Get the latest crypto news, updates, and reports by subscribing to our free newsletter.
Giấy phép số 4978/GP-TTĐT do Sở Thông tin và Truyền thông Hà Nội cấp ngày 14 tháng 10 năm 2019 / Giấy phép SĐ, BS GP ICP số 2107/GP-TTĐT do Sở TTTT Hà Nội cấp ngày 13/7/2022.
© 2026 Index.vn
A newly reported attack chain, dubbed GPUBreach, extends the long-known Rowhammer technique to target GPU memory on NVIDIA’s Ampere-generation graphics processors. Researchers say the method can break the GPU’s page table, enabling a non-privileged CUDA process to read and write across the entire GPU memory, and then escalate access to the CPU without disabling IOMMU.
Rowhammer is a vulnerability class that has existed for more than a decade and was previously demonstrated against system memory such as DDR modules. In the GPUBreach variant, the target is the memory chips inside the GPU, rather than ordinary RAM.
According to the research described by Ars Technica, the attack uses electromagnetic interference to affect individual memory chips within the GPU. The GPU memory is then misled into treating sensitive data as if it were stored in a safe location, while the data is actually exposed.
The research indicates that NVIDIA Ampere-generation GPUs are in the affected area, including:
The GPUBreach attack chain is described as capable of corrupting the GPU page table. Once that occurs, a non-privileged CUDA process can obtain arbitrary read and write access to all GPU memory.
From there, the researchers say attackers can leverage newly discovered memory-safety bugs in the NVIDIA driver to escalate privileges to the CPU, ultimately enabling root access across the entire system.
A key point highlighted in the reporting is that GPUBreach achieves its escalation without disabling IOMMU. IOMMU is a protection that is often disabled in similar attacks, and the researchers characterize this as making GPUBreach a more serious threat than contemporaneous work.
For individual users, the practical risk is described as low. GPU-targeted Rowhammer attacks are described as technically extremely complex, and the reporting states that there are no real-world cases. It also notes that there is no evidence showing that any GPU line has been exploited using this method.
Premium gym chains are entering a “golden era” that is ending or already in decline, as rising operating costs collide with shifting consumer preferences toward more flexible, community-based ways to exercise. Long-term memberships are shrinking, margins are pressured by higher rents and facility expenses, and competition from smaller, more personalized…