Arm has introduced a new mobile compute platform, and for once the follow-through arrived almost immediately. Just over a week after the announcement, a major chipmaker had already built and launched a chip on top of it.
On September 8, 2026, Arm announced CSS for Mobile 2, a compute platform combining a new GPU, a new CPU cluster, and the system-level plumbing to connect them, all aimed at the twin demands of agentic AI and AI-enhanced graphics on smartphones. Within days, MediaTek launched its Dimensity 9600 Pro chipset built on the platform’s CPU and GPU designs — giving the announcement something rarer than a roadmap: an actual chip.
That combination of a detailed architectural pitch and a fast, verifiable adoption is what makes this moment worth examining closely, along with what still isn’t settled: exactly which phones will carry this hardware, and when.
Table of Contents
What Arm Announced
CSS for Mobile 2 is Arm’s newest “Compute Subsystem” for smartphones — a bundled reference design that silicon partners like MediaTek license and adapt into their own chips, rather than a chip itself. According to Arm’s official announcement, the platform brings together the new Arm C2 CPU cluster, the Arm Mali G2-Ultra NX GPU, a system interconnect called SI L2, physical implementation work, and a supporting software stack.
Arm’s pitch, in the company’s own words, is that agentic AI — AI that manages tasks and coordinates apps on a user’s behalf, rather than just answering questions — needs a computing platform where the CPU, GPU, and system components are designed together, not treated as separate parts. Whether that framing holds up depends on how developers actually use it, but it is Arm’s stated rationale for bundling the CPU and GPU work into a single platform rather than releasing them separately.
Platform at a glance:
- Arm C2 CPU cluster (C2-Ultra + C2-Pro cores, two SME2 units)
- Arm Mali G2-Ultra NX GPU (neural accelerators + new execution engine + ray tracing unit)
- SI L2 system interconnect
- Developer software stack (KleidiAI, Neural Graphics Development Kit, AI Portal)
[Production note: insert platform-composition diagram graphic here per Phase 3 visual recommendations.]
Inside the Mali G2-Ultra NX GPU
The GPU is the more novel piece. Arm describes the Mali G2-Ultra NX as the first Mali GPU with neural accelerators built directly into its shader cores, rather than as a separate chip block. That integration matters because it lets neural processing and traditional graphics rendering share the same memory and control systems, which Arm says reduces the data movement that normally costs power and time.
Practically, this shows up as three named features: Neural Super Sampling (NSS), which reconstructs a higher-resolution image from a lower-resolution render; Neural Frame Rate Upscaling (NFRU), which generates intermediate frames to smooth out gameplay; and Neural Super Sampling and Denoising (NSSD), which combines upscaling with denoising for ray-traced scenes.
The GPU also gets what Arm calls its biggest instruction-set upgrade in seven generations, plus a third-generation ray tracing unit with hardware support for handling complex transparent geometry like foliage and layered fabric.
On performance, Arm reports — using its own internal testing — up to 4x higher performance-per-watt for neural graphics workloads compared with the previous-generation Mali G1-Ultra, along with a 24% uplift in general benchmark performance and 14% higher performance in existing (non-AI) games from architecture alone. The company also claims up to 13% lower memory traffic on ray-tracing workloads. These are Arm’s own figures; no independent lab has yet reproduced them.
| Metric (vs. Mali G1-Ultra) | Arm’s Claimed Improvement |
|---|---|
| Neural graphics performance-per-watt | Up to 4x |
| General benchmark performance | Up to 24% |
| Non-AI gaming performance | Up to 14% |
| Ray-tracing memory traffic | Up to 13% reduction |
| Sustained gameplay from combined architecture + NFRU | Up to 120 FPS |
Source: Arm, company-reported figures. Note: the 120 FPS figure reflects the combined effect of architectural improvements and NFRU, not a standalone generational comparison.
Inside the C2 CPU Cluster
If the GPU is about visuals, the CPU cluster is about coordination. Arm frames the CPU as the “orchestration engine” for agentic AI — the part of the chip that keeps track of what an AI assistant is doing, schedules work, and moves tasks between the CPU, GPU, and other accelerators as needed.
The new C2 cluster pairs Arm’s fastest mobile CPU core, C2-Ultra, with efficiency-focused C2-Pro cores, and doubles the number of SME2 units — a matrix-math extension used to accelerate on-device AI — compared with the prior generation. Arm says this combination is intended to keep AI features like speech processing, translation, and lightweight model inference responsive without leaning entirely on separate AI accelerators.
Arm’s reported gains, compared with the previous C1-Ultra generation, include up to 1.7x higher performance on current AI models, up to 15% higher single-thread performance, 15% faster web browsing, 12% faster app launches, and 12% higher multi-thread performance. In a representative end-to-end agentic workflow that Arm designed — spanning speech recognition, memory retrieval, reasoning, and web browsing — the company reports the new cluster finishing 24% faster than before.
As with the GPU figures, these are Arm’s own numbers, generated under its own test conditions. No independent benchmarking of the C2 cluster has been published.
First Silicon: MediaTek’s Dimensity 9600 Pro
The clearest evidence that CSS for Mobile 2 is more than a slide deck is MediaTek’s Dimensity 9600 Pro, a flagship smartphone chip built on TSMC’s 2-nanometer process. It uses two Arm C2-Ultra cores alongside six C2-Pro cores, paired with the Mali-G2 Ultra NX GPU — independently confirmed by multiple outlets including Notebookcheck, 9to5Google, and Guru3D, and further corroborated when Arm’s own social media account publicly congratulated MediaTek on the launch, specifically naming the C2 cluster and Mali G2-Ultra NX GPU.
