MediaTek has announced the Dimensity 9600 Pro, its first chip built on TSMC's 2nm process and the first to use ARM's new C2 cores.
The mobile silicon packs more than 33 billion transistors and adopts a 2+3+3 CPU layout.
MediaTek says the move from 3nm to 2nm alone delivers a 14% performance gain and a 23% reduction in power consumption at the chip level.
CPU, cache, and benchmark gains
The Dimensity 9600 Pro uses a fourth-generation all-big-core setup led by two C2-Ultra super-big cores clocked at 4.55GHz with 2MB of L2 cache each.
It also includes three C2 Pro big cores at 4.35GHz with 1MB of L2, plus three more C2 Pro cores at 3.10GHz with 512KB of L2.
With 16MB of L3 cache, the chip offers 34.5MB of total cache, which MediaTek says is 21% more L2 cache than the Dimensity 9500.
MediaTek claims single-core CPU performance is up 17% and multi-core performance is up 15% over the Dimensity 9500.
The ultra-large cores consume 37% less power, while multi-core power consumption drops by 61% at matching peak performance.
In Geekbench 6.4, the Dimensity 9600 Pro scores 4,276 in single-core and 12,650 in multi-core tests.
The Dimensity 9500 scored 3,666 and 11,014 in the same tests, matching MediaTek's generational claims.
The chip is the first mobile platform to support LPDDR6 memory and UFS 5.0 storage.
LPDDR6 delivers 33% higher performance than LPDDR5X, while UFS 5.0 doubles the sequential read and write speeds of dual-channel UFS 4.0.
MediaTek paired the CPU upgrades with the G2-Ultra NX GPU, which it says offers 27% higher peak performance and 24% lower power draw at matching peak output.
Ray tracing performance has climbed by 18%, and the chip can push game frame rates as high as 185fps.
The Imagiq 1290 ISP handles 4K recording at 240fps for slow-motion capture.
Built-in H.266 hardware decoding cuts power consumption by 23% compared to software decoding alone.
MediaTek built the Dimensity 9600 Pro around an AI-native architecture with a dual NPU setup.
That includes a 10th-generation high-performance NPU 1090 and a second-generation low-power NPU for always-on sensing tasks.
The high-performance NPU's prefill performance rises 51% over the previous generation, while the low-power NPU cuts its own power consumption by 40%.
The Generative AI Engine 3.0 doubles INT4 computing power compared to its predecessor.
This lets the chip run mixture-of-experts models with up to 30 billion parameters entirely on-device, without relying on cloud processing.
MediaTek's Tianji Intelligent Computing Fusion Architecture manages scheduling across the CPU, NPU, and memory as one system.
The company says this allows large AI models and regular apps to run simultaneously without resource starvation, keeping power draw down even as output speed holds steady.
