Imagination Technologies is presenting its graphics processing units of the electronic series (GPU) for graphics and processing of AI on the edge.
Imagination said the products redefine the design of the AI system and edge graphics with the launch, heprining a highly efficient parallel processing architecture to provide exceptional graphic performance and at the same time climb from INT8/FP8 tops for AI work collisions.
In an informative session, Kristof Beets, Vice President of Imagination Products Management, told me that GPUs have an intelligent mixture of AI processing components within the chip to more efficiently handle the processing tasks in question. That helps distinguish it from other GPUs in the market and allows imagination to go to specialized markets as Automotive.
He said that the GPU family sacrifices a versatile and programmed solution for future edge applications that include graphics, desktop applications, natural language processing on smartphones, vision of the industrial computer and vehicle autonomy.
Two new technologies support the potential of the electronic series to transform the design of the edge system:
- Neuronal nucleus: climbing up to 200 tops (INT8/FP8), these nuclei offer significant acceleration for AI and calculations workloads.
- Gusting processors: A highly innovative solution that offers an improvement of 35% in average energy efficiency for edge applications.
“The AI on the device is quickly evolving, but Edge’s Systems designers still face challenges to balance performance and flexibility efficiency,” said Phil Solis, IDC investigation director, in a statement. “The imagination has taken advantage of its long -standing experience in the development of GPU with electrical efficiency and has evolved to admit both graphic workloads and AI for AI in the AI device.
High performance acceleration for low power AI
The electronic series continues to offer the advanced graphics capabilities of the previous generations of imagination GPU, including the support for the lightning layout. To this, add a deeply integrated acceleration for low -precision and efficient the AI operations in each GPU nucleus. This creates the neuronal nuclei of the E dense series in computation that scale up to 200 tops int8 and unleashes up to 400% of the AI performance of the previous series D.
The neural nuclei admit a wide range of popular number formats, which allows developers to design networks that meet a broad spectrum of demand, precision and energy demands. One of its many performance efficiency measures is a friendly memory architecture with AI that prioritizes local memory for calculation, largely reducing power costs and performance to go to external memory.
Ai programmable for the design of future proof systems
GPUs are programs that the future -proof devices against the continuous evolution of the workloads of AI, computation and graphics. The neuronal nuclei of the electronic series are aligned with the widest GPU and the heterogeneous computing software ecosystem by deeply integrating the acceleration of AI in the GPU.
Their capacities can be unlocked by Popular API, such as OpenCl, and developers can easily move their work loads to neural nuclei using open standards and tools such as ONEPI, Apache TVM or Litert. Imagination computing libraries and the highly optimized graphic compiler maximize the efficiency of the GPU.
“Hardware and software integration Edge AI is crucial to unlock the potential of intelligence in the device,” said Parv Sharma, a senior analyst at Counterpoint Research. “The electronic series allows developers to implement AI algorithms in disposal in multiple applications and final devices.”
Efficient processing for the yield of AI and sustained edge graphics
The Imagination Powervr GPU architecture is recognized for its energy efficiency and has been implemented in energy -limited devices for almost twenty years. The new ESERIs rupture processors technology improves energy efficiency by an additional 35% for AI work loads, user games and interfaces. This improvement is achieved by reducing the depth of the pipe and minimizing the movement of data within the GPU.
The GPU that makes more
Modern devices are increasingly complex and the processors are requested to admit multiple graphs and workloads of the AI simultaneously. Ensuring the quality of the service (QOs) and the clear prioritization between the thesis workloads is essential for the user experience.
The E series improves the multitasking capabilities of the previous generations by doubling the number of virtual hardware overload machines backed by imagination GPU to sixteen, with sophisticated QO support. Multiple variants or electronic series GPUs can take advantage of additional nuclei for additional performance or improved flexibility. These GPUs are capable or handled multiple graphic workloads, multiple workloads of AI or a combination of both simultaneously.
“The electronic series will place the GPU in the center of the graphics and the IA EDGE systems,” said Tim Mutter, Chief of Innovation and Imagination Engineering, in a statement. “For systems designers who need to execute graphics and calculate workloads, an electronic series GPU is a versatile solution that eliminates the need for fixed function or additional function solutions, which provides flexibility costs to the proof of the future.”
The first IP GPU of the electronic series is available in the autumn of 2025 and already has a leg license. Automotive, consumption, desktop and mobile variations are under development.
