MediaTek: Dimensity 9400+ Higher Clock CPU and Optimize NPU

: Dimensity 9400+ CPU CLOC

MediaTek: Dimensity 9400+ CPU with higher clock speed and optimized NPU

Image: MediaTek

MediaTek is upgrading the Dimensity 9400 and transforming it into the new Dimensity 9400+. This is primarily accompanied by a higher peak clock speed for the Cortex-X925. This time, MediaTek has made further changes to the chip that should improve the NPU’s AI performance and connectivity.

You know this from MediaTek: The flagship SoC for will be unveiled in the fall, and the optimized Plus model usually follows in the spring. This time, it also works with the Dimensity 9400+, which follows the Dimensity 9400. The first smartphones with the Dimensity 9400+ are still scheduled for this month. Cortex-X925 goes up to 3.73 GHz

More CPU performance is achieved by increasing the upper clock, which is achieved through binning, which is adopted accordingly. MediaTek takes the top Dimension 9400, where slightly more clock speeds are possible than with the other chips in the wafer. Starting at a maximum of 3.62 GHz for the individual ARM Cortex-X925, which is located at the top of the CPU, MediaTek increases to 3.73 GHz in the Dimension 9400+.

The CPU interpretation in general with its 1 + 3 + 4 design and the other clock frequencies of the chip remain unchanged. The “All Big-Core SoC” continues to rely on three Cortex-X4s with 3.3 GHz and four Cortex-A720s with 2.4 GHz below the Cortex-X925. The GPU, the Immortalis-G925 MC12, also comes from MediaTek, with 1612 MHz at the Dimensity 9400.

Faster Cortex-x925 for the CPU

Faster Cortex-x925 for the CPU (Image: MediaTek) Faster speculative decoding for the NPU

However, the MediaTek NPU 890 has an upgrade achieved through 20% faster speculative decoding, which MediaTek calls “SPD+.” Speculative decoding should ensure that tokens are generated faster for the user when inferring large speech models (LLMs). Instead of generating one token after another, multiple tokens are speculatively generated via a smaller language model (SLM), which can act more quickly. These tokens are verified by the large language model, if necessary, and then issued.

NPU Optimization

NPU Optimization (Image: MediaTek) Corresponding support for this SLM and LLM mix is ​​available on MediaTek for the Chinese distillate Deepseek-R1 in the QWEN-1.5B, QWEN-7B, and LLAMA-8B variants. Accelerated Location Determination with BEIDOU

The other key chip data corresponds to that of the Dimensity 9400 and is explained in detail in the corresponding article. For the Dimensity 9400+, however, MediaTek still notes that the Chinese satellite navigation system now has a “33% time to first fix,” i.e., until the first determination of the user’s location.

Connectivity Optimization

Connectivity Optimization (Image: MediaTek)

Techastuce received information about this article from MediaTek under NDA. The only requirement was the earliest possible publication time. Topics: ARM MediaTek Process Source: MediaTek

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top