Today's cybersecurity headlines are brought to you by ThreatPerspective


Ethical Hacking News

Nvidia A6000 GPUs Vulnerable to Rowhammer Attack: Potential Threats to AI Model Accuracy


Nvidia's A6000 GPUs have been found vulnerable to Rowhammer attacks, which could compromise the accuracy of AI models used in critical applications. The attack exploits memory bit flipping vulnerabilities in DRAM chips used in GPUs and can be executed on Nvidia A6000 GPUs with GDDR6 DRAM.

  • The Nvidia A6000 GPU is vulnerable to the Rowhammer attack, which can manipulate weights of deep neural networks in GPU memory.
  • The attack can degrade AI model accuracy by up to 80 percent despite a defense mechanism present in GDDR6 memory.
  • Nvidia has issued a security advisory and provided guidance on mitigating the vulnerability.
  • Enabling Error Correction Codes (ECC) using the nvidia-smi -e 1 command can help prevent the Rowhammer attack, but at a performance cost.



  • The Register has recently reported on a concerning vulnerability discovered in Nvidia's A6000 GPUs, which could have significant implications for the accuracy of AI models. The Rowhammer attack, first identified in 2014, takes advantage of memory bit flipping vulnerabilities in DRAM chips used in many modern devices, including GPUs.

    According to research published by Chris Shaopeng Lin, Joyce Qu, and Gururaj Saileshwar from the University of Toronto, the attack can be executed on Nvidia A6000 GPUs with GDDR6 DRAM using a technique known as Rowhammer-induced bit-flips. This exploit allows attackers to manipulate the weights of deep neural networks (DNNs) resident in GPU memory, leading to significant degradation in AI model accuracy.

    The researchers demonstrated that they were able to degrade DNN accuracy by up to 80 percent using their Rowhammer attack, despite the presence of a defense mechanism called Target Row Refresh in GDDR6 memory. This finding has serious implications for organizations running AI applications in cloud environments with other tenants, as they may be vulnerable to similar attacks.

    Nvidia has since issued a security advisory warning its customers about the potential threat and providing guidance on mitigating the vulnerability. According to Nvidia, enabling Error Correction Codes (ECC) using the nvidia-smi -e 1 command and then rebooting can help prevent the Rowhammer attack. However, this approach comes with a performance hit of approximately 10 percent and a reduction in memory capacity of about 6.25 percent.

    The discovery of this vulnerability highlights the ongoing cat-and-mouse game between attackers and defenders in the world of AI security. As AI models become increasingly complex and rely heavily on specialized hardware like GPUs, it is essential that organizations take proactive measures to protect their infrastructure against emerging threats like Rowhammer attacks.

    In light of this new information, we will be monitoring the situation closely and providing updates as more information becomes available. In the meantime, IT professionals and organizations running AI applications should consider taking steps to mitigate potential vulnerabilities in their systems.



    Related Information:
  • https://www.ethicalhackingnews.com/articles/Nvidia-A6000-GPUs-Vulnerable-to-Rowhammer-Attack-Potential-Threats-to-AI-Model-Accuracy-ehn.shtml

  • https://go.theregister.com/feed/www.theregister.com/2025/07/14/nvidia_a6000_gpu_gpuhammer/


  • Published: Mon Jul 14 16:03:35 2025 by llama3.2 3B Q4_K_M













    © Ethical Hacking News . All rights reserved.

    Privacy | Terms of Use | Contact Us