MediaTek’s Dimensity 9300+ May Bring Faster AI to More Android Phones

Why Trust Techopedia
Key Takeaways

  • Mediatek introduced the Dimensity 9300+, the successor to last year’s 9300.
  • The semiconductor giant claimed the product will propel generative AI processing.
  • The chip could debut in a premium US Android smartphone, but no date has been confirmed for its release.

MediaTek has unveiled its latest flagship mobile chip, the Dimensity 9300+, pledging significant upgrades over its 9300 predecessor.

The new system-on-chip is said to boost generative AI processing at the edge to speed up large language models (LLMs) like Google’s Gemini Nano and Meta’s Llama 2.

The new chip centers on an all-big-core design with four Cortex-X4 cores, including a primary 3.4GHz core that improves on the 3.25GHz of the ordinary 9300.

MediaTek’s NeuroPilot Speculative Decode Acceleration technology is said to boost AI performance. In tandem with a more powerful “AI engine,” it will reportedly allow LLMs to operate with up to 33 billion parameters. The chip can run several billion parameters at a rate of up to 22 tokens per second. 

The company outlined how Dimensity 9300+ uses on-device LoRA Fusion and NeuroPilot LoRA Fusion 2.0, both of which can assist developers wanting to run generative AI apps.

Beyond AI, the Dimensity 9300+ is also said to outperform its ancestor in gaming while saving more battery life and data usage.

The new part’s release is not imminent. However, 9to5Google said an update might come within weeks as a high-end phone is slated to use MediaTek’s new Dimensity chip.

Last week (May 2nd), the company suggested one of its chips will be launched in a premium Android smartphone in the U.S. for the first time. While there were no details on the device or chip, the timing suggests the 9300+ is the likely candidate.

The upgrade could help spur the adoption of AI on Android phones in some countries. While Qualcomm’s Snapdragon line dominates North American and European markets, MediaTek has a large footprint in countries like China. This could make generative AI considerably more viable in some regions.