MediaTek Dimensity 9300+ Released, AI Processing Becomes Faster

Morrissey Technology – MediaTek released its newest system on a chip (SoC) called Dimensity 9300+, the successor to the Dimensity 9300. Compared to its predecessor, the clock speed of the Dimensity 9300+ is higher, this is because the main core (Cortex-X4) is set at 3.4GHz, not 3.25GHz like its predecessor. Meanwhile, the other three Cortex-X4 cores still have a speed of 2.85GHz, and the four power-saving cores are Cortex-A720 2GHz.

So even though the chip is still produced using the same 4nm+ process from TSMC, the performance can be faster. The GPU is still the same, namely the Immortalis-G720 which is equipped with ray tracing. The Dimensity 9300+ integrates a 2nd generation hardware raytracing engine with an Arm Immortalis-G720 GPU, providing gamers with a fast raytracing experience at a smooth 60 FPS, complemented by console-standard global lighting effects. Improvements also occurred in the AI ​​Processing Unit (APU), specifically the APU 790 FOR4D which promises a performance increase of 10% for AI processing because it uses NeuroPilot Speculative Decode Acceleration technology.

The powerful APU 790 AI engine enables the Dimensity 9300+ to not only support large language models (LLM) with parameters 1B, 7B, and 13B, with scalability up to 33B, but also run LLM quickly and efficiently. With NeuroPilot Speculative Decode Acceleration technology, Dimensity 9300+ can run LLM FOR4D with seven billion parameters at 22 tokens per second, faster than its rivals. With MediaTek’s Imagiq 990 ISP offering 18-bit RAW processing, the Dimensity 9300+ is promised to produce high quality photos and videos, even in low-light environments.

The Dimensity 9300+ is ideal for a wide range of AI applications, offering :

Support for LoRA Fusion and NeuroPilot LoRA Fusion 2.0 on devices so developers can quickly market new generative AI applications with text, images, music, and more. Support for the latest LLMs includes 01.AIYi-Nano, Alibaba Cloud Qwen, Baichuan AI, ERNIE-3.5-SE FOR4D, Google Gemini Nano, and Meta Llama 2 and Llama 3. Delegate ExecuTorch solutions to ecosystems that apply inference on devices.

“Dimensity 9300+ will help us continue to build a rich ecosystem of generative AI applications thanks to extensive LLM support on the chip and LoRA Fusion on the device,” said JS Hsu, Corporate Senior Vice President of MediaTek.