MediaTek's new chipset brings generative AI to mid-range smartphones
After launching the Dimensity 9300 as its answer to Qualcomm's Snapdragon 8 Gen 3 chipset, MediaTek has now unveiled the Dimensity 8300 smartphone chipset, marking its foray into the mid-range segment with on-device generative AI support.
The latest addition to the Dimensity 8000 series brings generative AI capabilities, energy-efficient performance, adaptive gaming technology, and swift connectivity, aiming to deliver flagship-level experiences to the mid-range 5G smartphone market.
Dr. Yenchi Lee, Deputy General Manager of MediaTek's Wireless Communications Business Unit, emphasized, "With MediaTek's optimized Dimensity 8000 series, consumers don't have to compromise between accessibility and premier experiences like flagship-grade memory or accelerated AI capabilities—they can have it all."
MediaTek highlights that the chip’s APU 780 AI silicon supports large language models (LLMs) with up to 10 billion parameters. The company asserts an 8x speed boost for transformer-based generative AI and a 2x improvement in integer and floating-point computations. According to AI Benchmark v5, there is a significant 3.3x increase in AI performance, promising a substantial leap in AI capabilities for mid-range mobile devices.
The Dimensity 8300 introduces support for LPDDR5X RAM at speeds up to 8,533Mbps, a necessity for running LLMs effectively. Fabricated on TSMC's 2nd generation 4nm process, the chipset boasts an octa-core CPU featuring four Arm Cortex-A715 cores and four Cortex-A510 cores built on Arm's latest v9 CPU architecture. This configuration results in a 20% faster CPU performance and a 30% peak gain in power efficiency compared to its predecessor.
Speaking of gaming–MediaTek’s new chipset incorporates "adaptive gaming technology," optimizing smartphone gaming through AI performance prediction. This not only reduces power consumption at maximum speeds but also offers a 24% faster scene change. Short video creation benefits as well, with 15% faster AR preview and 6% faster rendering compared to alternatives.
The Dimensity 8300 also brings enhancements in the camera department, supporting 4K/60fps HDR video (replacing its predecessor’s 4K/60fps SDR video), more power-efficient video recording, and AI-Color functionality. And that is not all, as AV1 decoding, Bluetooth 5.4, Wi-Fi 6E, and the ability to handle refresh rates of up to 120Hz at WQHD+ resolutions or 180Hz at FHD+ are among the chipset’s features.
MediaTek anticipates the first phones featuring the Dimensity 8300 to launch before the end of the year. According to GSMChina, the Redmi K70E, expected to debut later this month, will be among the first smartphones to showcase the Ultra variant of the new Dimensity 8300.
The latest addition to the Dimensity 8000 series brings generative AI capabilities, energy-efficient performance, adaptive gaming technology, and swift connectivity, aiming to deliver flagship-level experiences to the mid-range 5G smartphone market.
MediaTek highlights that the chip’s APU 780 AI silicon supports large language models (LLMs) with up to 10 billion parameters. The company asserts an 8x speed boost for transformer-based generative AI and a 2x improvement in integer and floating-point computations. According to AI Benchmark v5, there is a significant 3.3x increase in AI performance, promising a substantial leap in AI capabilities for mid-range mobile devices.
Moreover, the Mali-G615 MC6 GPU upgrade in the Dimensity 8300 results in a 60% boost in performance and a 55% improvement in power efficiency. Budget-friendly phones sporting this chip might be a good choice for savvy gamers watching their wallets.
MediaTek anticipates the first phones featuring the Dimensity 8300 to launch before the end of the year. According to GSMChina, the Redmi K70E, expected to debut later this month, will be among the first smartphones to showcase the Ultra variant of the new Dimensity 8300.
Featured Stories
Popular stories
FCC OKs Cingular\'s purchase of AT&T Wireless
Things that are NOT allowed: