{"id":167488,"date":"2026-04-04T17:11:55","date_gmt":"2026-04-04T15:11:56","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/ataki-gddrhammer-i-geforge-iskazhayushhie-pamyat-gpu-dlya-dostupa-ko-vsej-pamyati-cpu"},"modified":"2026-04-04T17:11:55","modified_gmt":"2026-04-04T15:11:56","slug":"ataki-gddrhammer-i-geforge-iskazhayushhie-pamyat-gpu-dlya-dostupa-ko-vsej-pamyati-cpu","status":"publish","type":"post","link":"https:\/\/prohoster.info\/en\/blog\/news\/ataki-gddrhammer-i-geforge-iskazhayushhie-pamyat-gpu-dlya-dostupa-ko-vsej-pamyati-cpu","title":{"rendered":"GDDRHammer and GeForge attacks distort GPU memory to access all CPU memory","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Researchers from several universities in the USA and Australia have developed two Rowhammer-class attacks \u2014 GDDRHammer and GeForge \u2014 that allow for bit flipping in GDDR memory chips while running an unprivileged CUDA kernel on NVIDIA GPUs. Unlike last year's GPUHammer method, these new attacks are not limited to affecting the data being processed in GPU memory; they enable full access to all main memory related to the CPU's address space. The researchers demonstrated exploits providing root access to the main system while executing an unprivileged CUDA kernel on the GPU. <\/p>\n

Both attacks achieve premature charge loss in specific memory cells of the video memory to which attackers do not have direct access, leading to changes in the bits stored in those cells. The ability to read and write to memory related to the CPU's address space is implemented by interfering with the operation of the GPU memory allocator (cudaMalloc) to break the isolation of GPU memory and map GPU virtual addresses to arbitrary physical memory addresses of the GPU or CPU. <\/p>\n

The intervention occurs through the distortion of bit values in the GPU memory, which stores the page tables responsible for translating virtual addresses into physical ones. The differences between the GDDRHammer and GeForge attacks lie in the fact that the GDDRHammer method modifies the last-level page table (PT \u2014 Last Level Page Table), while the GeForge method alters the last-level page directory (PD0 \u2014 Last Level Page Directory). <\/p>\n

Among other things, the specified address translation tables are used to organize direct access from the GPU to the CPU memory, so changing the address in the GPU page table to a physical address in the main RAM and setting the APERTURE flag, which enables CPU memory mapping mode, allows reading and writing data with access to main memory over the PCIe bus with IOMMU disabled (by default off).

\"GDDRHammer<\/center>
\"GDDRHammer<\/center> <\/p>\n

Successful attacks have been demonstrated on high-performance professional NVIDIA RTX A6000 graphics cards based on the Ampere microarchitecture (with new cards starting at $6850, and used ones at $4000) and consumer models like the NVIDIA RTX 3060 ($250-300). The technique developed to circumvent Rowhammer protection, combined with the use of parallel processing capabilities offered by the GPU, allowed for a 64-fold increase in the frequency of cell distortion compared to previous attacks. As a temporary measure to block attacks, the command 'nvidia-smi -e 1' can enable Error Correcting Codes (ECC) mode, but this comes with additional overhead and can potentially be circumvented using attack methods like ECCploit and ECC.fail.