I've started developing my own GPU monitoring system.
I'm conceiving the project as focused on the accuracy of GPU temperature and other information. The main goal is to implement hotspot display on Nvidia 50xx graphics cards, since the hotspot is disabled on them, and the driver doesn't provide this information. The solution in hwinfo is questionable, as it's unclear where this temperature is coming from. Core and video memory telemetry works on the cards I've tested, but I'm still working on independently verifying Hotspot readings before enabling them.
I try to design the software based on the principle that either telemetry is present or it isn't. That is, if the sensors are either inaccurate or provide questionable information, the program will return N/A.
The stack I chose to use for development is Native Win32 + Rust. The project itself is quite small (about 8 megabytes), but that's excluding further optimizations I plan to perform.
The telemetry provider itself is a separate component that is not a kernel module or driver.
The graphical interface is currently under development, and this is not its final form. :)
I plan to make it pleasant and minimalistic.
The software itself is currently only available for Windows, but I plan to create a Linux build in the future.
I've started developing my own GPU monitoring system.
I'm conceiving the project as focused on the accuracy of GPU temperature and other information. The main goal is to implement hotspot display on Nvidia 50xx graphics cards, since the hotspot is disabled on them, and the driver doesn't provide this information. The solution in hwinfo is questionable, as it's unclear where this temperature is coming from. Core and video memory telemetry works on the cards I've tested, but I'm still working on independently verifying Hotspot readings before enabling them.
I try to design the software based on the principle that either telemetry is present or it isn't. That is, if the sensors are either inaccurate or provide questionable information, the program will return N/A.
The stack I chose to use for development is Native Win32 + Rust. The project itself is quite small (about 8 megabytes), but that's excluding further optimizations I plan to perform.
The telemetry provider itself is a separate component that is not a kernel module or driver.
The graphical interface is currently under development, and this is not its final form. :) I plan to make it pleasant and minimalistic.
The software itself is currently only available for Windows, but I plan to create a Linux build in the future.
Repository and builds for Windows:
https://github.com/k1gs/Gpu-Shark
I would be interested in feedback from those who have worked with NVIDIA telemetry, especially on different motherboard variants.