Arm C2-Ultra Claims 15% Faster Single-Core and 38% Less Power for Phone Chips
Arm introduced its next mobile platform on September 8, and the headline figure is a CPU that claims 15 percent higher single-thread performance while drawing 38 percent less power at the same speed. The package, called CSS for Mobile 2, pairs the new C2-Ultra and C2-Pro CPU cores with the Mali G2-Ultra NX, the first Mali GPU to put neural accelerators inside its shader cores. These are the designs that end up inside next year's Android flagship chips, so the numbers matter well before any phone is announced.
What the C2-Ultra actually gains
Measured against last year's C1-Ultra, Arm quotes up to 15 percent more single-thread performance, 12 percent faster app launches, and up to 1.7x the AI performance. The efficiency claim is the more interesting one: 38 percent less power to hit the same performance level, which is the number that decides whether a phone throttles after ten minutes of sustained load.
The AI gain comes mostly from doubling the SME2 units in the cluster. Arm says that doubling produces a 70 percent speedup on current small language models — the compact models that run on the phone itself rather than in a data centre, for things like summarising, translation and on-device assistants.
On paper the C2-Ultra is an Armv9.3-A core with selected Armv9.4-A features, SME2, SVE2 and RAS support. A cluster can scale to 14 cores, with 2–3MB of L2 per core and an optional shared L3 that runs from 256KB up to 32MB. The C2-Pro sits below it as the mid-tier core, and Arm has published far less about it so far.
The GPU is the bigger structural change
The Mali G2-Ultra NX is the part Arm is calling AI-native, and the description is more literal than most marketing. Instead of bolting a separate NPU onto the graphics block, Arm has put neural accelerators directly inside the shader cores, so neural rendering runs in the same pipeline as ordinary graphics work. Arm describes the instruction set overhaul as the largest in seven Mali generations.
What that buys, according to Arm: up to 4x the performance per watt on neural graphics workloads such as Neural Super Sampling, frame-rate upscaling and combined super-sampling with denoising. Ray tracing gets a third-generation ray tracing unit with hardware occlusion mapping, which Arm says cuts DRAM traffic on ray-traced scenes by 70 percent. The GPU scales from 10 to 24 cores and supports Vulkan 1.4, OpenGL ES 3.2 and OpenCL 3.0.
Two different gaming numbers, and what each one means
Coverage of the launch has quoted two separate GPU figures, and they are not the same claim. Arm's own comparison against the Mali G1-Ultra is 14 percent higher performance on existing game content — games shipping today, with no changes made for the new hardware. The 24 percent figure circulating alongside it is a benchmark result, not a game result.
So the practical expectation for a game you already have installed is closer to the 14 percent number. The 4x figure applies only to neural rendering, and only in titles built to use it. Arm named NetEase's Where Winds Meet, Tencent Games, Unity China and Infold Games' Infinity Nikki among the studios working with its neural graphics tools, so the first real demonstrations will be narrow.
Which phones get it, and when
Arm did not name a launch chip, a launch partner or a date. That is normal for an IP announcement — Arm sells designs to chipmakers, who then take roughly a year to ship silicon, so devices built on this generation realistically arrive through 2027 rather than this autumn.
It also will not reach every flagship equally. Qualcomm designs its own CPU cores under an architecture licence, so its phone chips do not use Arm's Cortex-class cores or Mali graphics — its next GPU is the separately announced Adreno Neural Fusion design, and its immediate lineup is the pair of Snapdragon 8 Elite Gen 6 chips due this month. Arm's cores and Mali GPUs reach phones mainly through MediaTek's Dimensity line and Samsung's Exynos, which is also why MediaTek's role in Android hardware keeps widening — it now supplies the modem in Google's Pixel 11.
What to watch next
The useful checkpoint is the first Dimensity or Exynos flagship announcement that names the C2-Ultra or Mali G2-Ultra NX, because that is when Arm's percentages get tested against a real phone with a real thermal budget. Until then, treat the 38 percent power figure as the one worth remembering: sustained performance, not peak benchmark scores, is where recent mobile chips have struggled most.
Frequently Asked Questions
What is the Arm C2-Ultra?
The C2-Ultra is Arm's new highest-performance mobile CPU core, announced on September 8, 2026 as part of the CSS for Mobile 2 platform. Arm claims up to 15 percent higher single-thread performance and 38 percent lower power at matched performance versus the previous C1-Ultra. It is a design Arm licenses to chipmakers, not a chip you can buy directly.
Which phones will use the C2-Ultra and Mali G2-Ultra NX?
Arm did not name any device or chip at announcement. Its CPU cores and Mali GPUs reach phones mainly through MediaTek's Dimensity chips and Samsung's Exynos line, and silicon built on a new Arm generation usually ships around a year after the design is announced. That points to phones in 2027 rather than this year.
How much faster are games on the Mali G2-Ultra NX?
Arm quotes up to 14 percent higher performance on existing game content compared with the Mali G1-Ultra. A separate 24 percent figure that has circulated is a benchmark result rather than a real-game one. The much larger 4x figure applies only to neural rendering workloads in games specifically built to use them.
Does Snapdragon use Arm's new C2-Ultra cores?
No. Qualcomm holds an architecture licence and designs its own CPU cores, and it uses its own Adreno graphics rather than Mali. Arm's new cores and GPU will not appear in Snapdragon chips, so this announcement affects MediaTek and Samsung silicon first.
What are neural accelerators in a GPU?
They are dedicated hardware blocks for running small neural networks, and in the Mali G2-Ultra NX they sit inside the shader cores instead of in a separate processor. That lets techniques like neural super sampling and AI upscaling run in the same pipeline as normal graphics, which Arm says delivers up to four times the performance per watt on those tasks.
