Imagination Technologies has unveiled its new E-series GPU architecture, designed to integrate AI capabilities directly into graphics processing. This new architecture features a "Neural Core" that can perform AI upscaling in 2.3 milliseconds for a 540p to 1080p resolution increase, and it supports low-precision formats like MXFP8 and MXFP4, boosting Prefill performance by 4.7 times for large language models. The E-series aims to reduce data movement and improve power efficiency by handling both graphics and AI tasks on a unified, programmable platform, though it still requires collaboration with other processing units in complex SoCs. AI
IMPACT This integrated AI approach in GPUs could lead to more efficient on-device processing for AI tasks and potentially reduce reliance on separate NPUs.
RANK_REASON Imagination Technologies, a GPU IP vendor, announced a new E-series GPU architecture with integrated AI capabilities. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- AI accelerator
- E series
- GDDR5
- GPU
- High Bandwidth Memory
- Imagination Technologies
- Jake Kochnowicz
- LLM
- LPDDR
- Markus Mosen
- MXFP4
- MXFP8
- OpenCL
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
