AMD is taking a new step toward improving AI performance in desktop PCs. The company is evaluating the development of a discrete NPU (Neural Processing Unit) card that would function similarly to graphics cards. This hardware could open the door to a new hardware class for desktop users by enabling more efficient and faster execution of AI workloads.

AMD Prepares for the AI Revolution

Rahul Tikoo, AMD’s client CPU manager, commented on the development, stating that the idea of a discrete NPU card is still in its early stages, but that the company is closely monitoring developments in this area. Tikoo stated that their current technological infrastructure is capable of rapidly entering this field.

AMD has recently attracted attention with its AI-focused developments, particularly in mobile processors. Using powerful integrated NPU engines in laptops equipped with Strix Point and Strix Halo APUs, the company has introduced solutions that enable the execution of large language models with 128 billion parameters in compact systems. However, these solutions are primarily focused on portable devices.

The new discrete NPU card could be a valuable alternative, particularly for professional content creators, AI developers, and researchers working with large language models. These cards are planned to be installed externally in computers, similar to graphics cards, and to execute AI tasks independently of the CPU or GPU.

Unlike traditional CPUs and GPUs, NPUs are specifically designed for running algorithms such as artificial neural networks. Their highly parallel processing capabilities offer significant speed and efficiency advantages, particularly in tasks such as matrix multiplication and multidimensional tensor operations, which are used in deep learning.