GeForce RTX 5000: Modders Unlock Individual GDDR7 Chip Temperatures

After making the temperature information of NVIDIA Blackwell GPUs' hotspots accessible, some independent developers have managed to obtain data from the thermal sensors of GDDR7 memory as well. This result further expands the monitoring tool capabilities on the new GeForce RTX 5000, providing information that NVIDIA does not officially offer to users.

The work has been carried out by a group of enthusiasts and developers, including, among others, the team led by Paulo Gomes, the developer of MSI Afterburner and RivaTuner Statistics Server known as "Unwinder"; Overclock.net member "asder00"; Martin "Mumak" Malík, the author of HWiNFO, and "olealgoritme".

To obtain the data, it was necessary to analyze the memory logs and study the hardware configurations used by the Blackwell cards. One of the main difficulties was the presence of memory configurations where multiple chips share certain lines and detection information. Indeed, NVIDIA does not provide public documentation on these internal logs.

To allow for reading the new data, the Hotspot.dll plug-in used by MSI Afterburner has been updated. In addition to the six temperature channels of the Blackwell dies, the calculated hotspot value, the delta from the GPU temperature, and the memory junction temperature, the software can now also display values from the thermal sensors of the GDDR7 memory.

Initial tests show interesting results. The NVIDIA GeForce RTX 5090 features 16 GDDR7 modules of 2 GB each, totaling 32 GB of memory, but the system returns eight detection points, each associated with a pair of chips. Through HWiNFO64, maximum temperatures of around 64 °C were recorded during measurements performed on a Gigabyte model.

A NVIDIA GeForce RTX 5060 Ti 16 GB also provided additional data on memory: during the 3DMark Steel Nomad stress test, two values related to the GDDR7 chips were recorded, standing at 68 °C and 64 °C respectively.

While the new information does not allow for precise identification of the physical sensor locations on the card, it can be useful for pinpointing any anomalies in heat distribution. A module that reaches significantly higher temperatures than others could indicate issues related to heatsink pressure, suboptimal thermal pad contact, or deterioration of the material used for heat transfer.

At the moment, NVIDIA does not offer official access to these parameters through its public software interfaces and has not provided documentation on the logs used by the Blackwell GPU sensors. The company has also not responded to inquiries regarding a possible restoration of official access to hotspot data or the publication of technical specifications for the sensors. Consequently, the information obtained by the community is a sort of reverse engineering operation that, unless NVIDIA intervenes in some way, could be refined over time and made even more useful, both for those who love to monitor every parameter and in the event of comparing the effectiveness of individual cooling systems.