Furiosa: 3.5x efficiency over H100s (furiosa.ai)

FuriosaAI has unveiled its NXT RNGD Server, a groundbreaking turnkey solution for AI inference designed to optimize performance at data center scale. Utilizing RNGD accelerators, this system promises 3.5 times the efficiency of the NVIDIA H100s, enabling enterprises to transition swiftly from experimentation to full deployment. With preinstalled software, including the Furiosa SDK and LLM runtime, organizations can hit the ground running without the need for specialized infrastructure. The server operates at just 3 kW, making it compatible with most current data center power and cooling capabilities, thus offering a scalable and cost-effective solution for AI workloads. Significantly, the NXT RNGD Server is poised to address the considerable demands of the growing AI market, particularly as data center energy needs are predicted to double by 2030. It not only supports diverse data types (BF16, FP8, INT8, INT4) but also maintains security and management features, providing enterprises complete control over sensitive data. The platform's successful validation by LG AI Research for applications like the EXAONE model further highlights its practicality across various sectors, including finance and biotechnology. As the industry shifts towards more efficient AI infrastructure, Furiosa’s innovation represents a vital step in leveraging existing facilities without incurring excessive energy costs.