Yesterday, NVIDIA rushes to release a critical hot program to curb the consequences of a previously released driver who triggered alerts from the AI and gaming communities by causing the system to incorrectly report safe GPU temperatures, even as cooling demand quietly escalates towards potentially critical levels.
In NVIDIA’s official post in the Hotfix version, the issue is called “” despite being ranked third in the stated fix listGPU monitoring utility may stop reporting GPU temperature after PC wakes up’.
Shortly after the affected game ready driver 576.02, the fixed stable diffuser Redit’s thread is fixed with the title Read to save your GPU!becoming a resource for anecdotal questions about new drivers and updated user reports. From these and other reports around the network, a timetable for some urgent issues can be established.
The first REDDIT report for the error appears to have occurred on Friday afternoon UTC, on Zephyrusg14 Subreddit, User Freeze 81 quoted a post on the NVIDIA forum (Archive):
Users of the NVIDIA forum found problems after updating 576.02. Source: https://www.nvidia.com/en-us/geforce/forums/game-ready-drivers/13/563010/geforce-grece-grd-57602-feedback-weedback-thread-read-rreas-41625/3524072/
NVIDIA Forum users report that after installing driver updates, MSI Afterburner and in-game monitors (e.g. call of Duty (Usually accessing native system readings, just like the GPU panel of Task Manager does in Windows) Stop updating GPU temperature readings and freeze around 35-36°C.
Users say that restarting the monitoring software has no effect and that accurate readings can be restored only by restarting the complete system. Tools such as Hwinfo and Nvidia’s own monitoring applications continue to report temperature correctly. Users stressed that the problem occurs during normal use, not just after waking the system from sleep.
User feedback on various forums highlights the general disruption of normal fan curve behavior and changes in core thermal regulation, resulting in the graphics processing unit idle at unexpectedly high temperatures and shockingly overheating in situations that are often considered standard operating loads, as detailed in this review:
“I can say there is something wrong. The weather outside was about 55°F/12°C, but I was still alive in the room. My windows were open, but I felt like there was no difference. All fans ran at maximum speed, and at first, the temporary temperature was good, lasting for a while after 68°C to 72°C.
“At first, it seemed normal – until the next morning, when I realized those weren’t idle temperatures, fans were still [kicking].
“After fixing a few things recently, I did some AI overclocking, so I’m not sure if these values happen to be high. After installing Asus AI Suite 3, it happened once before – due to this setup, it doesn’t even work properly.
“Anyway, I’m going to go back to an old driver now.”
Second-best
The official version of the 576.02 driver update PDF provides some clues about changes that could lead to new problems. In Section 5.5, NVIDIA acknowledges that GPU temperature can be reported incorrectly on NVIDIA Optimus Prime The system, when the application is not running, specifically displays zero degrees.
The official 576.02 update note solves the temperature monitoring problem that appears to affect more systems than Optimus Prime systems. Source: https://us.download.nvidia.com/windows/576.02/576.02-win11-win11-release-notes.pdf
Release notes:
5.5 GPU temperature is incorrectly reported on Optimus system
5.5.1 Problem
On Optimus systems, temperature reporting tools such as Speccy or GPU-Z report that the NVIDIA GPU temperature is zero when not running.
5.5.2 Explanation
On Optimus systems, when the NVIDIA GPU is not used, put it in a low power state. This causes the temperature reporting tool to return an incorrect value. Waking up the GPU to query the temperature will result in meaningless measurements because the GPU temperature changes accordingly.
These tools report accurate temperatures only when the GPU is awake and running.
NVIDIA Optimus is a GPU switching technology that switches between integrated and discrete graphics according to application needs to automatically balance performance and power consumption, designed to save battery life and reduce power consumption. For tasks like gaming or HD video playback, Optimus activates discrete GPUs for performance; in lightweight activities such as web browsing, it will be restored to integrated (on-vehicle) graphics.
The update appears to have extended the behavior previously limited to Optimus Prime systems to limits, allowing the affected GPU to enter a low-power state when idle, even if not hosted on Optimus Prime systems, thus destroying temperature reports in third-party tools.
Risk adjustment
In most cases, it can be said that the VBIO of the graphics card may prevent permanent GPU damage. VBIO performs thermal and power limits at the firmware level independently of the driver.
Therefore, even if the driver causes incorrect fan behavior or false alarms of temperature, VBIO should still be dynamically resistant, improve fan activity, or turn off the GPU to prevent hardware failure.
This does not mean that the risk is trivial – the constant high temperatures can reduce performance or pressure on adjacent components over time; in addition, if there is no common understanding that updated drivers can cause problems (especially in systems where drivers are updated “silently”), problems of this nature can mislead a large proportion of affected users who may try to resolve non-existent problems and may even cause system damage by applying uncertain “fixes.”
The wrong behavior caused by update 576.02 is particularly shocking for those engaged in AI workflows, where high-performance hardware is often pushed to thermal limits for extended durations.
The problematic 576.02 driver was inspired by more complaints after its release in mid-April, although preliminary reports say it provides some beneficial performance improvements. Despite providing Hotfix, and the level of damage that 576.02 seems to cause, it is still available for download on NVIDIA’s website at the time of writing*.
afterglow
There are multiple types of damage and/or inconvenience reports regarding the impact of a fault update: User Frankie_t9000 reported that his GPU crashed at startup due to heat buildup under the fault update and was only stable after underestimation. He commentedIt seems it is not permanently damaged, but needs to be made a remodel as soon as possible (I’m coming Wednesday) suspecting that the old hot paste will age due to heat buildup, so I put a new paste pad.‘
Yesterday, another user in the same thread said: I’m using a custom fan curve witty MSI combustion chamber, which keeps showing that my GPU temperature is always at 27°C, so the fan is not on, which causes overheating issues. I think this is my problem, but after installing the previous drivers, all this solves well. Also, Temps Arent is displayed correctly in Taskmanager.
Though NVIDIA (as it states persistently in each hotfix release) often provides hotfixes for particular video-games or platforms, the risk of heat damage to or around a GPU is higher for AI practitioners than for videogamers, since intense machine learning processes such as training or sustained inference place a GPU under consistent long-term load – an event likely to be triggered only periodically in a game, which may ‘spike’ into high usage for a boss-battle or specially requested part of the map, but otherwise design it as a tradeoff between GPU development and system stability.
* Archive: https://archive.ph/ylvr1
First published on Tuesday, April 22, 2025
