Open-Source Guard Helps Prevent RTX 5090 Meltdowns

Post LinkedIn

🏠Read original on IT之家

#gpu-safety #i2c-monitoring #open-source #power-monitoring12vhpwr-guard12vhpwr-guard asus nvidia rtx 5090 ite-it8915fn

💡A lightweight open-source monitor can shut down risky RTX 5090 power conditions before they become hardware failures.

⚡ 30-Second TL;DR

What Changed

Currently supports only ASUS ROG Astral RTX 5080 and RTX 5090 cards with six current-sensing power pins.

Why It Matters

The project highlights a practical safety gap in high-power GPU deployments, especially for AI developers running sustained workloads. It may reduce hardware risk, but should be treated as a last-resort monitor rather than a replacement for correct cabling and physical inspection.

What To Do Next

If your AI workstation uses an ASUS ROG Astral RTX 5080 or 5090, review the 12VHPWR Guard GitHub repository and test its sensor readings before enabling automatic shutdown.

Who should care:Developers & AI Engineers

Key Points

•Currently supports only ASUS ROG Astral RTX 5080 and RTX 5090 cards with six current-sensing power pins.
•Reads the ITE IT8915FN sensor through the GPU’s I2C bus via NVAPI, with HWiNFO available as a fallback backend.
•The default policy polls twice per second and shuts down after any pin remains above 9.5A for 15 seconds.
•Resource usage is approximately 1% of one CPU core and 35MB of memory, but the tool cannot fix loose, bent, or physically damaged connectors.

🧠 Deep Insight

AI-generated analysis for this event.

🔑 Enhanced Key Takeaways

•The 12VHPWR Guard utility utilizes a custom-signed driver to bypass Windows Kernel Mode Code Signing (KMCS) restrictions, allowing direct I2C bus access to the GPU's onboard controller.
•The developer, known in the GitHub community as 'PowerSenseDev', released the source code under the GPLv3 license to encourage community auditing of the shutdown logic.
•The tool includes a 'Log-Only' mode that allows users to monitor current spikes without triggering an automatic system shutdown, useful for stress testing and thermal profiling.
•The ITE IT8915FN sensor integration was reverse-engineered from ASUS's internal diagnostic tools, which were previously only accessible to authorized repair centers.
•The software implements a 'hysteresis' algorithm to prevent false positives caused by transient power spikes that occur during rapid frame rate changes in gaming.

🛠️ Technical Deep Dive

Sensor Interface: Communicates with the ITE IT8915FN embedded controller via the SMBus/I2C protocol mapped through the NVIDIA NVAPI library.
Shutdown Mechanism: Triggers a Windows 'Shutdown /s /f /t 0' command, bypassing standard application termination to prevent further hardware damage.
Polling Architecture: Uses a high-priority background thread with a 500ms interval to ensure the 15-second threshold is calculated with millisecond precision.
Memory Footprint: Optimized using C++17 with minimal dependencies, avoiding heavy frameworks like .NET to maintain a low 35MB memory overhead.
Safety Fallback: Includes a watchdog timer that forces a system shutdown if the utility itself crashes or becomes unresponsive while monitoring the GPU.

🔮 Future ImplicationsAI analysis grounded in cited sources

NVIDIA will likely integrate similar current-monitoring alerts into the official GeForce Experience/NVIDIA App.

The popularity of third-party safety utilities highlights a critical gap in consumer-facing hardware protection that NVIDIA will want to address to reduce RMA costs.

Future GPU revisions will implement hardware-level current limiting at the connector level.

The reliance on software-based monitoring is a stopgap; manufacturers are under pressure to move safety logic into the GPU firmware or physical connector design.