Arm Unveils AI-Native GPU and CPU Tech to Transform Mobile Gaming and AI on 2027 Android Phones

Arm Introduces CSS for Mobile 2 with AI-Native GPU

Arm unveiled its Compute Subsystem (CSS) for Mobile 2 platform at the Arm Everywhere China event in Shanghai, marking a strategic shift toward AI-powered mobile graphics and processing3

3

. The platform centers on two key innovations: the Mali G2-Ultra NX, Arm's first AI-native GPU, and the C2 CPU cores designed specifically for agentic AI workloads1

1

. These technologies target next-gen Android phones expected to launch in 2027, with Xiaomi's 18 Fold already featuring the Mali G2-Ultra NX in its custom Xring O3 chip2

2

.

The Mali G2-Ultra NX represents a fundamental reimagining of mobile gaming graphics. By integrating neural acceleration directly into the GPU pipeline, Arm eliminates the need to transfer graphics data to separate neural processing units, reducing latency and improving power efficiency1

1

. Chris Bergey, executive vice president of Arm's Edge AI Business unit, explains the strategic rationale: "If we can provide an amazing 540p, 30 frames per second experience, and it allows you to use a smaller battery, that allows for a smaller form factor and potentially allows for a lower-cost handheld"1

1

.

Neural Graphics Deliver 4x Performance Boost

The Mali G2-Ultra NX employs three AI-powered techniques collectively called neural graphics. Neural super sampling upscales low-resolution renders from 540p to 1080p, while neural frame rate upscaling generates intermediate frames to double the frame rate from 30fps to 60fps1

1

. Neural super sampling and denoising combines upscaling with artifact removal for graphically demanding scenes featuring ray-traced reflections and complex lighting1

1

.

Arm demonstrated these capabilities using a game co-developed with Sumo Digital, showcasing a fantasy character descending into a cave system illuminated by neon-bright giant mushrooms. The demonstration revealed 4x higher performance efficiency and up to 70% lower external memory traffic compared with conventional graphics rendering1

1

. These improvements translate to a 24% boost in benchmark performance and a 14% improvement in ray tracing1

1

.

The technology operates within strict power constraints. An entire phone with these chipsets will consume no more than 2.5 watts, with only 1.5 watts allocated to the GPU2

2

. Bergey emphasizes the practical impact: "It might be like, hey, you can now play 1080p, 60 frames per second for three hours. In the past, it was 30 minutes"2

2

. The Xiaomi Pad 9 Pro Max tablet, which pairs the GPU with a larger battery and cooling surface, demonstrates even more impressive capabilities, delivering 3.2K resolution at 120 frames per second2

2

.

Desktop-Quality Graphics Within Mobile Power Budgets

The Mali G2-Ultra NX features completely redesigned execution engines supporting larger desktop-class shaders with up to twice the registers per warp—128 instead of 64—and more dynamic allocation5

5

. This architectural change allows advanced graphical shader technologies like Unreal 5's Lumen and Nanite to run without performance drops from reaching out to DRAM5

5

. The GPU also sports Arm's third-generation ray-tracing unit, which introduces a more compact triangle representation using shared edges to reduce data duplication, lowering DRAM traffic by 13%5

5

.

Arm's approach enables the GPU to sustain a 30 frames per second gaming session on content with intricate ray-traced scenery, enabled by a 70% reduction in overall ray tracing workload, translating to a 30% boost in performance3

3

. The reference design promises 18-24% better ray-tracing performance than its predecessor, depending on the benchmark5

5

. Gaming performance increases range from 9% in Honkai Star Rail to 14% in Fortnite when comparing the 14-core Mali G1-Ultra reference platform to the G2-Ultra equivalent5

5

.

C2 CPU Cores Optimized for Agentic AI Workloads

Arm's new C2 CPU cores address the computational demands of agentic AI, which interprets objectives, coordinates applications and models, and executes sequences of actions3

3

. The C2-Ultra core delivers up to 15% higher single-thread and 12% higher multi-thread performance compared to last year's C1 cores, with 1.7x performance across AI models while using 38% less power1

1

. A clock speed boost to 4.45GHz accounts for 8% of the performance improvements, with the remaining 7% coming from core architecture enhancements4

4

.

The C2-Ultra boosts its execution window by 30%, meaning approximately 2,600 instructions are in flight at any time, compared to 2,000 in C1-Ultra4

4

. Arm has enhanced out-of-order execution mechanisms with smarter speculation and faster recovery, enabling earlier detection and execution of code dependencies4

4

. The flagship cluster features two C2-Ultra cores, six efficiency-focused C2-Pro cores, and two Scalable Matrix Extensions (SME2) units, compared with one SME2 unit in last year's platform3

3

.

Doubling the SME2 capability enables up to a 70% speedup on the latest small language models3

3

. For real-world applications, Arm benchmarks show a 40% reduction in speech-to-text latency, 41% faster memory retrieval search latency, and 25% improvement in prompt generation speed for time-to-first-token4

4

. Across a sample agentic workflow combining speech processing, memory retrieval, reasoning, application execution and web browsing, C2-Ultra cores completed the workflow 24% faster than last year's cores1

1

.

Rollout Timeline and Device Availability

Neural graphics capabilities will arrive in phases. Bergey confirms that "the first titles will largely be super sampling," with frame generation coming in early 2027 and denoising "soon after that"2

2

. While neural graphics will be available across Premium and Pro GPU configurations, bringing the technology to a broader range of mobile segments and tiers2

2

, the Mali G2-Ultra NX targets flagship smartphones initially.

The technology isn't limited to Chinese devices. Bergey hints that Google's upcoming Chromebooks might use this graphics tech as well, stating, "You will be able to buy it; you will be able to get it in phones next year, you will be able to get it in other types of larger screen devices, whether that be tablets or Chromebooks"2

2

. The usual cadence of silicon releases suggests this new core is likely headed to next-gen Samsung Exynos and MediaTek Dimensity chipsets for phones around the turn of 2027, potentially including Google's Tensor G74

4

.

To encourage adoption, Arm has published its neural graphics SDK and sample code on GitHub, with support for the cross-platform Vulkan graphics API3

3

. Game developers will need to integrate these features, though participating game engines like Unreal Engine are already preparing support2

2

. The move to 2nm manufacturing processes enables these performance improvements while maintaining or reducing power consumption4

4

, positioning edge devices to handle increasingly complex AI workloads locally rather than relying on cloud processing.