Summary
- Growing need for advanced Security Operation Centers (SOCs) to counter new types of cyberattacks
- Developed the AiR appliance, an NPU-based AI security assistant
- Achieved GPU-level performance with the ATOM NPU while cutting power consumption by 50%
- Verified significant TCO savings across hardware acquisition, floor space, and long-term power and operating costs
Challenge
Today’s Security Operation Centers (SOCs) face a trilemma: They require deep threat intelligence and AI-driven tools to understand new types of attacks, they must ensure airtight data security to prevent information leaks and they must operate cost-efficiently.
As a leader in Korea’s security market, IGLOO Corporation set out to address this challenge by developing AiR (AI Road), a security assistant powered by generative AI. The first priority in development was data protection. Many enterprise clients were concerned about sensitive logs and internal data being exposed. This made it necessary to build an on-premise solution instead of relying on cloud services. At the same time, the system had to support real-time LLM inference within this secure environment, while also overcoming the high power consumption and costs associated with GPUs.
Solution
IGLOO Corporation created AiR as a security assistant capable of assessing threats, explaining the reasoning behind each judgment, and generating predictive analysis. These capabilities allow SOC operators to cut through complexity and make informed decisions more quickly and accurately.
For deployment, IGLOO required a solution that maintained on-premise security, supported LLM workloads, and provided strong energy efficiency. Rebellions’ NPU and SDK proved to be the optimal choice.
On Rebellions’ NPU infrastructure, IGLOO successfully built AiR in appliance form, meeting strict security requirements. Thanks to the Rebellions SDK, which supports a wide range of classification and LLM models, IGLOO easily migrated its classification AI, explanatory AI, and generative AI workloads from GPU to the NPU-based on-premise environment. The result was a system that delivered strong performance with far lower power consumption than GPUs.
Result
Prototype testing of the AiR appliance showed that Rebellions’ NPU outperformed expectations. It delivered higher tokens-per-second (TPS) throughput than GPGPUs, proving faster inference speeds, and also demonstrated superior energy efficiency by consuming only half the power. On a performance-per-watt basis, the NPU was 1.6 times more efficient than GPUs.
With these results, IGLOO succeeded in developing a generative AI-based security appliance that offers new value to customers. Combining NPU servers with AiR improved the accuracy of generative AI analysis, reinforced the stability of the SOC environment, and significantly reduced TCO. Furthermore, because AiR can be deployed as a compact appliance rather than requiring large servers or costly security hardware, it substantially lowers the barrier to adopting AI-driven SOC operations.
“By leveraging Rebellions’ NPUs, we accelerated the development of an AI security appliance that delivers both high efficiency and high performance. Through this collaboration, we will help more security organizations strengthen their ability to detect and respond to cyber threats.”
— Head of AI Research, IGLOO Corporation