Xiaomi’s 3rd generation in-house mobile processor, the XRing O3, has been officially unveiled, and the major jump in CPU and GPU performance alongside a stronger focus on on-device AI and gaming are going to challenge the existing players and their market share.
Built on TSMC’s N3P process, the XRing O3 measures 133 mm² and packs 24 billion transistors. Xiaomi has also moved away from traditional efficiency cores, instead using an all-big-core CPU design made up of two C1-Ultra cores running at up to 4.35 GHz, four C1-Premium cores reaching 3.68 GHz, and four C1-Pro cores topping out at 3.15 GHz.
The CPU setup comes with 12 MB of private L2 cache, 16 MB of shared L3 cache and another 16 MB of system-level cache, giving the chip 44 MB of on-chip cache in total. It also includes two SME units for accelerating certain workloads
In Xiaomi’s testing, the XRing O3 reached 3,945 points in Geekbench 6.5 single-core and 15,221 points in multi-core. That’s a claimed 31% and 61% improvement over the previous XRing O1, respectively. They also claimed the chip’s multi-core score leads the mobile SoC market, while power consumption under low-to-medium workloads can drop by up to 25% compared with the previous generation.
The GPU is another major change, in which the unreleased Mali-G2 Ultra NX MC16 has been mentioned here, and it features 16 compute units and 4 MB of dedicated L2 cache. Eight of those compute units come with dedicated NX tensor accelerators, allowing AI workloads to be handled directly inside the GPU.
Xiaomi claims the GPU can deliver the same peak performance as the XRing O1 while using 64% less power. In Steel Nomad Lite, it scored 4,567 points, an 85% improvement over the O1, while 3DMark Solar Bay Extreme reached 3,628 points – 182% higher than the previous chip and 63% ahead of Apple’s A19 Pro in Xiaomi’s comparison.
That setup is particularly useful for gaming. Xiaomi says the XRing O3 can perform AI upscaling and frame generation within the GPU without sending the image data back to the system-level NPU or main memory. This helps reduce memory bandwidth demands and latency, while also cutting power consumption.
The GPU can also upscale games from 540p to 1080p while consuming 20% less power than native 1080p rendering, according to Xiaomi. Its AI-based rendering features support upscaling from 1080p to 3.4K and frame generation from 60 FPS to 120FPS.
For AI workloads outside the GPU, the XRing O3 features a dedicated quad-core NPU capable of 200 TOPS using A8W4 precision, alongside 3.13 TFLOPS of vector throughput. Xiaomi has also worked with its MiMo model team on a five-value quantization model specifically optimized for the new NPU.
The chip’s fifth-generation flagship ISP gets an upgrade as well, supporting a single camera sensor of up to 432 MP or a triple-camera setup consisting of 64 MP + 64 MP + 50 MP sensors. It can process 4K60 AI noise-reduced night video and includes hardware optical flow, multi-camera and multi-frame processing, plus a Mini ISP for always-on camera functions.
Another notable addition is native LPDDR6 support. The XRing O3 supports memory speeds of up to 10,667 MT/s across four 24-bit channels, providing up to 113.8 GB/s of bandwidth. Xiaomi says its unified fusion bus, metal routing and prefetch technology help achieve a memory access latency of 82 ns.
The SoC also supports H.266 hardware decoding and uses a RISC-V-based security core for trusted execution, with support for SM-series cryptography and CCRC EAL5+ certification. Xiaomi has additionally implemented hybrid hardware-software scheduling with a granularity as low as 1 ms.
The XRing O3 is set to make its debut in the Xiaomi 18 Fold and Xiaomi Pad 9 Pro Max, both of which are scheduled to launch in September.
The post Xiaomi’s new XRing O3 chip could shift the mobile chip market if it can deliver everything it promises appeared first on Nasi Lemak Tech.