GPUHammer - First Rowhammer Attack Targeting NVIDIA GPUs

Cybersecurity researchers at the University of Toronto have achieved a breakthrough in hardware-level attacks by successfully demonstrating GPUHammer, the first Rowhammer attack specifically targeting discrete NVIDIA GPUs. The research, which focuses on the popular NVIDIA A6000 GPU with GDDR6 memory, represents a significant expansion of the decade-old Rowhammer vulnerability beyond traditional CPU memories. The attack required developing novel techniques, including reverse-engineering proprietary GDDR DRAM row mappings and creating GPU-specific memory access optimizations to amplify hammering intensity. Unlike CPUs where physical addresses are accessible, NVIDIA GPUs keep these mappings private, forcing researchers to develop new methods for identifying vulnerable memory locations1. The timing is particularly significant as cloud providers increasingly offer GPU time-sharing services, creating multi-tenant environments where malicious actors could potentially target other users’ AI models or sensitive data residing in GPU memory. This technology is automatically enabled on supported devices, including RTX 50 series consumer cards and the latest data center products, offering built-in resistance to Rowhammer attacks. The team noted that A100 GPUs with HBM2e memory and RTX 3080 devices showed no bit-flips in their testing, though this may be due to different threshold levels or enhanced mitigations rather than immunity. The team demonstrated that these bit-flips can cause devastating accuracy degradation in machine learning models, with drops of up to 80% observed across popular neural networks, including AlexNet, VGG16, ResNet50, DenseNet161, and InceptionV31. Their GPUHammer attack successfully induced 8-bit flips across 4 DRAM banks on the A6000 GPU, demonstrating that Graphics-DDR (GDDR) memories are indeed vulnerable to the same disturbance attacks that have plagued CPU memories for years. GPUs have approximately 4× higher memory latency compared to CPUs and faster refresh rates, making conventional hammering approaches ineffective. The company strongly recommends enabling System-Level ECC on professional and data center products, noting that it’s enabled by default on Hopper and Blackwell data center GPUs. Cyber Security News is a Dedicated News Platform For Cyber News, Cyber Attack News, Hacking News & Vulnerability Analysis. “This is the first systematic Rowhammer campaign on NVIDIA GPUs,” the researchers stated in their paper. The company emphasized that System-Level ECC effectively mitigates the attack when enabled, though this protection comes with approximately 6.5% memory overhead and 3-10% performance impact1. The research underscores the importance of hardware-level security considerations in AI system design and the need for robust mitigation strategies as GPU computing continues to expand across critical applications. Following responsible disclosure on January 15, 2025, NVIDIA issued a comprehensive security advisory acknowledging the vulnerability. While the researchers focused on the A6000 GPU, similar vulnerabilities may exist across other GPU architectures and memory types.

This Cyber News was published on cybersecuritynews.com. Publication date: Sat, 12 Jul 2025 17:20:10 +0000


