AI BriefingHot Chips 2024 Review: RNGD Global Debut
Key point
FuriosaAI globally unveiled RNGD, an AI accelerator for LLM inference, at Hot Chips 2024.
Details
FuriosaAI globally unveiled its new AI accelerator RNGD (Renegade) at Hot Chips 2024, designed for high-performance, high-efficiency LLM and multimodal model inference in data centers. On the Hot Chips stage, where technical experts gather, the company presented the product's design, specifications, software details, and benchmarks in depth.
On-site, a live demo of RNGD running Llama 3.1 8B and 70B models was conducted, and physical hardware featuring 4 RNGD units mounted in a Supermicro server was also on display.
Key questions from engineers and technical highlights include:
- An architecture capable of efficient computation without FlashAttention
- A proprietary general-purpose compiler that processes the entire model as a single fused operation
- An architecture designed for tensor contraction
This announcement received focused coverage from major semiconductor media outlets, including the EE Times main page, as well as Semiconductor Engineering and IEEE Spectrum.
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
Underdog Releases 'Husky', a Model-Specific Inference Engine Up to 4.5x Faster than MLX