The Numenta Platform for Intelligent Computing (NuPIC™) is a software AI platform that draws on Numenta's decades of neuroscience research to enable the deployment of Large Language Models (LLMs) on CPUs with simplicity, scalability, and speed.

Deploy Large AI Models on CPUs

Though GPUs have been the dominant hardware platform for running large-scale AI models, the movement to CPU-based AI has begun. NuPIC incorporates our decades of neuroscience research to take advantage of modern CPU architectures. With NuPIC, CPUs have become not just an alternative, but the ideal solution, offering the flexibility that is necessary for today’s diverse computational needs.

Leverage the Flexibility of CPUs in Your AI Workloads

Let's get real: would anyone choose to run AI models on a GPU if they didn't have to? No, they wouldn't. And neither should you. NuPIC lets you take advantage of the flexibility of CPUs, which provides numerous advantages:

• CPUs do not rely on batching - While batching allows you to run larger models with higher throughput requirements, it can delay feedback and introduce unnecessary complexity and latency.
• CPUs allow you to run multiple models concurrently on the same server - Many applications today require a mix of generative and non-generative models, which you can run asynchronously and independently on CPUs.
• CPUs are easy to scale and manage - With their straightforward memory management system and flexible programming environment, CPUs integrate seamlessly with operating systems and scale easily.
• CPUs are widely available - GPUs have a 12 to 18 months waiting list, which has resulted in many companies to pausing their AI initiatives.

Accelerated computing on CPUs with NuPIC

When you combine Numenta's neuroscience-based performance enhancements and optimizations with the flexibility of CPUs, the benefits multiply

• AMX acceleration - Our strategic partnership with Intel gave us early access to the AMX instruction set designed for AI workloads. Running NuPIC on AMX provides the ultimate performance boost.
• Optimize for throughput or latency - With NuPIC, you decide whether you want to configure for batch processing and optimize for high throughput or configure for real-time applications and optimize for low latency.
• NuPIC models run 1.3X to 20X faster than their unoptimized counterparts.

NuPIC is 100% Private

NuPIC is fully containerized and runs completely on your private infrastructure, with support for on-premise, private cloud, and hybrid environments. You have full control over your models and when they are updated, your data, and data governance policies. Your data and models never leave your company walls, and Numenta never sees them.