On-Device AI for Business with AMD Ryzen™ AI PRO Processors

An AMD AI PC is a system that includes a dedicated neural processing unit (NPU) in addition to the CPU and GPU. While many traditional PCs can run AI workloads using the CPU or GPU, an AMD AI PC is specifically designed to execute AI tasks across all three processing engines based on the characteristics of the workload. Non-AI PCs typically rely on the CPU, integrated graphics, or a discrete GPU to run AI workloads. In contrast, AMD AI PCs include an on-die AI engine that enables additional flexibility and potential efficiency by offloading suitable AI workloads to the NPU. This allows AI tasks to run with lower power impact while leaving the CPU and GPU available for other system processes. The distinction is not whether a PC can run AI workloads, but whether it includes dedicated AI hardware designed to support them efficiently and concurrently.

AMD Ryzen AI refers to the combination of AI-capable hardware and supporting software integrated into AMD Ryzen processors. This includes a dedicated AI engine (NPU), AMD Radeon graphics, and Ryzen CPU cores working together to enable AI capabilities across a wide range of workloads. In practical terms, AMD Ryzen AI encompasses both the silicon features that accelerate AI processing and the software stack required to take advantage of them. AI workloads may run on the CPU, GPU, or NPU depending on performance, latency, and power requirements. AMD Ryzen processors with NPUs use AMD XDNA™ and AMD XDNA2™ architectures, which are designed to support evolving AI workloads over time. Offloading AI tasks to the NPU enables multitasking with minimal performance impact, keeping the system responsive while AI processing runs in parallel.

AI workloads generally fall into two categories: training and inference. Training involves creating or refining an AI model using large datasets and typically requires substantial compute resources, often using clusters of GPUs or specialized accelerators. This work is usually performed in data centers or cloud environments. Inference is the process of applying a trained model to new data. Inference workloads are far more common in day-to-day use and can run across a wide range of hardware, including CPUs, GPUs, and NPUs. Many AI applications are designed to perform inference locally on PCs rather than relying on the cloud.

AMD Ryzen AI-based systems are optimized to support local inference workloads by running them on the most appropriate processing engine. The NPU is particularly well suited for sustained, low-power AI inference, while the CPU and GPU handle tasks that benefit from general-purpose or highly parallel processing.

“100+ AI-enabled experiences” refers to the growing ecosystem of applications and services that take advantage of AI acceleration on AMD Ryzen AI-based systems. This includes collaboration with independent software vendors (ISVs), operating system partners, and platform developers to enable AI features across productivity, security, creation, and technical workflows. These experiences span applications that use local inference, hybrid local-and-cloud processing, and OS-level AI integration. The breadth of this ecosystem reflects the AMD approach to AI enablement across multiple layers of the software stack rather than support for a single application or use case. As additional AI-enabled software is released, support for AMD Ryzen AI continues to expand through software updates and ecosystem collaboration, allowing systems to adapt as AI models and workloads evolve.

Running AI workloads locally on the PC can enable lower latency, improved responsiveness, and greater control over data. On-device AI allows inference to occur without reliance on network connectivity or cloud services, which can be important for privacy-sensitive or time-critical tasks. AMD Ryzen AI-based systems support on-device AI by distributing workloads across the CPU, GPU, and NPU. The NPU is designed to handle sustained AI inference efficiently, reducing power consumption while keeping other system resources available. This approach enables a broader range of AI use cases to run directly on the PC, complementing cloud-based services rather than replacing them.