lseurttyuu/GPU-T v0.2.0

GPU-T v0.2.0

on GitHub

The NVIDIA Update

The wait is over! GPU-T v0.2.0 is officially here, bringing full, comprehensive support for NVIDIA graphics cards.

This release introduces a massive architectural upgrade, bringing GPU-T's feature set for NVIDIA users perfectly in line with our AMD support. From hardware lookup and real-time sensors to deep-dive API telemetry, GPU-T is now a true multi-vendor diagnostic powerhouse.

✨ Highlights & Architecture

• Full NVIDIA Support: GPU-T now natively detects and monitors NVIDIA GPUs! To safely interact with proprietary NVIDIA libraries (libnvidia-ml.so / libnvidia-api.so), I've introduced the GPU-T.Nvapi sidecar. This secondary, Native AOT-compiled binary runs in the background to fetch advanced telemetry without compromising the stability or memory safety of the main UI.
• Advanced Tab Deep Dive (NVIDIA): The Advanced tab is now fully populated for team green. It features a completely new CUDA drop-down with detailed telemetry, Power & Limits readouts, and a hybrid diagnostic approach to accurately decode NVENC and NVDEC hardware encoding/decoding capabilities based on official NVIDIA headers.
• Async Sensor Telemetry: The entire sensor polling pipeline for NVIDIA (Hotspot, VRAM temps, Fan RPM, Bus Interface load) has been moved into background asynchronous tasks. This ensures the UI remains buttery smooth and completely unblocked, even when polling the driver at high refresh rates.

🚧 Known Limitation: NVIDIA Core/Memory/Boost Clocks

You may notice that the static GPU Core/Boost and Memory clocks displayed in the Graphics Card tab for consumer factory overclocked NVIDIA GPUs might not be 100% perfectly aligned with official spec sheets.

• The "Why": NVIDIA does not expose this specific static hardware data through standard user-space APIs. Accurate detection requires dumping and decoding the GPU BIOS directly, which requires root privileges.
• The Trade-off: A core philosophy of GPU-T is that it runs cleanly in user-space without requiring sudo or root access. For now, I have chosen to bypass BIOS dumping to keep the app as easy to run as possible. This is a conscious trade-off, but dynamic real-time clocks in the Sensors tab are fully functional!

🎨 UI / UX Improvements

• Smart Dependency Checks: The startup missing utilities popup has been rewritten. It is now "Vendor Aware" and will only alert you about missing tools relevant to the hardware actually installed in your PC (e.g., asking for nvidia-smi only if you have an NVIDIA card).
• Vendor-Specific Terminology: The UI now correctly adapts to the GPU vendor. For example, the Graphics Card tab will accurately display "SM Count" for NVIDIA cards instead of AMD's "Compute Units".
• Dynamic PCIe Status: Added real-time dynamic PCIe link status and width readouts to the Graphics Card tab for both NVIDIA and AMD GPUs.

🗄️ Database Updates

• Complete NVIDIA Database: Added a massive, comprehensive NVIDIA GPU database spanning from 2010 to 2026.
• Smart Model Detection: Implemented a highly accurate NVIDIA model detection algorithm using devID + subsysID matching.
• Max-Q Detection: Added specific database matching logic to correctly identify "Max-Q" variants for RTX 3000-series mobile GPUs and newer.

Installation: Download the .AppImage, make it executable (chmod +x), and run! For detailed instructions or alternative installation methods, check the README.