NeuReality Launches Accelerator-Agnostic NR1 Chip for AI Inference at Scale
June 6, 2025 -- NeuReality is touting its NR1 Chip as the first true AI-CPU purpose built for inference orchestration. It pairs with any GPU or AI Accelerator to boost effective GPU utilization to near 100% compared to the average 30-50% with traditional host CPU/NIC architecture in today’s inference servers.
The AI Accelerator-agnostic NR1 Chip replaces traditional CPUs and NICs that bottleneck AI workloads except with six times the processing power to drive maximum GPU throughput and AI inference at scale.
For years, GPUs evolved to meet AI's demands, becoming faster and more powerful. But traditional CPUs—designed for the Internet era, not the AI age—have remained mostly unchanged, creating a growing bottleneck as AI models grow increasingly complex and multiple AI queries grow in volume. Our accelerator-agnostic AI-CPU features a low-power engine that combines essential CPU functionality with dedicated media and data processors, a hardware hypervision layer, and comprehensive networking and connectivity IP—delivering drastically better performance, lower energy consumption and business ROI. In fact, using the same generative AI model running on the same AI accelerator, proof of concept demonstrations show that the NR1 achieves 6.5x greater AI token output for the same cost and power envelope - compared to x86 CPU-centric architecture running the same GPU.
In line with the current movement towards separating storage and compute resources, the disaggregation of AI resources enables a streamlined isolation of AI compute from the broader system. This separation is particularly crucial in data center and cloud workflows. Conventional software-operated CPU-centric platforms face challenges such as high costs, power consumption, and system bottlenecks when handling AI inference. The complexities and cost barriers of today's infrastructure often impede the complete realization and deployment of various inference possibilities.
Addressing these concerns, the NR1 Chip is ingeniously designed with comprehensive AI pipeline offload capabilities. Its hardware-based NR1 AI-Hypervisor hardware IP takes charge of data-path processing and job scheduling, encompassing pre- and post-processing engines, the NR1 AI-over-Fabric networking engine, and a built-in management and abstraction controller. The outcome is redefined price/performance and the lowest operational costs, characterized by low power consumption, minimal latency, and linear scalability. To enable DevOps and MLOps, the NR1 is accompanied by a complete software development kit (SDK) and a K8s-based service layer for ease of use and deployment.
Target Markets and Technologies
- Finance and Insurance
- Healthcare and Pharma
- Government and Education
- Telecommunications
- Retail & eCommerce
- Generative and Agentic AI
- Conversational AI
- Computer Vision
- Single & Multi-Modal AI Models
Media and Data Compute
Kernel libraries for processing of Vision, Audio, Text, Recommendations
- 4x Video/JPEG decoders
- 16x Audio/Speech DSPs
- 16x General Purpose vector DSPs
NR1 AI-over-Fabric embedded network engine
- 2x 10/25/50/100 GbE
- AIoF (over TCP / ROCEv2) for high efficiency and reduced latency
- Client-server & server-server links support
- Line rate cryptography
- 2 tiers of isolated network functions (CSP networking)
About NeuReality
Founded in 2019 by a seasoned team of system engineers, NeuReality Ltd. is an AI technology innovation company that creates purpose-built AI Inference system architecture, silicon, hardware, and software for the ultra-scalability of current and future AI applications. Its cutting-edge technology transforms how companies operate daily AI inferencing with its holistic, ready-to-use NR1 AI Inference Solution that supports limitless deep learning models and customer choice in hardware providers and open-source software. In its quest to democratize AI and unleash greater human achievements, NeuReality AI solutions are easily accessible, adaptable, and affordable for all governments and businesses large and small – with a robust set of leading industry partners to deliver and deploy. For more information, visit https://neureality.ai.
Source: NeuReality
