Slashdot: NVIDIA Warns Its High-End GPUs May Be Vulnerable to Rowhammer Attacks

Jul 12, 2025

—

Source URL: https://hardware.slashdot.org/story/25/07/12/199238/nvidia-warns-its-high-end-gpus-may-be-vulnerable-to-rowhammer-attacks
Source: Slashdot
Title: NVIDIA Warns Its High-End GPUs May Be Vulnerable to Rowhammer Attacks

Feedly Summary:

AI Summary and Description: Yes

Summary: The text discusses a new security notice from NVIDIA regarding vulnerabilities in GDDR6 memory on high-end GPUs due to Rowhammer attacks, showcasing the critical need for enabling Error Correction Code (ECC) to protect against these exploits. This finding is significant for data center administrators and users of high-performance workstations.

Detailed Description:
The report highlights a crucial security issue found in NVIDIA’s GPUs, particularly concerning GDDR6 memory. Researchers have discovered that Rowhammer attacks, traditionally a threat to CPU memory, can also target GPUs when ECC is disabled. This situation prompts a serious evaluation of memory protection protocols within the infrastructure of data centers and advanced computing systems.

Key points include:

– **Rowhammer Attacks**: A well-known vulnerability that allows malicious actors to alter data in memory through repeated access to specific locations, leading to potential data corruption.
– **Vulnerability in GDDR6**: The research indicates that this vulnerability affects GDDR6 memory in NVIDIA GPUs, environments not previously highlighted as susceptible.
– **Importance of ECC**: ECC (Error Correction Code) is crucial for safeguarding memory integrity. The research found that enabling ECC on the A6000 GPU effectively prevents Rowhammer attacks.
– **NVIDIA’s Guidance**: NVIDIA has released a reminder emphasizing the importance of using ECC where supported, stating that certain GPU lines and systems (e.g., Blackwell, Hopper, Ada, Ampere) incorporate this protection.
– **Built-in Protections**: Newer memory types like GDDR7 and HBM3 come with On-Die ECC, providing a level of automatic protection that cannot be disabled.
– **Performance Considerations**: Users may disable ECC due to perceived performance impacts or lack of clarity in configurations, adding a layer of risk.

NVIDIA encourages system administrators and enterprise users to verify their ECC settings utilizing tools such as nvidia-smi or management APIs, reflecting a proactive approach to security within cloud and data center environments. The report underscores the discipline required in managing hardware configurations to maintain security and performance integrity in high-stakes environments.

1 2 3 5 7 a access Act administrators advanced advanced computing AGI AI alt Ampere and API APIs app Arch art as at ated attack attacks Auto Bi black Blackwell built C centers CERN CI CIA clarity in Cloud co code Col Computing computing systems Configuration configurations core CPU critical D data data center data centers data corruption de Die ECC DoT e effective end enterprise enterprise use enterprise users environment ERP error error correction Error Correction Code (ECC) evaluation event exp exploit exploits for g GPU GPUs gs guidance H hardware hardware co hardware configuration hardware configurations HBM3 high high-performance Highlight Hopper HR http HTTPS in infrastructure integrity io Iron issue k Key l leading led level Li low M malicious actors man management memory memory protection mini N new NIST no Nvidia NVIDIA GPUs o of on ory over per performance performance considerations performance impact Performance Workstations point potential pre pro proactive prompt prompts protection protocol protocols ps Q R rag rate RCE red release report research researchers Risk Ro RoT row Rowhammer Attacks s Sable safe search sec security settings side Sig source specific SSE Station support system system administrators systems T ted text the threat to tool tools Tor TP type UI under up US use user Users V val Valuation vents vulnerabilities vulnerability Ware Well Wi workstation x z