AI BriefingFuriosaAI's RNGD Unveiled at Hot Chips 2024
Key point
FuriosaAI unveiled RNGD, a new AI accelerator optimized for LLM and multimodal model inference, at Hot Chips 2024.
Details
FuriosaAI unveiled its new AI accelerator RNGD (Renegade) at Hot Chips 2024, designed for high-performance LLM and multimodal model inference in data centers.
RNGD combines a 180W TDP, a proprietary chip architecture, and advanced memory technology such as HBM3. Through this, it aims to simultaneously achieve the three core elements of high performance, power efficiency, and programmability that existing GPUs and AI chips have struggled with.
The chip development and verification process proceeded very rapidly. After receiving its first silicon samples from TSMC in May 2024, the hardware was up and running within a week, and by early June it achieved the milestone of running the Llama 3.1 model. The company is currently delivering the product to early access customers while preparing for data center deployment.
FuriosaAI plans to focus on advancing its software stack going forward. It is currently processing about 12 queries per second on the GPT-J 6B model, and plans to continue improving performance through software optimization.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Qwen3.5-based Small Decision Model 'Kev' Released with TypeSafe API Compatibility and Performance Metrics