Imagination Technologies targets AI with E-Series GPUs
Imagination Technologies has unveiled its E-Series GPUs, designed to handle both traditional graphics and AI workloads with up to 200 trillion operations per second for INT8 and FP8 tasks. According to Kristof Beets, vice president of product management at the company, the architecture leverages tile-based rendering—a technique commonly used in graphics—to enhance efficiency in AI computing by keeping data on-chip during matrix operations.
Unlike dedicated neural processing units, which offer high efficiency but limited flexibility, Imagination’s approach integrates AI acceleration directly into the GPU. This allows for the reuse of existing memory and processing resources without requiring separate hardware. The E-Series features a new Burst Processor that enhances power efficiency by up to 35 percent for edge AI applications by optimizing common operations, such as multiplication and addition.
As an intellectual property vendor, Imagination lets customers customize GPU configurations for specific needs—whether for automotive, mobile, or data center use—unlike NVIDIA, which sells fixed chip designs. The E-Series supports standard APIs, such as Vulkan and OpenCL, and is compatible with major AI frameworks through optimized libraries. It also features optional hardware ray tracing and matrix multiply units, scalable from small single-core designs to multi-core setups targeting high-performance computing.
