If you are a Cloud Service Provider dealing with high ownership costs and inefficient power consumption—this project developed the NR1 Server on Chip that increases AI accelerator utilization from ~30% to 100%. This allows for more economically viable AI applications at scale.
High-Efficiency AI Server Chip to Reduce Data Center Costs and Power Use
Imagine a computer where the brain is constantly waiting for a slow assistant to hand over data; that's how current AI servers work. This project replaces that slow assistant with a direct-access system, letting the AI processor work at full speed. It's like moving from a slow drive-thru window to a high-speed conveyor belt for data.
What needed solving
Current AI computing is CPU-centric, creating bottlenecks that lead to high power consumption and expensive ownership costs. This prevents companies other than hyperscalers from deploying AI at scale.
What was built
The NR1 Server-on-a-Chip (SoC), the NR1-M module, the NR1-S server, and a cloud-enabled SDK. These are bundled into a turnkey AI Inference Appliance.
Who needs this
Who can put this to work
If you are a software provider dealing with the complexity of deploying Large Language Models—this project developed a turnkey AI Inference Appliance. It provides a Private GPT and Agentic AI ready platform that works out of the box.
If you are an integrator dealing with power constraints at the edge—this project developed the NR1-M module and SoC. It delivers an order-of-magnitude improvement in energy efficiency for real-life AI applications.
Quick answers
How does this reduce the total cost of ownership (TCO)?
It removes the CPU from the critical data path and uses hardware-based control engines. This increases AI accelerator utilization from ~30% to 100%, reducing waste and cost.
Can this be scaled for large data centers?
Yes, the architecture enables linear scalability and the roadmap includes specific configurations for clustered data centers.
What is the IP or licensing model for the technology?
Based on available project data, the technology is developed by NeuReality Ltd as a proprietary Server on Chip (SoC) and SDK, though specific licensing terms are not listed.
How does it integrate with existing AI software?
The project includes a cloud-enabled SDK and a turnkey appliance that integrates compute, software, and orchestration into one platform.
What is the timeline for commercial availability?
The project period runs from 2023-07-01 to 2025-06-30, with the company already moving into functional inference appliance verification as of mid-2024.
Who built it
The project is led by a single SME, NeuReality Ltd, based in Israel. With a 100% industry ratio and no university or research partners, the project is purely commercially driven, focusing on rapid productization of the NR1 chip and its associated SDK.
Contact NeuReality Ltd regarding NR1 SoC and AI Inference Appliance
Talk to the team behind this work.
Contact us to find similar AI-centric hardware accelerators for your infrastructure.