MediaTek’s own reported performance claims for the Dimensity 9600 Pro include 17% higher single-core and 15% higher multi-core CPU performance compared with its previous Dimensity 9500 chip, alongside up to 61% lower power consumption under peak multi-core workloads. On the graphics side, MediaTek claims up to 27% higher peak GPU performance, 24% lower power draw at peak performance, and 18% faster ray tracing, with support for game frame rates up to 185 frames per second.
It’s worth being precise about what this confirms and what it doesn’t. It confirms that a major silicon vendor has built and announced a real chip using Arm’s new architecture, with independently reported specifications broadly consistent with what Arm described. It does not confirm which phones will use that chip or when they will go on sale — that information has not been officially released by any device maker.
| Metric (Dimensity 9600 Pro vs. Dimensity 9500) | MediaTek’s Claimed Improvement |
|---|---|
| Single-core CPU performance | 17% higher |
| Multi-core CPU performance | 15% higher |
| Peak multi-core power consumption | Up to 61% lower |
| GPU peak performance | 27% higher |
| GPU power at peak performance | 24% lower |
| Ray tracing performance | 18% higher |
Source: MediaTek, company-reported figures
Software and Developer Ecosystem
New hardware capabilities only matter if developers can use them, and Arm has been building toward this for a while. The company points to its KleidiAI software libraries, a Neural Graphics Development Kit, and a newer “AI Portal” meant to give developers a single place to find optimized AI models and code for Arm-based devices.
On the game-development side, Arm names several concrete partners already integrating its neural graphics technology: Tencent Games Central Tech is bringing it into its MagicDawn engine, and Unity China is integrating it natively into its Tuanjie Engine. NetEase has said its game Where Winds Meet plans to ship an NSS-enabled version this year, and Infold Games is integrating NSS into Infinity Nikki. Arm also built a technology demo called Neural Dawn with developer Sumo Digital to showcase the graphics features in a production-style game pipeline.
Separately, Arm notes that its SME2 CPU extension — which predates this specific platform — already has traction with companies including Alipay, Google AI Edge Gallery, OPPO, and vivo. That’s a real and useful data point about SME2’s broader momentum, but it describes adoption of a CPU feature in general, not confirmed adoption of CSS for Mobile 2 specifically. vivo is quoted directly by Arm committing to bring Arm’s neural graphics technology to its flagship phones “built on Arm’s latest compute platform” — a meaningful signal of intent, but a company statement rather than a confirmed device specification.
What’s Still Unconfirmed
Here’s where the story gets more cautious. No phone maker has officially announced an on-sale date for a device using the Dimensity 9600 Pro or the Mali G2-Ultra NX GPU. Industry chatter and unofficial leaks point to vivo’s X500 series and Oppo’s Find X10 series as likely early candidates, but those reports trace back to tipsters and rumor-tracking outlets rather than official statements from vivo or Oppo. Readers should treat any specific phone name or launch date circulating online as speculation until the manufacturers confirm it themselves.
It’s also not publicly known whether other major silicon vendors — including Qualcomm and Samsung, which design their own CPU cores rather than licensing Arm’s — have any plans to use this platform. Their absence from Arm’s announcement isn’t evidence of exclusion; it simply hasn’t been addressed either way.
Conclusion
What the evidence supports right now is fairly clean: Arm has announced a real, technically detailed mobile compute platform, and at least one major chipmaker has already shipped silicon built on its core components, with independent reporting broadly matching Arm’s own descriptions of the architecture. What remains open is everything downstream of the chip — which specific phones will use it, when they’ll arrive, and how the performance claims hold up once independent reviewers get hardware in hand. For now, this is a platform with confirmed silicon behind it, not yet a platform with confirmed products in stores. That gap is worth watching as vivo, Oppo, and other device makers make their own announcements in the coming months.
FAQ
What is Arm CSS for Mobile 2? It’s a compute platform that Arm licenses to chipmakers, combining a new CPU cluster (C2), a new GPU (Mali G2-Ultra NX), a system interconnect, and supporting software, aimed at agentic AI and AI-enhanced graphics on smartphones. It was announced by Arm on September 8, 2026.
What makes the Mali G2-Ultra NX GPU different from previous Mali GPUs? Arm says it’s the first Mali GPU with neural accelerators built directly into its shader cores rather than as a separate component, allowing neural graphics processing to share memory and control systems with traditional rendering.
What is the Arm C2 CPU cluster and what is SME2? The C2 cluster combines Arm’s high-performance C2-Ultra cores with efficiency-focused C2-Pro cores and doubles the number of SME2 units versus the prior generation. SME2 is a CPU extension for accelerating matrix-heavy AI computations directly on the processor.
Which chip is the first to use Arm CSS for Mobile 2? MediaTek’s Dimensity 9600 Pro, a 2-nanometer flagship smartphone chip, is the first publicly confirmed chip built on the platform’s C2 CPU cluster and Mali G2-Ultra NX GPU.
What are Neural Super Sampling (NSS), Neural Frame Rate Upscaling (NFRU), and NSSD? They’re neural-graphics features on the Mali G2-Ultra NX GPU. NSS reconstructs higher-resolution images from lower-resolution renders; NFRU generates intermediate frames to raise frame rates; NSSD combines upscaling with denoising for ray-traced scenes.
Which phones will use the MediaTek Dimensity 9600 Pro? No phone maker has officially confirmed this. Industry rumors point to vivo’s X500 series and Oppo’s Find X10 series as likely candidates, but these come from unofficial leaks, not confirmed announcements.
Are Arm’s and MediaTek’s performance claims independently verified? No. All of the performance figures cited for both the CSS for Mobile 2 platform and the Dimensity 9600 Pro come from the companies’ own internal testing. No independent benchmarking results have been published yet.