Arm brings neural upscaling to Android phones with Xiaomi’s Xring O3

• Xiaomi’s 18 Fold is the first announced device with Arm’s Mali G2-Ultra NX GPU, inside the company’s Xring O3 chip. ^[1]
• Arm’s neural graphics stack includes Neural Super Sampling, Neural Frame Rate Upscaling and Neural Super Sampling and Denoising, with support through Vulkan and Unreal Engine tools. ^[2]
• Confirmed integrations include Where Winds Meet, Infinity Nikki and Arena Breakout: Infinite, but support is still limited to specific games and demonstrations. ^[2]
• Xiaomi’s 18 Fold starts at 10,999 yuan, with Chinese sales scheduled to begin on September 10. ^[3]

Arm’s first AI-native mobile GPU has reached a shipping device. Xiaomi’s 18 Fold, launched in China on September 7, uses the company’s Xring O3 system-on-chip with Arm’s Mali G2-Ultra NX graphics processor, giving Android developers a hardware path for AI upscaling and frame generation similar in function to technologies such as Nvidia’s DLSS. [1]

The launch does not mean that every Android game will receive an automatic graphics upgrade. Developers must integrate Arm’s neural graphics technology into a game or engine, and the first announced support is concentrated among a small group of Chinese studios. [2]

A GPU built around neural graphics #

The Mali G2-Ultra NX combines a conventional graphics pipeline with dedicated neural accelerators, a redesigned execution engine and a third-generation ray-tracing unit. Arm says the GPU delivers up to 14% higher performance on non-AI gaming workloads and up to 24% higher peak benchmark performance than the previous generation. Those are Arm’s own comparisons, not independent tests. [2]

Its neural features divide the work among three techniques. Neural Super Sampling, or NSS, reconstructs a higher-resolution image from a lower-resolution render using color, motion and depth information. Neural Frame Rate Upscaling, or NFRU, generates an intermediate frame between two rendered frames. Neural Super Sampling and Denoising, or NSSD, combines upscaling with denoising for ray-traced scenes. [2]

Arm says its Neural Dawn demonstration delivered up to four times higher performance efficiency and up to 70% lower external-memory traffic than native rendering when the techniques were combined. Those results come from an Arm demonstration, so they should not be read as a device-wide or independently verified performance guarantee. [2]

The software decides whether the hardware matters #

Arm’s public Neural Graphics SDK supports NSS and NFRU on Android AArch64 through Vulkan. Its documentation requires developers to integrate the SDK into a game or engine, enable Vulkan machine-learning extensions where available and use an emulation layer when the platform lacks native support. [4]

The SDK’s samples show that NSS receives lower-resolution color, velocity and depth data before producing an upscaled output. NFRU uses rendered frames and motion information to generate an additional frame during post-processing. That makes the system more than a display-level filter, but it also means the game engine must provide the required data and scheduling. [4]

Arm has announced integrations with Unity China’s Tuanjie Engine and Tencent Games’ MagicDawn technology. The company also identified NetEase’s Where Winds Meet as a planned NSS-enabled title, cited an NSSD demonstration using Tencent’s Arena Breakout: Infinite, and listed NSS integration with Infold Games’ Infinity Nikki. These announcements establish development and demonstration support; they do not show that all of the features are already available in public, shipping versions of those games. [2]

The open SDK is also not a guarantee of equal performance across devices. Arm says its emulation layer is intended for platforms without the relevant native Vulkan extensions, while its documented performance targets apply to supported hardware configurations. [4]

Xiaomi supplies the first device, not a global rollout #

Xiaomi’s 18 Fold is a wide-format foldable with a 7.58-inch inner display and the company’s Xring O3 chip. Xiaomi CEO Lei Jun said the processor contains 24 billion transistors, delivers 60% higher CPU performance than the previous Xring O1 and reduces power consumption by 25% in certain tasks. Those figures are Xiaomi’s claims and concern selected comparisons; they do not establish a corresponding gaming improvement. [3]

The phone starts at 10,999 yuan, or about $1,639 based on the cited launch coverage, with Chinese sales scheduled to begin at 10 a.m. local time on September 10. The sources reviewed for this article do not provide a global launch date. [3]

Arm’s approach enters a market that already has competing mobile rendering tools. Apple’s MetalFX provides temporal and spatial upscaling through the Metal API. Qualcomm offers Snapdragon Game Super Resolution and has announced Adreno Neural Fusion, which combines neural processing, super resolution and frame generation in a newer Adreno pipeline. [5][6]

The distinction is that Arm is offering technology intended for adoption across multiple chip vendors and device makers, while Apple controls both the hardware and software stack on its platforms. Whether Mali G2-Ultra NX becomes a broad Android gaming standard will depend on image quality, latency, thermal behavior and the number of games that ship with the required integration. That remains an open question after the Xiaomi debut.

Companies mentioned #

Further sources #

The stories that matter, in one email. Free — unsubscribe anytime.