On May 8, 2025, Imagination Technologies launched its new E-Series GPUs, designed to enhance graphics and AI processing at the edge.

These GPUs utilize a highly efficient parallel processing architecture, capable of scaling from two to 200 TOPS INT8/FP8 for demanding AI workloads.

Tim Mamtora, chief of innovation and engineering at Imagination, emphasized that the E-Series will centralize graphics and edge AI functions, which will reduce the need for additional AI solutions and lower overall system design costs.

The first E-Series GPU IP is anticipated to be available in Autumn 2025, with variations already in development for automotive, consumer, desktop, and mobile applications.

The architecture includes Neural Cores that support various AI number formats, enabling developers to optimize networks for performance, accuracy, and power efficiency.

Kristof Beets, vice president of product management at Imagination, pointed out the unique combination of AI processing components within the E-Series chip, setting it apart from competitors.

Phil Solis from IDC noted that the E-Series effectively addresses the challenge of balancing performance, efficiency, and flexibility for edge AI system designers.

With a legacy of energy efficiency, Imagination's PowerVR architecture is further enhanced by the E-Series Burst Processors, which contribute to lower power consumption for both AI and graphics tasks.

The E-Series maintains advanced graphics capabilities from previous Imagination GPUs, including ray tracing support, while integrating AI acceleration into each GPU core.

Key technologies in the E-Series include Neural Cores, which provide up to 200 TOPS for AI tasks, and Burst Processors, which improve average power efficiency by 35% for edge applications.

The programmable GPU architecture ensures adaptability to evolving AI and compute demands, supporting APIs like OpenCL and tools such as oneAPI and Apache TVM.

Designed for a wide range of applications, the E-Series GPUs cater to graphics, desktop applications, natural language processing, industrial computer vision, and vehicle autonomy.

Additionally, the E-Series doubles the number of hardware-backed virtual machines to sixteen, enhancing multitasking capabilities and allowing for the simultaneous handling of multiple graphics and AI workloads.