Cyber News related to GPUHammer - First Rowhammer Attack Targeting NVIDIA GPUs

GPUHammer - First Rowhammer Attack Targeting NVIDIA GPUs - Cybersecurity researchers at the University of Toronto have achieved a breakthrough in hardware-level attacks by successfully demonstrating GPUHammer, the first Rowhammer attack specifically targeting discrete NVIDIA GPUs. The research, which focuses ...
2 days ago Cybersecuritynews.com Inception
NVIDIA issues guidance to defend GDDR6 GPUs against Rowhammer - NVIDIA is warning users to activate the System Level Error-Correcting Code  mitigation to protect against Rowhammer attacks on graphical processors with GDDR6 memory. Rowhammer represents a real security concern that could cause ...
3 days ago Bleepingcomputer.com
Nvidia To Build Network Of AI Chip Plants In Japan - Nvidia chief Jensen Huang says company to work with local companies to build network of AI chip plants in Japan. Nvidia is to collaborate with local companies to build a network of semiconductor manufacturing facilities in Japan to meet demand for ...
1 year ago Silicon.co.uk
Apple, AMD, Qualcomm, Imagination GPUs open to data theft The Register - A design flaw in GPU drivers made by Apple, Qualcomm, AMD, and likely Imagination can be exploited by miscreants on a shared system to snoop on fellow users. On a non-shared system, malware that manages to run on the box could abuse the weakness to ...
1 year ago Go.theregister.com Hunters
Nvidia sued after video call mistake showed 'stolen' data - According to a lawsuit filed against tech giant Nvidia, senior staff member Mohammad Moniruzzaman made this error with disastrous consequences. In the course of it, Valeo claims he accidentally displayed a file proving he stole its tech secrets. The ...
1 year ago Bbc.com
CVE-2021-42114 - Modern DRAM devices (PC-DDR4, LPDDR4X) are affected by a vulnerability in their internal Target Row Refresh (TRR) mitigation against Rowhammer attacks. Novel non-uniform Rowhammer access patterns, consisting of aggressors with different frequencies, ...
3 years ago
AMD, Apple, Qualcomm GPUs leak AI data in LeftoverLocals attacks - A new vulnerability dubbed 'LeftoverLocals' affecting graphics processing units from AMD, Apple, Qualcomm, and Imagination Technologies allows retrieving data from the local memory space. Tracked as CVE-2023-4969, the security issue enables data ...
1 year ago Bleepingcomputer.com CVE-2023-4969
Cohesity partners with NVIDIA to harness the power of generative AI - Cohesity announced a collaboration with NVIDIA to help organizations safely unlock the power of generative AI and data using the recently announced NVIDIA NIM microservices and by integrating NVIDIA AI Enterprise into the Cohesity Gaia platform. ...
1 year ago Helpnetsecurity.com
CVE-2020-10255 - Modern DRAM chips (DDR4 and LPDDR4 after 2015) are affected by a vulnerability in deployment of internal mitigations against RowHammer attacks known as Target Row Refresh (TRR), aka the TRRespass issue. To exploit this vulnerability, the attacker ...
5 years ago
Five AI topics to discuss with your CEO - At Cisco Live EMEA in Amsterdam in early February, you couldn't swing an Ethernet cable without hitting someone who was talking about it. Even though AI comes with especially useful applications and some very practical downsides, it's important to ...
1 year ago Feedpress.me
Palo Alto Networks Prevents Data Loss at Enterprise Scale with NVIDIA - With NVIDIA accelerated computing and AI software, cybersecurity leaders like Palo Alto Networks can safeguard vast amounts of sensitive information with unprecedented speed and accuracy, ushering in a new era of AI-driven data protection. The ...
9 months ago Paloaltonetworks.com
Attack Vector vs Attack Surface: The Subtle Difference - Cybersecurity discussions about "Attack vectors" and "Attack surfaces" sometimes use these two terms interchangeably. This article guides you through the distinctions between attack vectors and attack surfaces to help you better understand the two ...
2 years ago Trendmicro.com
Volt Typhoon Ramps Up Malicious Activity Against Critical Infrastructure - China-backed cyber espionage group Volt Typhoon is systematically targeting legacy Cisco devices in a sophisticated and stealthy campaign to grow its attack infrastructure. In many instances, the threat actor, known for targeting critical ...
1 year ago Darkreading.com Volt Typhoon
Attackers Could Eavesdrop on AI Conversations on GPUs - Researchers at cybersecurity research and consulting firm Trail of Bits have discovered a vulnerability that could allow attackers to read GPU local memory from affected Apple, Qualcomm, AMD and Imagination GPUs. In particular, the ...
1 year ago Techrepublic.com
9 Best DDoS Protection Service Providers for 2024 - eSecurity Planet content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More. One of the most powerful defenses an organization can employ against distributed ...
1 year ago Esecurityplanet.com
Building Data Center Infrastructure for the AI Revolution  - This is part two of a multi-part blog series on AI. Part one, Why 2024 is the Year of AI for Networking, discussed Cisco's AI networking vision and strategy. This blog will focus on evolving data center network infrastructure for supporting AI/ML ...
1 year ago Feedpress.me
Fortifying confidential computing in Microsoft Azure - I wrote about how Microsoft used Intel's secure extensions to its processor instruction sets to provide a foundation for confidential computing in Azure a few years ago. In the years since, the confidential computing market has taken a few steps ...
1 year ago Infoworld.com
Tech upgrades for first responders are a necessity, not a luxury - Public safety professionals want technology upgrades and adoption of federal standards for first responder IT security, reporting and efficiency, according to Mark43. First responders face growing concerns amidst cybersecurity surge. The ...
1 year ago Helpnetsecurity.com
Ten new Android banking trojans targeted 985 bank apps in 2023 - This year has seen the emergence of ten new Android banking malware families, which collectively target 985 bank and fintech/trading apps from financial institutes across 61 countries. Banking trojans are malware that targets people's online bank ...
1 year ago Bleepingcomputer.com
Multiple NVIDIA CUDA Toolkit Vulnerabilities Let Attackers Trigger DoS - Attackers could crash analysis tools or leak memory contents by distributing weaponized cubin files – a critical risk for AI development teams sharing pre-trained models. These vulnerabilities, spanning the cuobjdump and nvdisasm utilities, expose ...
4 months ago Cybersecuritynews.com
NVIDIA Riva Vulnerabilities Let Attackers Escalate Privileges - The update, detailed in a March 10, 2025 security bulletin, impacts all Linux deployments running Riva versions ≤2.18.0 and follows coordinated disclosure with  Trend Micro’s David Fiser and Alfredo Oliveira researchers. NVIDIA has ...
4 months ago Cybersecuritynews.com CVE-2025-23242
Vultr Cloud Inference simplifies AI deployment - Vultr launched Vultr Cloud Inference, a new serverless platform. Leveraging Vultr's global infrastructure spanning six continents and 32 locations, Vultr Cloud Inference provides customers with scalability, reduced latency, and enhanced cost ...
1 year ago Helpnetsecurity.com
Nvidia Targets Insider Threats with Digital Fingerprinting Technology - Nvidia recently announced a new technology to help detect and prevent insider threats. The tech, known as Digital Fingerprinting, is designed to detect unauthorized attempts to access sensitive data or systems within a company's network. The ...
2 years ago Csoonline.com
Nvidia Promises Japan AI Support Amidst Heavy Demand - Nvidia chief executive Jensen Huang says company will try to prioritise Japan AI requirements amidst heavy worldwide demand. Nvidia chief executive Jensen Huang said the company would do its best to prioritise Japan for artificial intelligence chips ...
1 year ago Silicon.co.uk
Trends: Hardware gets AI updates in 2024 - This includes the use of specialized neural engines in devices like the iPhone 15 Pro, which are optimized for AI tasks such as machine learning and natural language processing. This configuration allows for new experiences such as real-time AI image ...
9 months ago Securityintelligence.com

Latest Cyber News


Cyber Trends (last 7 days)