Pull to refresh
Logo
Arm puts neural accelerators inside mobile GPU with CSS for Mobile 2

Arm puts neural accelerators inside mobile GPU with CSS for Mobile 2

New Capabilities

AI-native compute platform pairs neural-accelerated Mali GPU with C2 CPUs for agentic AI

Today: Arm launches CSS for Mobile 2

Overview

Updated 48 minutes ago

Arm's new mobile platform puts neural accelerators inside the GPU for the first time in its Mali graphics line. The Mali G2-Ultra NX pairs with a new C2 CPU cluster that doubles its AI processing units to keep on-device AI agents responsive.

Arm says the combination runs neural graphics at up to four times the efficiency of standard rendering and cuts power by up to 38 percent at equal performance. If phone makers adopt it, flagship Android devices could get console-grade graphics and always-on AI on one chip.

Why it matters

On-device neural upscaling could bring console-quality mobile games and fast AI agents to phones without draining batteries.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

1.7x
AI performance gain over prior C1-Ultra CPU
C2-Ultra with doubled SME2 units delivers the gain.
70%
Small-language-model speedup
Two SME2 units per cluster accelerate on-device language models.
38%
Power reduction at equal performance
Claimed C2-Ultra efficiency versus the previous generation.
4x
Neural graphics efficiency gain
Neural Frame Rate Upscaling versus native rendering.
24%
Faster end-to-end agentic workflow
C2-Ultra completes agentic tasks 24% faster than prior gen.
120 FPS
Sustained mobile gaming frame rate
Neural frame rate upscaling enables longer sessions at high framerates.

Voices

Curated perspectives — historical figures and your fellow readers.

Ever wondered what historical figures would say about today's headlines?

Sign up to generate historical perspectives on this story.

Play

Exploring all sides of a story is often best achieved with Play.

Most of these play right now — no account needed. Sign up to save scores, keep a streak, and unlock Debate and Predict. Log in Sign Up
Predict 3 ways this could play out. Back the one you believe — contrarian picks score more when a scenario has a resolution date. Log in to play

People Involved

Organizations Involved

Timeline

May 2024 September 2026

2 events Latest: Today
  1. Arm launches CSS for Mobile 2

    Today Product Launch

    Arm launches CSS for Mobile 2 with neural-accelerated Mali GPU and C2 CPU cluster.

  2. Arm unveils first Compute Subsystem

    Product Launch

    Arm unveils its first Compute Subsystem, bundling CPU, GPU, and system IP into one platform.

Historical Context

2 moments from history that rhyme with this story — and how they unfolded.

September 2018

NVIDIA DLSS launches (2018)

NVIDIA shipped Deep Learning Super Sampling, using tensor cores to upscale lower-resolution frames in real time and cut the rendering load. It debuted as a flagship feature of its GeForce RTX GPUs.

Then

Neural upscaling became a headline PC graphics feature and reshaped player expectations for image quality.

Now

Rivals AMD and Intel shipped their own versions (FSR and XeSS); neural upscaling is now a standard PC graphics feature.

Why this matters now

Arm's Neural Super Sampling and Neural Frame Rate Upscaling are the same play moved to phones, where power and heat budgets are far tighter.

November 2020

Apple Silicon transition (2020)

Apple replaced Intel processors in its Macs with its own integrated chips combining CPU, GPU, and a Neural Engine on one die. The first M1 chip doubled battery life in entry-level laptops.

Then

The transition completed by 2023 and made integration the centerpiece of Apple's computing pitch.

Now

It proved tightly coupled, integrated platforms outperform modular component ecosystems for consumer devices.

Why this matters now

Arm's CSS bundles CPU, GPU, and system IP the same way, betting total integration beats stitching together separately licensed cores.

Sources

(5)