Arteris Interconnect IP Deployed in NeuReality Inference Server for Generative AI and Large Language Model Applications

FlexNoC network-on-chip IP seamlessly provides connectivity across the NR1 chip within the inference server to efficiently meet high-density, low-latency AI performance needs at a minimal total cost of ownership.

Arteris,a leading provider of system IP which accelerates system-on-chip (SoC) creation announced that NeuReality has deployed Arteris FlexNoC interconnect IP as part of the NR1 network addressable inference server-on-a-chip to deliver high-performance, disruptive cost and power consumption improvements for machine and deep learning compute in its AI inference products. This integration is architected in an 8-hierarchy NoC with an aggregated bandwidth of 4.5TB/sec, meeting low latency requirements for running AI applications at scale and lower cost. The NeuReality inference server targets Generative AI, Large Language Models (LLMs) and other AI workloads.

Recommended:  Five Things You Should Do to Ace your Customer Service Strategy with Artificial Intelligence

“The new era of Generative AI with LLMs requires large-scale computing that is faster, easier, and less expensive.  We created a category of microprocessors for today’s AI-centric data centers supporting sustainability,” said Moshe Tanach, co-founder and CEO of NeuReality. “Arteris has earned a notable reputation in the market which together with their AI-ready network-on-chip technology were determining factors in our decision to adopt their FlexNoC IP for our AI server. This IP enabled us to successfully address AI performance requirements, scalability, high density, and low latency, all with a minimal total cost of ownership.”

NeuReality’s innovative NR1 server-on-a-chip, is the first Network Addressable Processing Unit (NAPU), which is a workflow-optimized hardware device with specialized processing units, native network and virtualization capabilities. It provides native AI-over-fabric networking, including full AI pipeline offload and hardware-based AI hypervisor capabilities. The ability to offload CPUs, GPUs and even deep learning accelerators to multiple NR1 chips is what makes it possible for NeuReality’s inference server to effectively deliver up to 10 times the performance with less power consumption and at a fraction of the cost in its inference server.

Recommended: Predictions Series 2022: AiThority Interview with Dr. Jack Zeineh, Co-Founder and CTO at PreciseDx

“Developing inference platforms for advanced AI and machine learning applications, such as Generative AI, is a complex process that requires a deep understanding of both software and hardware, along with state-of-art connected chip development,” said K. Charles Janac, president and CEO of Arteris. “We are thrilled to be working with NeuReality, and deploying Arteris IP to provide AI connectivity, supporting their vision of cost-effective, high-performance AI at scale.”

Recommended: Why Managers Should Train More with AI Devices and Intelligent Virtual Assistants

[To share your insights with us, please write to sghosh@martechseries.com]