Arm unveils Mali G2-Ultra NX GPU bringing dedicated AI neural rendering to mobile
Making its commercial debut in Xiaomi's 18 Fold, the architecture integrates dedicated neural accelerators directly into shader cores.
Original cover art, generated for this story. THE VISSION does not republish third-party press imagery.
- Arm has introduced its CSS for Mobile 2 platform, featuring the Mali G2-Ultra NX graphics processor.
- The GPU integrates dedicated neural accelerators (NX) to handle AI-driven image upscaling and frame rate generation.
- The Xiaomi 18 Fold is the first device to utilize the chip, powered by Xiaomi's custom XRING O3 silicon.
Arm has officially announced its next-generation mobile compute platform, CSS for Mobile 2, headlined by the Mali G2-Ultra NX graphics processing unit. The release marks a significant architectural shift toward AI-native mobile graphics, moving away from offloading image processing to a separate central NPU. Instead, the Mali G2-Ultra NX integrates dedicated Neural Accelerators (NX) directly into its shader cores, enabling low-latency, real-time AI-driven rendering.
The dedicated neural hardware enables a suite of high-fidelity graphics features, including Neural Super Sampling (NSS) and Neural Frame Rate Upscaling (NFRU). NSS upscales lower-resolution mobile games to high definition, while NFRU dynamically doubles framerates using motion estimation. Arm claims this neural rendering pipeline delivers up to a fourfold improvement in frame rates and power efficiency while reducing ray-tracing computational workloads by 70 percent compared to conventional hardware pipelines.
The Mali G2-Ultra NX has made its commercial debut inside the newly launched Xiaomi 18 Fold folding smartphone, which is powered by Xiaomi's custom XRING O3 system-on-chip. The XRING O3 integrates a 16-core configuration of the new Arm GPU alongside up to 16 gigabytes of high-bandwidth LPDDR6 memory. The Xiaomi 18 Fold features a wide passport-style folding form factor, Leica-tuned camera systems, and dual 4,000-nit AMOLED displays, retailing in China starting at 10,000 yuan.
By placing dedicated neural accelerators directly into GPU shader cores, Arm is establishing a new paradigm for mobile silicon design. For developers and game engines, this integration transitions AI upscaling from an expensive software-level trick into a highly optimized hardware-level primitive, allowing developers to target desktop-class lighting, ray-tracing, and physical simulation on compact, thermally constrained mobile devices.
Will other major mobile chipmakers, such as Qualcomm or MediaTek, release competing GPUs with dedicated shader-core neural accelerators by the end of 2026?
Still open. When the paper finds out, it will say so here and on the open questions page — including if it got this wrong.