Here’s What Nvidia Has to Say About Rowhammer and If You Need to Be Concerned

Challenges in cybersecurity are a fundamental aspect of our digital age. Some threats are more concerning, whereas others are less critical than commonly believed. Today, we will explore Rowhammer, a recognized exploit impacting dynamic random-access memory (DRAM), specifically a variation called GPUHammer that focuses on graphics cards equipped with GDDR6 memory.

Recent findings from the University of Toronto indicate that GPUHammer is the first Rowhammer variant that specifically targets GPUs (graphics processing units). The research shows that GPUHammer can change bits on GDDR6 memory, potentially resulting in problems like data leaks and diminishing the GPU’s performance in select tasks. For example, the researchers have shown that modifying even a single bit can drop the accuracy of deep neural network models from 80 percent to 0.1 percent, with additional accuracy loss as more bits are changed. Considering the widespread use of Nvidia GPUs in data centers and workstation computers, this vulnerability could impact many businesses.

SEE ALSO:

Everything Unveiled at Nvidia’s 2025 Computex Press Conference in 19 Minutes

Nvidia has released a security advisory concerning the exploit. Fortunately, the remedy is uncomplicated. Activating System-Level ECC, or error correcting code, introduces redundancy in the bits, enabling the system to automatically rectify any flipped bits before problems arise.

Should I worry about GPUHammer?

In summary, no. Nvidia and the researchers agree that this mainly impacts workstation and data center GPUs. The researchers even tried this exploit on an Nvidia RTX 3080 without any success. Therefore, for everyday users, there is no reason for alarm. Furthermore, BleepingComputer highlights that the exploit necessitates conditions that are generally not present in consumer-grade computers, so even if consumer GPUs were vulnerable, the risk remains low.

If you operate a business that utilizes Nvidia workstation or data center GPUs, you should check the list of affected GPUs. If yours is listed, follow <a href="https://www.nvidia.com/content/Control-Panel-Help/vLatest/en-us/mergedProjects/3D%20Settings/To_turn_your_GPU_ECC_on_or_off.htm" target="_blank"