Sunday, July 12th 2026
NVIDIA GeForce RTX 50 Series Hotspot Sensor Can Be Accessed with NVIDIA's Internal Tool
It seems that NVIDIA's removal of the hotspot temperature sensor from the GeForce RTX 50 series "Blackwell" GPUs is not permanent, as the sensor data can still be accessed and the actual sensor is physically present. However, this is only possible with NVIDIA's internal tool called Modular Diagnostic Software (MODS), which isn't available to the public. According to Brazilian repair specialist Paulo Gomes, NVIDIA's MODS tool can still access the hotspot sensor. Readers may recall that back in January 2025, when the GeForce RTX 5090 GPU was reviewed, it was confirmed that NVIDIA had indeed removed the hotspot sensor reading API from "Blackwell." However, the company has kept access to "GPU Temperature" and "Memory Temperature" features for measuring the overall GPU temperature and its memory, making up for the absence of the hotspot sensor by keeping other features available.
For those who want to access the hotspot sensor on NVIDIA's GeForce RTX 50 series, Paulo Gomes shows that it is possible through the use of MODS, which is an internal tool available only to NVIDIA employees in testing labs. Some builds have been circulating among repair technicians. The tool requires a bootable Linux drive and likely specific packages, making it difficult for the average enthusiast to set up. By connecting the tool from a remote server, the repair crew discovered that NVIDIA's thermal limit of 107°C had been reached five times in some hotspots on the GPU die, despite the average GPU temperature of the tested GIGABYTE GeForce RTX 5070 Ti being around 68°C.
This indicates that the hotspot reading revealed the issue that led to the GPU being brought to the repair shop, while the average GPU temperature did not highlight these problems. Our RTX 5090 review also highlighted that the hotspot sensor was useful in situations like this, where a misaligned cooler or water block was causing trouble that couldn't be detected otherwise. It's important to note that hotspot readings are still available on older GeForce RTX 40 series "Ada Lovelace" graphics cards.
Source:
Paulo Gomes on YouTube
For those who want to access the hotspot sensor on NVIDIA's GeForce RTX 50 series, Paulo Gomes shows that it is possible through the use of MODS, which is an internal tool available only to NVIDIA employees in testing labs. Some builds have been circulating among repair technicians. The tool requires a bootable Linux drive and likely specific packages, making it difficult for the average enthusiast to set up. By connecting the tool from a remote server, the repair crew discovered that NVIDIA's thermal limit of 107°C had been reached five times in some hotspots on the GPU die, despite the average GPU temperature of the tested GIGABYTE GeForce RTX 5070 Ti being around 68°C.
35 Comments on NVIDIA GeForce RTX 50 Series Hotspot Sensor Can Be Accessed with NVIDIA's Internal Tool
If a company wants to void your warranty, it goes without saying that it should have to prove you did something that damaged the product.
You might as well just buy an Nvidia GPU and get it over with already.
The closet is open, you can come out now.
This made me think that someone needs to make a meme of one of these hot GPUs heating up a Utah teapot as seen in en.wikipedia.org/wiki/Utah_teapot because the Utah teapot is used as a test 3D model, making it a fitting 3D model for something related to a GPU that is supposed to render 3D images.
The high hotspot temperature makes me think about the meme of Fermi GPUs grilling food when they proved to be super hot as well.
Who would have thought that RTX 5000 series could be overheating when they
are same process as RTX 4000 with just scaled up transistors count and no IPC gains.
The smaller area in which heat is produced (concentrated), the worse heat menagement (cooling).
RTX 5090 has funny PCB. Yep, it's tiny and sexy, but for electronics it's a royal nightmare.
Average GPU temp reading is bullshit, that one should be completely removed instead of hotspot.
You should always take the highest temperature of a chip as baseline for cooling, not average one.
On my RX 9070 XT, when GPU temp is around 50°C, that means hotspot is +15...20°C more. And so is VRAM.
But piece of shit heat management will not start fans even in lowest pwm duty cycle, when they are inaudible
and it's enough rpms to drop hotspot and VRAM temps to ~60°C.
GPU manufacturers are fighting for quietest cards, not giving shit about hardware lifespan.
Why go passive and let temps reach 70-80°C when you can spin fans in inaudible speeds
and at the same time lower temps significantly (and extend lifespan and reduce possibility of paste pump-outs).
So Jensen hid the HotSpot readout to hide the fact that a GPU core at 67C is a fully cooked 107C instead.
Next week NVidia will press release that Hotspot readouts have a serious rounding Error lol
But we did get along just fine, for decades, with cooling behavior tuned to average temperature. Though I'm not a fan of Nvidia hiding the hot spot reading, I can see why they did it: more often than not, such readings are little more than an attractive nuisance, needlessly alarming people.
Also, FWIW, I'm not seeing crazy deltas between average/hotspot on my 9070xt under zero RPM mode. The delta increases under full load, but in that situation the fan's already at well over audible levels. If you want to address the 9070xt's high hotspot/memory temps, dialing down the power limit by 10-15% is the way to go. The perf difference will be negligible, but if you're worried about perf, you can also toss on a small (under-)voltage offset (say, -20 or -30 mV), to compensate.