AI infrastructure Product
FuriosaAI unveiled RNGD as an accelerator for running AI models
FuriosaAI introduced RNGD at Hot Chips in August 2024 while early customers were sampling it. In July 2025, the company reported that LG AI Research had adopted RNGD to run EXAONE models. [1][2]
What it does
The launch described a data-center chip intended for language and multimodal model inference. FuriosaAI later documented a PCIe card and a named deployment with LG AI Research. [1][2][3]
Why it matters
Inference hardware is evaluated for how it serves particular models after training. A named EXAONE deployment gives buyers a specific workload to consider alongside the card specifications. [2][3]
Who it’s for
Data-center operators and machine learning teams selecting hardware for model serving can check supported software, system fit and performance on their own workloads. [1][3]
Key dates
Where it fits
The manufacturer provides a PCIe card specification and development software. Teams evaluating it should match the server and model requirements to the documented configuration. [1][3]
Limits and trade-offs
The original wider availability date was a forecast. The later LG announcement verifies one adoption, with its reported performance tied to the stated model and test configuration. [1][2]
Sources and verification
Catalog evidence checked How we verify records
- Primary source FuriosaAI’s dated Hot Chips RNGD introduction and sampling disclosure
- Primary source FuriosaAI reports named LG AI Research RNGD deployment
- Primary source Furiosa RNGD PCIe Card — FuriosaAI
- Other source FuriosaAI RNGD at Hot Chips 2024
Related profiles
AI infrastructure
Azure Cobalt 200 virtual machines
Microsoft Azure
The Arm-based CPU platform supports databases, applications and tool execution around AI models.
AI infrastructure
Sarvam Token Factory
Sarvam AI
It exposes hosted model inference through supported developer interfaces.
AI infrastructure
NVIDIA Rubin HPC
NVIDIA
It is designed to accelerate high-performance computing workloads in compatible systems.