Furiosa AI's 'RNGD (Renegade)' will support the implementation of LG's large-scale language model (LLM) 'EXAONE' infrastructure, and will begin to attack the AI ecosystem of LG affiliates and partners in earnest. Expectations are high that it will emerge as an alternative to GPU in the future.

AI semiconductor company Furiosa AI (CEO Baek Jun-ho) announced on the 22nd that its second-generation AI inference accelerator 'RNGD (Renegade)' has been fully implemented in LG's large-scale language model (LLM) 'EXAONE'.

Building on this collaboration, the two companies will soon launch an enterprise ExaOne solution based on Renegade.

This collaboration is seen as a strategic turning point in improving AI data center power efficiency and providing an alternative to GPUs.

In approximately eight months of pilot testing, Renegade met all of LG AI Research's high-performance requirements while delivering a 2.25x performance-per-watt improvement over existing GPUs.

By overcoming the limitations of GPUs, which were previously criticized for excessive power consumption, we simultaneously achieved a reduction in the total cost of ownership (TCO) of AI data centers and maximum power efficiency.

The solution can be applied across all areas of AI infrastructure, including top-of-rack switches, power supply units (PSUs), battery backup systems, and DC-DC networking.

Furiosa AI cited significant improvements in power density and stability through optimization of the 48V intermediate bus converter and power module design as innovation points.

In particular, as major technology leaders focus on securing AI sovereignty, the introduction of Renegade has high strategic value in that it demonstrates the possibility of self-reliance in AI infrastructure based on domestically produced NPUs.

It is attracting attention as an important model that can reduce dependence on GPUs and operate AI workloads with its own solution.

This case marks the first landmark domestic introduction of hardware other than GPUs into a large-scale enterprise LLM inference environment.

Through this, Furiosa AI has gained technological prowess and reliability.Zeng plans to launch a full-scale attack on the enterprise AI solution market targeting Fortune Global 500 companies.

“Renegade offers outstanding absolute performance and outstanding infrastructure TCO reduction,” said Jeon Ki-jeong, head of the LG AI Research Product Unit. “The fast and easy model support process is also a major strength.”

"ExaOne is emerging as a national foundation model for Korea," said Baek Jun-ho, CEO of Furiosa AI. "Through continued collaboration, we will contribute to building a high-performance national AI infrastructure." He added, "This collaboration will serve as a model for companies around the world to proactively design and operate AI infrastructure."

FuriosaAI is preparing to support the ExaOne 4.0 model, the successor to the ExaOne 3.5, and will soon launch an on-premise, turnkey enterprise-grade ExaOne AI solution based on Renegade.

This is expected to accelerate the spread of domestic NPUs that focus on replacing GPUs and power efficiency across the AI ecosystem.