Nuked 
Hello followers! Today, we’re diving into some fascinating tech news—Nvidia’s GPUs are now victims of a rare type of attack called Rowhammer, which can flip bits in their memory.
Originally used for rendering graphics, GPUs like Nvidia’s RTX-A6000 are now powerful tools for machine learning and high-performance computing. Recently, researchers showed that these GPUs can be targeted with a Rowhammer attack, causing bit flips in their onboard memory.
This attack, dubbed GPUhammer, is significant because it’s the first time a successful Rowhammer exploit has been demonstrated on a discrete GPU. It can alter neural network models used in self-driving cars, medical diagnosis, and security systems by flipping a single bit, which drastically reduces the model’s accuracy.
To counteract this threat, Nvidia suggests enabling error-correcting code (ECC), which can detect and fix some errors. However, using ECC can slow down performance by up to 10%, especially in high-memory workloads like medical imaging.
The vulnerability primarily affects GDDR6 memory modules used in many Nvidia GPUs. While newer models with on-die ECC might be more secure, haven’t been thoroughly tested against Rowhammer yet. This new attack expands the scope of Rowhammer from CPU memory chips to GPUs, raising concerns about security in cloud environments, where resources are shared among users.
Performing such attacks is complex due to the physical design of GPU memory, but the proof of concept shows that hardware vulnerabilities remain a critical concern. The researchers, including Gururaj Saileshwar, plan to present their findings at an upcoming security conference, highlighting the importance of ongoing security measures in hardware design.
So, stay vigilant and keep an eye on how hardware manufacturers respond to these emerging threats. Tech keeps evolving, and so must our defenses!
