Rowhammer attack targets NVIDIA GPUs with GDDR6

New research shows Rowhammer exploits can target NVIDIA GPUs using GDDR6 memory and extend beyond the graphics subsystem into host CPU memory. The attacks can corrupt GPU page tables and lead to full system compromise.

Rowhammer attacks are no longer limited to CPUs and DDR memory. New research indicates that NVIDIA GPUs using GDDR6 are also vulnerable, and the impact reaches beyond graphics hardware into the host system. Two independent teams, GDDRHammer and GeForge, developed working exploits that use Rowhammer-induced bit-flips in NVIDIA GPUs to gain complete control over the host CPU’s memory.

The attack applies to some NVIDIA GPU models spanning the Ampere to Ada Lovelace families. An attacker who succeeds can read and write anything stored in the machine’s main memory. Both teams introduced Rowhammer techniques tailored to GPU architecture, and these methods achieved a significantly higher rate of bit-flips on GDDR6 memory than previous approaches.

The key stage in both exploit chains targets the GPU memory allocator. Controlled bit-flips are used to corrupt the GPU’s page tables, and once those page tables are compromised, the attacker gains arbitrary read and write access to CPU memory. That collapses the security boundary between the graphics subsystem and the rest of the machine, enabling a full system compromise and root access without interacting with privileged software paths.

The affected GPUs include the GeForce RTX 3060, which experienced 1,171 bit-flips, and the RTX 6000 ‘Ada’ GPU, which saw 202 bit-flips from the attack. The findings show that the fragile nature of GDDR6 memory can be exploited in ways that directly compromise the CPU host, turning a long-known DRAM weakness into a broader platform security problem.

82

Impact Score

Artificial Intelligence agent struggles to run cafe in Sweden

Andon Labs handed a Google Gemini-powered Artificial Intelligence agent control of launching and operating a cafe in Stockholm. The experiment produced permit problems, strange purchases, and fresh questions about oversight and accountability.

Policymakers weigh pause on Artificial Intelligence data center construction

Federal, state, and local officials are moving to slow or condition large data center development as concerns grow over electricity costs, grid strain, environmental effects, and labor standards. Proposed moratoriums and tax incentive changes are creating new uncertainty for developers, hyperscalers, and financiers.

European Union delays key Artificial Intelligence Act obligations

European Union lawmakers have agreed to revise the Artificial Intelligence Act, delaying major high-risk compliance obligations and easing some overlapping requirements. The changes give businesses more time to prepare while preserving the law’s core framework for high-risk systems and transparency rules.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.