- Researchers at Zhejiang University have warned that GPU workloads could destabilize local networks and cause power outages.
- Attackers could leverage around 1,000 GPUs to drain power and generate excess heat in systems.
- Theoretical attack called Bit2Watt; mitigation measures include detection of malicious patterns and energy buffering systems
Anytime an AI data center thinks really hard, it can increase its power consumption to the point of triggering disruptions, or even power outages and equipment malfunctions. So, is it possible for a malicious actor to deliberately trigger this scenario, in order to cause physical harm?
Several researchers from Zhejiang University in Hangzhou, China, wrote a research paper titled “Bit2Watt: A Cyber-Physical Vulnerability Exploiting GPU Workloads on Power and IT Infrastructure.”
In it, they claim that a malicious cloud tenant is, in theory, capable of launching GPU workloads so intensive that they cause physical damage.
Suggest mitigation measures
“Our results indicate that GPU loads can achieve modulation frequencies above 6,000 Hz, compared to only a few hertz observed in conventional household loads such as air conditioners,” the team wrote in their research paper.
“Such high-frequency modulations can significantly induce voltage excursions, harmonic distortion and damping degradation.”
An attacker could use around 1,000 GPUs to target a one-megawatt local power grid composed primarily of distributed power sources (such as solar panels), causing it to lose nearly half of the electrical power, while generating around 20% more heat than usual.
“This not only threatens the availability of IT equipment, but also produces a negative damping ratio of -0.27, introducing an unstable mode into the system,” the paper adds.
“Once protections are triggered and IT loads are removed, this can trigger cascading outages, potentially leading to power outages exceeding 80% in large-scale power systems.”
AI data centers creating huge swings in power consumption is nothing new, and it’s a challenge that some of today’s brightest minds are trying to solve.
Fortunately, the attack is (still) purely theoretical and the researchers published the paper to warn against possible misuse. They also suggested mitigation measures: Defenders could look for malicious computer patterns, while operators should create energy storage systems for unusual peaks in demand.
Via The register
Follow TechRadar on Google News And add us as your favorite source to get our news, reviews and expert opinions in your feeds.