Loading story…
University of Toronto researchers described GPUThor, a Rowhammer-class attack targeting GDDR6 memory in NVIDIA Ampere GPU accelerators. The technique characterizes the platform’s Targeted Refresh Rate (TRR) behavior and uses non-uniform memory-access patterns to induce substantially more bit flips than earlier GPU Rowhammer methods, extending the DRAM disturbance-error threat first established by Rowhammer research.
GPUThor generated double- and triple-bit memory corruptions, which can exceed ECC’s single-bit correction capability. Researchers demonstrated a denial-of-service outcome: a GPU reset accompanied by data loss and a hardware-replacement warning. They did not demonstrate arbitrary code execution or compromise newer GPU generations; ECC bypass leading to code execution remains theoretical.
See affected versions and whether adversaries are exploiting it.
2 events from the most recent confirmed update back to the earliest known activity.
Three studies published in spring 2026 demonstrated Rowhammer attacks against NVIDIA GPUs using GDDR6 memory. Their results were more limited on the NVIDIA A6000 than on consumer GPUs, and the attacks did not work with ECC enabled.
University of Toronto researchers described GPUThor, a Rowhammer-class technique for GDDR6 memory in NVIDIA Ampere accelerators. By exploiting Ampere TRR behavior with non-uniform access patterns, they reported double- and triple-bit corruptions and demonstrated denial of service causing a GPU reset, data loss, and a hardware-replacement warning; arbitrary code execution was not demonstrated.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
See whether adversaries are exploiting this yet, and where the affected versions run in your environment.
5 references tracked. Mallory keeps watching after this page renders.
kaspersky.com
Open sourcekaspersky.ru
Open sourcegputhor.com
Open sourcegpubreach.ca
Open sourceusers.ece.cmu.edu
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
University of Toronto researchers disclosed **GPUThor**, a GPU-based Rowhammer technique that can bypass practical protections provided by SECDED ECC on certain NVIDIA Ampere workstation GPUs using GDDR6 memory, including the RTX A6000. By using a non-uniform memory-hammering pattern that accounts for undocumented request coalescing and Target Row Refresh behavior, the technique substantially increases bit-flip rates and can trigger uncorrectable double-bit errors or incorrect repair of triple-bit errors. An attacker needs to execute an unprivileged CUDA kernel, creating risk in shared GPU-tenancy environments and from untrusted local GPU workloads. The researchers demonstrated data corruption and GPU denial of service and described a path to host root privilege escalation through GPU page-table corruption. NVIDIA recommends enabling System-Level ECC (SYS-ECC), enforcing IOMMU/DMA isolation, monitoring GPU ECC telemetry, and restricting untrusted or cross-tenant GPU workloads; no CVE, patch, or confirmed in-the-wild exploitation was reported.
Aug 28, 2026
Researchers disclosed two GPU-based Rowhammer techniques, **`GDDRHammer`** and **`GeForge`**, that can induce bit flips in GDDR6 VRAM on certain Nvidia GPUs and break memory isolation. Tests showed the attacks against Ampere-era cards including the **RTX 3060** and **RTX A6000/RTX 6000**, where corrupted GPU page tables or page directories allowed attackers to gain read/write access to protected VRAM regions and, in some cases, remap access into host memory. The researchers said this could escalate from GPU compromise to **full control of the machine**, including root-level access to system memory. The reports said the attacks require an adversary to already run code on the target system, making shared GPU environments such as AI clusters a more plausible risk than a purely remote attack path. No in-the-wild exploitation has been reported, but the findings highlight a hardware security gap in systems where GPUs can directly access host memory. Recommended mitigations include enabling **IOMMU** in BIOS to restrict device DMA access to sensitive system memory and turning on **ECC** on supported GPUs, though both measures can impose performance or memory-capacity tradeoffs and ECC may not block every Rowhammer variant.
Jun 29, 2026
Researchers reported that **Rowhammer** remains a practical hardware security threat, with the newer **Phoenix** technique showing that **DDR5** memory can still be manipulated despite built-in protections. The attack revives a flaw first highlighted by Google Project Zero, which demonstrated that repeatedly accessing DRAM rows could trigger adjacent bit flips and be weaponized for **privilege escalation** on x86 systems, including a Linux kernel exploit that altered page table entries and a sandbox escape from **Google Native Client**. The newer findings indicate that DDR5 mitigations have not eliminated the underlying risk, extending concerns that began with widespread **DDR3** exposure across tested laptop models. Earlier research had already shown Rowhammer was more than a reliability issue, and the latest reports suggest modern memory modules remain susceptible to carefully engineered disturbance attacks that can undermine isolation boundaries and system integrity, renewing pressure on hardware vendors to disclose affected products and strengthen defenses.
Jun 29, 2026