New Vulnerability in Nvidia GPUs Could Allow Hackers to Seize Full Control of Systems
A significant new vulnerability has been discovered in Nvidia’s high-end GPUs, with models like the RTX 3060 and RTX 6000 exposed to an attack method that allows hackers to gain root control over systems. This vulnerability arises from an exploit of the GDDR memory in these GPUs, which is usually difficult to manipulate.
Exploit Mechanism: GDDRHammer and GeForge
Researchers have identified two main attack techniques: GDDRHammer and GeForge. The flaw targets GPUs built on the Ampere architecture, allowing for the manipulation of critical data structures like the Page Tables in the system. The attack uses rapid electrical signal interference to cause bit-flipping in the memory, allowing hackers to bypass security systems and write to CPU memory directly.
The severity of the issue is amplified by the ability of the GDDRHammer attack to induce bit errors up to 129 times per memory cache, which is 64 times more than previous tests. This enables hackers to break through isolation boundaries between users, making systems highly vulnerable. The GeForge attack can cause data corruption up to 1,171 times on the RTX 3060, opening up a root shell to execute commands on the target system.
Impact and Mitigation:
The cause of the vulnerability lies in the increasing density of newer memory types, which makes it easier for electrical signal interference to occur. Furthermore, many systems disable IOMMU (Input-Output Memory Management Unit) by default to prioritize performance. This makes it easier for the GPU to access private areas of the CPU’s memory.
To mitigate the risk, Nvidia and researchers recommend enabling IOMMU in the BIOS settings and turning on Error Correcting Codes (ECC) on the GPU. Although these measures may reduce performance and available memory space, they are essential to prevent hackers from exploiting this vulnerability.
Conclusion
While no real-world attacks have been reported using this Rowhammer technique yet, this research serves as a wake-up call for both chip manufacturers and cloud service providers to improve their security standards. As hackers develop more user-friendly tools, expensive GPUs could become entry points for intruders, putting sensitive systems at risk.
New Vulnerability in Nvidia GPUs Could Allow Hackers to Seize Full Control of Systems
A significant new vulnerability has been discovered in Nvidia’s high-end GPUs, with models like the RTX 3060 and RTX 6000 exposed to an attack method that allows hackers to gain root control over systems. This vulnerability arises from an exploit of the GDDR memory in these GPUs, which is usually difficult to manipulate.
Exploit Mechanism: GDDRHammer and GeForge
Researchers have identified two main attack techniques: GDDRHammer and GeForge. The flaw targets GPUs built on the Ampere architecture, allowing for the manipulation of critical data structures like the Page Tables in the system. The attack uses rapid electrical signal interference to cause bit-flipping in the memory, allowing hackers to bypass security systems and write to CPU memory directly.
The severity of the issue is amplified by the ability of the GDDRHammer attack to induce bit errors up to 129 times per memory cache, which is 64 times more than previous tests. This enables hackers to break through isolation boundaries between users, making systems highly vulnerable. The GeForge attack can cause data corruption up to 1,171 times on the RTX 3060, opening up a root shell to execute commands on the target system.
Impact and Mitigation:
The cause of the vulnerability lies in the increasing density of newer memory types, which makes it easier for electrical signal interference to occur. Furthermore, many systems disable IOMMU (Input-Output Memory Management Unit) by default to prioritize performance. This makes it easier for the GPU to access private areas of the CPU’s memory.
To mitigate the risk, Nvidia and researchers recommend enabling IOMMU in the BIOS settings and turning on Error Correcting Codes (ECC) on the GPU. Although these measures may reduce performance and available memory space, they are essential to prevent hackers from exploiting this vulnerability.
Conclusion
While no real-world attacks have been reported using this Rowhammer technique yet, this research serves as a wake-up call for both chip manufacturers and cloud service providers to improve their security standards. As hackers develop more user-friendly tools, expensive GPUs could become entry points for intruders, putting sensitive systems at risk.
Source:
Arstechnica
