South Korean AI startup FuriosaAI has partnered with LG Electronics to offer its RNGD (pronounced “Renegade”) AI accelerator to enterprise clients. FuriosaAI’s chips have passed performance tests conducted by LG AI Research using their EXAONE models. LG plans to provide the RNGD Server to customers across various industries, including electronics, finance, telecommunications, and biotechnology.
The RNGD architecture boasts a Tensor Contraction Processor (TCP) capable of delivering up to 512 TFLOPS of FP8 performance with a Thermal Design Power (TDP) of just 180W, significantly more efficient than existing GPU solutions. "After extensively testing a wide range of options, we found RNGD to be a highly effective solution for deploying EXAONE models," stated Kijeong Jeon, lead of the product unit at LG AI Research.
The RNGD Server design allows for up to eight accelerators housed in a single, air-cooled 4U chassis. While LG is best known for its consumer products like refrigerators and televisions, LG AI Research serves as the company’s hub for cutting-edge artificial intelligence research. The RNGD technology, which is now available for enterprise clients, achieved 2.25 times better performance per watt during LG’s evaluations compared to a GPU-based solution, generating significantly more tokens for LLMs under equivalent power constraints.
FuriosaAI has captured attention in the competitive AI chip market, previously attracting a buyout offer from Meta for $800 million, which the startup ultimately declined. Despite numerous AI chip startups, industry experts express skepticism about any of them truly competing with established giants like Nvidia. However, the market remains vibrant with opportunities for significant wins among the emerging companies.
