News Posts matching #Rowhammer

Return to Keyword Browsing

"GPUBreach" is a Rowhammer Attack for GDDR6-Based NVIDIA GPUs That Bypasses IOMMU

Late last week, we reported on a new series of rowhammer bit-flip attacks targeting GDDR6-based NVIDIA GPUs. Most of these attacks can be mitigated by enabling IOMMU through the BIOS, which restricts the memory regions the GPU can access on the host system, thereby closing the primary attack path. However, researchers from the University of Toronto have introduced "GPUBreach," which can bypass IOMMU and enable CPU-side privilege escalation, unlike the previous "GDDRHammer" and "GeForge" attacks. In most typical server, workstation, and even PC configurations, IOMMU restricts the GPU's access to the CPU's physical addresses, preventing direct memory access. These are the typical DMA-based attacks that the Input-Output Memory Management Unit protects users from. However, the new "GPUBreach" operates differently.

For example, "GPUBreach" exploits memory-safe bugs in the actual GPU driver and corrupts them. When IOMMU confines the GPU's direct memory access to driver-assigned buffers, the new exploit corrupts metadata within these permitted buffers. This causes the driver, which has kernel privileges enabled on the CPU host, to perform out-of-band writes to the buffer, effectively bypassing any protection IOMMU can offer. This logic is built into the kernel by default, as the GPU driver is one of the most trusted components of the operating system. Hence, IOMMU bypass is possible when the metadata is corrupted. Since "GPUBreach" grants an attacker full root privilege escalation, the attack differs significantly from previous rowhammer attacks.

New Rowhammer Attack Puts GDDR6-Powered NVIDIA GPUs at Risk

Rowhammer attacks exploit vulnerabilities in DRAM hardware by causing targeted bit-flips, allowing attackers to bypass memory isolation and gain control over a device. These attacks were initially focused on CPUs and their associated DDR memory, such as DDR4. However, recent research indicates that NVIDIA GPUs are also vulnerable due to the fragile nature of the GDDR6 memory they use, which directly compromises the CPU host. Two independent research teams have discovered ways to exploit this decades-old memory vulnerability against modern graphics hardware, with trouble extending beyond the GPU itself. The "GDDRHammer" and "GeForge" groups have each developed functional exploits that use Rowhammer bit-flips in NVIDIA GPUs with GDDR6 memory to gain complete control over the host CPU's memory. This attack can perform bit-flips on some NVIDIA GPU models, ranging from the "Ampere" to "Ada Lovelace" families of cards.

An attacker who succeeds in this can read and write anything stored in the machine's main memory. Both teams have also introduced new Rowhammer techniques specifically designed for GPU architecture, achieving a significantly higher rate of bit-flips on GDDR6 memory than previous methods. The critical step in both exploit chains involves targeting the GPU's memory allocator, using controlled bit-flips to corrupt the GPU's page tables. Once these page tables are compromised, the attacker gains arbitrary read and write access to CPU memory, breaking down the security boundary between the graphics subsystem and the rest of the machine. The end result is a full system compromise, as the attacker can manipulate memory at will and gain root access, achieving total control 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.

SK Hynix DDR5 DIMMs Vulnerable to "Phoenix" Rowhammer Attack, On-Die ECC Exposed Too

Researchers from ETH Zurich and Google have shown that SK Hynix DDR5 modules remain vulnerable to a new Rowhammer variant they call Phoenix (CVE-2025-6202), even with on-die ECC memory modules. The attack can be executed in only 109 seconds, making it a very realistic threat. By reverse engineering the built-in DRAM mitigation (TRR) for the already existing Rowhammer exploit, the team found a blind spot in the refresh-sampling logic: the mitigation's sampling period repeats every 128 tREFI intervals and, within those intervals, the first two of four sub-intervals are "lightly sampled." Leveraging this, the researchers crafted two novel long-running hammer patterns: a shorter 128-tREFI pattern and a much longer 2608-tREFI pattern. They later coupled them with a new self-correcting refresh synchronization method that realigns the attack whenever a refresh is missed. This synchronization is what lets Phoenix maintain alignment across thousands of refresh intervals, something prior routines like Zenhammer could not do reliably.

On 15 SK Hynix DDR5 modules, manufactured between late 2021 and late 2024, every module flipped bits under one of the two patterns. The 128-tREFI pattern was about 2.62× more effective on average. The team observed thousands of bit flips per DIMM and turned those flips into real-world primitives: page-table corruption, yielding arbitrary read/write access, extraction of valuable RSA-2048 keys from co-located VMs in many cases, and local privilege escalation by corrupting the sudo binary on some DIMMs. To increase the probability of hitting a vulnerable refresh offset (only 2 of 128 offsets are vulnerable, which is 1.56%), they run four shifted pattern instances across each of the four banks in parallel, boosting the hit probability by 16× to roughly 25%.
Return to Keyword Browsing
Nov 18th, 2026 07:09 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts