Lecture at Seoul National University AI Semiconductor Forum on the 3rd
Inference-specialized AI Semiconductor Renegade
Currently Under Performance Evaluation by LG and Aramco
Baek Junho, CEO of FuriosaAI, said on the 3rd, “Starting January next year, North American clients will begin full-scale performance evaluations of the next-generation AI semiconductor 'Renegade (RNGD)',” adding, “Samsung Electronics will start performance evaluations from mid-January.”
At the ‘2nd Seoul National University AI Semiconductor Forum (SAISF)’ held at Seoul National University Gwanak Campus in Gwanak-gu, Seoul, hosted by the SNU Graduate School of AI Semiconductor, CEO Baek stated, “Currently, LG is evaluating Renegade’s performance using the ultra-large AI model Exaone, and Saudi Aramco is also testing the Renegade 70B model.” He further added, “Mass production sales are expected to begin from the second quarter of next year.”
On the 3rd, Baek Junho, CEO of FuriosaAI, is holding and explaining the inference-strong AI semiconductor 'Renegade (RNGD)' at the '2nd Seoul National University AI Semiconductor Forum (SAISF)' held at Seoul National University Gwanak Campus in Gwanak-gu, Seoul. Photo by Choi Seoyoon
Renegade is the world’s first inference AI semiconductor equipped with HBM3, developed by FuriosaAI over approximately four years. The first samples were received from TSMC in May, and the product was first unveiled at the semiconductor industry academic event ‘Hot Chips 2024 Conference’ held at Stanford University in the U.S. last August. A demo was also showcased for the representative large language model (LLM) Llama 3.1 70B model, proving its technological capabilities in the global AI market.
Renegade offers a processing speed of 12.4 QPS (Queries Per Second), similar to NVIDIA A100, while providing 50% better power efficiency. A higher QPS means it can handle more user requests. Queries refer to questions, requests, or commands input by users. CEO Baek emphasized Renegade’s performance and efficiency by stating, “By adopting HBM3 memory, we solved the memory bottleneck issue, which is considered the biggest problem in LLMs.”
CEO Baek said, “Renegade has a TDP (Thermal Design Power) level of 150 to 200W, maintaining high compatibility with existing data center environments and delivering optimal performance in LLM inference tasks.” TDP refers to the maximum amount of heat generated when a semiconductor or processor operates at full load. The lower the TDP, the less power consumption and the higher the energy efficiency of the chip.
In the NVIDIA-centered GPU ecosystem, FuriosaAI is expected to create a new competitive landscape with its proprietary technology. CEO Baek stated, “AI semiconductors go beyond simple performance competition; sustainability through energy efficiency and cost-effectiveness is key,” adding, “FuriosaAI focuses on providing sustainable AI computing solutions.”
CEO Baek previously worked as an engineer in Samsung Electronics’ Memory Division and as an engineer in the GPU team at the U.S. semiconductor company AMD before founding FuriosaAI in 2017. The company develops system semiconductors optimized for AI inference operations. This year, it has completed the selection of underwriters for its IPO and is reviewing mid- to long-term listing plans.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
