnosing a very intermittent RTX 5090 problem that has become significantly more frequent over the last few months. The system was essentially stable for the **first \~6 months after it was built**. After that, GPU-related hard-locks started occurring occasionally. Over the following **\~5 months**, they gradually became more frequent. At this point the problem can happen **every 1–2 days and sometimes almost daily**. The symptom I am most concerned about is not just the black screen itself, but what sometimes happens **after the crash**: **black screen → GPU fans 100% → complete hard lock → hard reset → no POST/display → normal reboot does not recover it → complete AC power removal for \~20–30 minutes → system POSTs normally again** I am trying to determine whether this points toward an RTX 5090 hardware/GSP/PCIe/power problem rather than a normal Windows display-driver crash. # GPU * ASUS TUF Gaming GeForce RTX 5090 * 32 GB VRAM * Non-OC model * Current NVIDIA driver: **616.92** # Other hardware * Intel Core Ultra 9 285K * ASUS Z890-F Gaming WiFi * 64 GB G.Skill DDR5 * XMP currently OFF * ASUS ROG STRIX 1200P Gaming * 1200 W / 80+ Platinum / ATX 3.1 * Native 12V-2x6 / 16-pin GPU cable * Windows 11 * Latest motherboard BIOS / Intel ME # The GPU is mounted on a Gen5 riser This may be important. The RTX 5090 is vertically mounted using: **Lian Li VG4v4 — PCIe Gen5, 200 mm** The current connection is: `RTX 5090 → Lian Li VG4v4 Gen5 riser → ASUS Z890` GPU-Z has shown a normal PCIe x16 Gen5 link. However, I have **not yet tested the RTX 5090 directly in the motherboard's primary PCIe x16 slot**. That is my next major test. # What happens during the crash? The typical sequence is: 1. Gaming or a GPU workload is running normally. 2. The display suddenly goes completely black. 3. GPU fans immediately ramp to 100%. 4. Other system fans also ramp up. 5. The computer becomes completely unresponsive. 6. `Win + Ctrl + Shift + B` does nothing. 7. A hard shutdown is required. There is no normal recovery from the event. This has happened in multiple applications, including: * Gaming * ComfyUI / CUDA workloads * Ollama / local LLM inference Normal Windows desktop usage is extremely stable. # The strange part: sometimes the system will not POST after the crash This has happened **twice in a row during the most recent incidents**. After the black-screen crash and hard reset: * PC powers on * No BIOS screen * No display signal * Cannot enter BIOS * Normal reboot does not fix it The machine only became usable again after: **disconnecting AC power completely and leaving the PC without AC for approximately 20–30 minutes.** After reconnecting power, the system booted normally. Once it boots, everything appears normal again. Windows is stable, gaming works, and AI workloads can work normally. This temporary **failure to POST after a GPU hard-lock** is the part I find most concerning. # Frequency / timeline The system was stable for roughly the first **6 months**. The problem then appeared and gradually increased over approximately **5 months**. Initially: * rare crashes Now: * sometimes every 1–2 days * sometimes almost daily * dozens of hard resets over the lifetime of the problem This appears to be getting worse over time. # It happens under completely different GPU workloads The crashes are not limited to one game or one CUDA application. I have seen the same basic failure during: # Gaming For example, **007 First Light** has produced multiple hard black-screen crashes on separate days. # CUDA / ComfyUI There have been numerous crashes during AI image/video generation. # Ollama I have also seen GPU-related failures during local LLM inference. This makes me wonder whether the common factor is the RTX 5090 / PCIe / driver / power path rather than a particular application. # NVIDIA driver history I previously used: **610.88** I am currently on: **616.92** The problem is still present. I have **not yet performed a DDU cleanup and tested an older driver such as 596.49**. # Power / undervolt testing I have already experimented with: * GPU undervolting * 70–80% power limit * NVIDIA Power Management = Prefer maximum performance Interestingly, reducing the GPU power limit sometimes appeared to reduce the frequency of crashes. For example, I was able to run roughly 30 minutes of AI video generation at around **80% power limit** without crashing. However, the issue later returned. So lowering power consumption may affect the frequency, but it has **not eliminated the problem**. I have not yet completed the cleanest possible test with: * GPU stock * no undervolt * no power-limit reduction * GPU directly connected to motherboard # PCIe testing I have previously tested: * PCIe forced to Gen4 * ASPM / PCIe power-management-related settings disabled The riser is still installed. I therefore have not yet completely ruled out: * riser signal integrity * PCIe link stability * motherboard PCIe/root-complex interaction # Power cable / PSU The GPU uses the PSU's native 12V-2x6 / 16-pin cable. I have checked and reseated it. There is: * no visible melting * no discoloration * no burning smell * no obvious connector damage * no obvious cable damage The connector appears fully seated. The PSU is a **1200 W ATX 3.1 unit**, so this is not an obvious undersized-PSU situation. I have not yet tested with another PSU. # Event Viewer / crash logging Logging is inconsistent. Some crashes produce: `nvlddmkm` **Event ID 153** Other hard-locks produce almost no useful information for the exact crash. I have checked: * Event Viewer * NVIDIA-related events * Windows System logs * minidumps * LiveKernelReports For some crashes there is no useful dump at all. That makes me wonder whether some of these incidents are deeper GPU/PCIe failures where Windows never gets a chance to record a normal driver recovery. # Current theories I am trying to separate At the moment I am trying to distinguish between: * RTX 5090 hardware instability/failure * NVIDIA driver issue * GPU GSP / firmware issue * PCIe link/signal-integrity problem * Gen5 riser problem * motherboard PCIe/root-complex issue * PSU/power-delivery/transient issue # My next test My next major test will be: **Remove the Lian Li VG4v4 riser and install the RTX 5090 directly into the motherboard's primary PCIe x16 slot.** Then I plan to run: * PCIe Gen4 * XMP OFF * GPU completely stock * no undervolt * no power-limit reduction and reproduce the same workloads that have previously triggered the crashes. This should at least eliminate the riser as one variable. # Questions for RTX 5090 owners I would particularly appreciate input from anyone who has experienced something similar. # 1. Has anyone seen an RTX 5090 produce: **black screen + 100% fans + complete system lock** followed by a **temporary failure to POST after the crash**? # 2. Can a crashed RTX 5090 remain in some kind of hardware/GSP/firmware state where a normal reboot is insufficient, but **removing AC power completely** allows it to initialize again? # 3. Can a PCIe Gen5 riser cause this type of intermittent failure even when GPU-Z reports a perfectly normal x16 Gen5 link? # 4. How useful are NVIDIA **Xid / GSP / PCIe-related logs** for distinguishing GPU hardware failure from a driver or PCIe problem? Are there particular Xid codes, GSP logs, WHEA events, or other diagnostics I should be specifically looking for? # 5. At what point would you consider this behavior sufficient evidence to treat the RTX 5090 as a potential **RMA candidate**? The fact that the problem has become progressively more frequent, occurs across unrelated workloads, and has now twice been followed by a failure to POST until AC power was removed is what concerns me most. I can provide: * exact BIOS version * Event Viewer / `nvlddmkm` events * GPU-Z screenshots * HWiNFO sensor logs * WHEA events * LiveKernelReports * minidumps * NVIDIA driver details Any experience with **RTX 5090 + black screen + 100% fans + failed warm reboot / temporary no-POST** would be especially useful.
document fetched 2026-09-21 00:32 · canonical https://reddit.com/r/nvidia/comments/1wli3c2/rtx_5090_random_black_screen_100_fans_hard_lock