Two newly discovered Rowhammer-style attacks—GDDRHammer and GeForge—can give attackers full control over machines equipped with Nvidia GPUs, according to research published in April 2026. The attacks exploit vulnerabilities in GDDR5 and GDDR6 memory, bypassing both software and hardware defenses and exposing a massive new attack surface across consumer and enterprise devices.

Why does this matter? Nvidia GPUs are everywhere: powering AI workloads, gaming rigs, and cloud infrastructure. Until now, Rowhammer-style exploits were mostly a CPU-side problem. This research overturns that assumption, showing that GPU-attached memory is just as vulnerable—and that the industry’s current mitigations are inadequate for this new threat vector.

How GDDRHammer and GeForge Work

Rowhammer attacks, first documented in 2014, flip bits in DRAM by rapidly accessing (or "hammering") adjacent memory rows. Traditionally, these attacks targeted CPU-attached DRAM. GDDRHammer and GeForge apply the same principle to Nvidia GPU memory, specifically GDDR5 and GDDR6 chips, which are widely used in desktops, laptops, and data center servers.

The new attacks exploit the physical properties of these memory chips, inducing bit flips that can be chained to escalate privileges or execute arbitrary code. According to the researchers, the attacks can bypass both software-level protections and hardware isolation mechanisms, effectively breaking down the security boundaries that many systems rely on.

Scope: Consumer, Enterprise, and Cloud

The vulnerabilities affect a broad swath of Nvidia hardware, from gaming PCs to data center GPUs used in AI and cloud deployments. There are currently no comprehensive software or firmware mitigations for these GPU-based Rowhammer attacks. Existing defenses—such as memory scrubbing or error-correcting code (ECC) memory—are either ineffective or unavailable for the targeted GPU memory architectures.

This means every Nvidia-powered system with GDDR5 or GDDR6 memory is potentially exposed, regardless of whether it’s running in a home office or a hyperscale data center.

Industry Context: Hardware Security Under Fire

This research arrives at a time when hardware-level vulnerabilities are under increasing scrutiny. Rowhammer attacks have already forced chipmakers and cloud providers to rethink DRAM security. Now, with GPUs at the heart of AI, gaming, and scientific computing, the attack surface is broader—and the stakes are higher—than ever.

Nvidia has not yet released a public mitigation strategy for these specific vulnerabilities. The lack of immediate fixes leaves a significant window of exposure for organizations relying on Nvidia hardware for sensitive workloads.

“The attacks demonstrate that GPU memory is not immune to Rowhammer-style exploitation, and that current industry practices are insufficient,” the researchers wrote in their April 2026 paper.

For context: Rowhammer was first reported in 2014, and since then, mitigation has been a slow, iterative process—often lagging behind new attack techniques. The extension of Rowhammer to GPU-attached memory is a wake-up call for the entire hardware ecosystem.

What This Means

For founders building in this space: Hardware security can no longer be an afterthought. If you’re building AI infrastructure, cloud platforms, or even consumer-facing applications that rely on GPU acceleration, you must assume that hardware vulnerabilities are part of your threat model. The lack of available mitigations means startups have an opportunity—but also a responsibility—to innovate on detection, isolation, and rapid response strategies at the hardware-software boundary.

For the industry trajectory: This is a shot across the bow for GPU vendors and cloud providers. The assumption that GPUs are "safe by default" is dead. Expect a new wave of hardware security research, increased scrutiny on GPU memory architectures, and—inevitably—more pressure on Nvidia and its competitors to deliver real, hardware-level mitigations. The days of treating GPU memory as a black box are over.

The non-obvious second-order effect: As GPU vulnerabilities become public, regulators and enterprise buyers will demand more transparency and security guarantees from hardware vendors. This could slow down AI and cloud adoption in regulated industries, or force a shift toward alternative architectures—potentially opening the door for new entrants who can prove their security story. The trust gap in hardware is widening, and the winners will be those who close it fastest.