AI BriefingRNGD Preview: FuriosaAI's Renegade AI Accelerator
Key point
FuriosaAI has unveiled **RNGD**, a highly efficient AI accelerator optimized for LLM and multimodal model inference.
Details
FuriosaAI has developed RNGD (Renegade), a highly efficient AI accelerator for large language model (LLM) and multimodal model inference. The chip is currently undergoing hardware sample testing, targeting release by the end of this year.
Unlike the latest GPUs, which consume up to 1,200W, RNGD is designed with a TDP (Thermal Design Power) of 150W, offering very high power efficiency. Notably, when running advanced LLMs, it achieves 3x better performance per watt compared to Nvidia H100.
Key technical specifications are as follows:
- Compute Performance: 512 TFLOPS (FP8) or 256/512/1024 TOPS (BF16/FP8/INT8/INT4)
- Memory: 48GB HBM3, 1.5 TB/s bandwidth
- Software: Native PyTorch 2.x support, model quantization API, Kubernetes support, etc.
By adopting a tensor-based hardware architecture, it can automatically optimize even new model structures, and a data multicast feature reduces SRAM access while increasing data reusability.
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