Xiaomi 18 Fold first to ship Arm's Mali G2-Ultra NX, bringing DLSS-style neural graphics to phones

The Xiaomi 18 Fold goes on sale in mainland China as the first phone to carry Arm's new Mali G2-Ultra NX graphics processor, built into Xiaomi's custom Xring O3 chipset. According to The Verge, the device marks the debut of an AI graphics accelerator that Arm spent five years embedding into its GPU designs, and the company expects the technology to spread well beyond a single Chinese foldable.

DLSS-style rendering comes to mobile

Arm's strategy mirrors the one Nvidia pioneered on the desktop. The new GPU supports a suite of hardware-accelerated techniques Arm brands "neural graphics": AI upscaling that reconstructs a high-resolution image while the GPU renders fewer real pixels, frame generation that synthesises extra frames between rendered ones, and a denoising feature comparable to the ray reconstruction in Nvidia's DLSS 4.5. As The Verge notes, Arm uses the term more narrowly than Nvidia does, focusing on speed and efficiency rather than reimagining how light and shadow behave.

Stack all three together, Neural Super Sampling, Neural Frame Rate Upscaling and Neural Denoising, and Arm claims up to four times the framerate. The report adds two caveats. Games will not feel four times as responsive, because roughly half of the gain comes from generated frames and frame generation typically introduces added latency. And the features arrive in stages: Chris Bergey, Arm's EVP of Edge AI, told The Verge that early titles will mostly use super sampling, with frame generation landing in early 2027 and denoising following soon after.

On conventional throughput, Arm cites an 85 percent performance improvement over the GPU Xiaomi used previously, and a 14 percent gain over its own prior flagship in non-AI gaming. These are the company's own figures.

A two-and-a-half-watt ceiling

The more striking numbers involve power. Bergey told The Verge that an entire phone built on these chips should draw no more than 2.5 watts, with the GPU itself limited to 1.5 watts, enough in his framing for sustained sessions without a hot device or a scramble for a charger. He described playing at 1080p and 60fps for three hours, where comparable experiences previously lasted around 30 minutes. Xiaomi's Pad 9 Pro Max tablet, which pairs the GPU with a larger battery, screen and cooling surface, is being shown running 3.2K resolution at 120fps.

Bergey also frames this as mainstream technology rather than an enthusiast overclocking feature, and says device makers can alternatively run the GPU at lower wattage to extend battery life.

Developers hold the keys

The techniques require per-game integration, though support in engines such as Unreal Engine and Unity eases the work, and Bergey does not expect a universal toggle players can flip for any title. The first three confirmed games are Where Winds Meet, Infinity Nikki and Arena Breakout: Infinite, with more titles to be announced in the coming weeks.

Arm is not alone in pushing PC-style upscaling into mobile-class hardware. As The Verge points out, the developers of the graphically demanding Stellar Blade are targeting 60fps on the Nintendo Switch 2, an Arm-based chip paired with a custom Nvidia GPU, by using DLSS alongside a version of AMD's frame generation.

Beyond one foldable

Bergey says the technology will be purchasable in phones next year, and in larger-screen devices including tablets and Chromebooks. He hinted that some of Google's upcoming Googlebooks may adopt the graphics tech as well. While the G2-Ultra NX itself targets flagship smartphones, Arm says neural graphics will also be available in Premium and Pro GPU configurations, extending it across more mobile segments and price tiers.

Why it matters

Mobile graphics have always been constrained by thermal and battery limits rather than raw silicon capability. Arm's bet is that AI can break that trade-off: render fewer pixels and frames natively, let neural networks fill the gaps, and deliver PC-like image quality and framerates inside a 2.5-watt envelope. If the claims hold, ordinary phones could run visually demanding games at settings once reserved for consoles or laptops. The open questions are equally clear. Gains depend on developers opting in title by title, generated frames carry latency costs, and the performance numbers are Arm's own. The Xiaomi 18 Fold is the first real-world test of whether neural rendering becomes the default path for mobile gaming or another feature that lives and dies by per-game support.