Arm Community
Arm Community
  • Site
  • User
  • Site
  • Search
  • User
Arm Community blogs
Arm Community blogs
Mobile, Graphics, and Gaming blog Start experimenting with Neural Super Sampling for mobile graphics today
  • Blogs
  • Mentions
  • Sub-Groups
  • Tags
  • Jump...
  • Cancel
More blogs in Arm Community blogs
  • AI blog

  • Announcements

  • Architectures and Processors blog

  • Automotive blog

  • Embedded and Microcontrollers blog

  • Internet of Things (IoT) blog

  • Laptops and Desktops blog

  • Mobile, Graphics, and Gaming blog

  • Operating Systems blog

  • Servers and Cloud Computing blog

  • SoC Design and Simulation blog

  • Tools, Software and IDEs blog

Tags
  • vulkan
  • Neural Super Sampling (NSS)
  • Artificial Intelligence (AI)
  • Arm Accuracy Super Resolution
  • Mali GPUs
  • Mobile (Mobile Gaming and Mobile AI)
Actions
  • RSS
  • More
  • Cancel
Related blog posts
Related forum threads

Start experimenting with Neural Super Sampling for mobile graphics today

Sergio Alapont Granero
Sergio Alapont Granero
August 12, 2025
6 minute read time.

Mobile game developers around the world face increasing pressure to meet user expectations for sharper visuals, smoother gameplay, and longer battery life. Balancing these goals on constrained mobile devices often means making trade-offs. Traditional upscaling methods offer limited flexibility. Real-time AI rendering remains complex, power-hungry, or hardware dependent.

Neural Super Sampling (NSS) directly addresses these challenges. NSS is a next-generation AI-powered upscaling solution. It is the first use case of Arm neural graphics technology, as announced 12 August 2025 at SIGGRAPH 2025.

NSS brings desktop-class visual fidelity to mobile gaming using dedicated neural accelerators. It will be available in future Arm GPUs through Arm neural technology. This is a major innovation in mobile graphics. It enables real-time super-resolution that combines temporal coherence with perceptual quality far beyond traditional techniques.

Devices with this technology will ship in late 2026. Developers can begin exploring NSS today using the Arm Neural Graphics Software Development Kit (SDK). This SDK includes an Unreal Engine plugin and a Vulkan emulation layer. These tools help developers experiment with neural upscaling workflows, game integration prototyping, and preparing for Vulkan ML-based rendering pipelines, all before the first compatible hardware arrives.

Real-time upscaling with neural performance

Traditional upscaling techniques rely on handcrafted heuristics or limited spatial data. NSS uses neural inference to resolve detail, motion, and lighting changes with higher fidelity.

In a typical pipeline, NSS involves three steps:

  1. Preprocessing: Prepares input data (color, motion vectors, depth) into tensors
  2. Neural inference: Performed on the Neural Accelerator using Vulkan ML extensions
  3. Post-processing: Reconstructs the final image using model output.

This allows NSS to scale performance gradually. NSS uses fewer resources in higher-resolution passes, making it ideal for the hardware constraints of mobile devices. For more information about how NSS works, read this blog post:

How Arm Neural Sampling works

Built on Arm ASR, elevated by machine learning

NSS is designed for real-time performance on future mobile devices with Arm neural technology. However, latency depends on implementation factors such as GPU configuration, resolution, and use case. In our Enchanted Castle demo video below, NSS reduced GPU workload by 50 percent. The model rendered at 540p and upscaled to 1080p in just 4ms in sustained performance setup.

Developers interested in exploring model performance and image quality comparisons can find detailed results in our technical whitepaper:

Learn more about Neural Super Sampling

NSS builds on the success of Arm Accuracy Super Resolution (Arm ASR). ASR is our open source solution for temporal upscaling across a wide range of mobile devices. With NSS, our goal is to provide developers with a state-of-the-art neural upscaling solution for mobile devices equipped with neural accelerators. Arm ASR also remains a highly optimized alternative for devices without those capabilities.

Internal testing shows NSS reaching DLSS2-level fidelity using a smaller, mobile-optimized model. This is supported by metrics such as PSNR (Peak Signal-to-Noise Ratio), SSIM (Structural Similarity Index), and STRRED (Spatio-Temporal Reduced Reference Entropic Differencing). These metrics were collected during side-by-side comparisons with Arm ASR, DLSS2 (NVIDIA’s Deep Learning Super Sampling 2.0), and FSR2 (AMD’s FidelityFX Super Resolution 2.0).

Arm ASR vs NSS: What you need to know

Feature Arm ASR NSS
Method Shader-based, spatio-temporal AI-based with neural inference
Hardware requirements Runs on Arm GPUs Requires 2026 Arm GPUs with Neural Accelerator
Performance Low latency, optimized for a wide range Designed for real-time execution with future neural hardware
Image quality Strong baseline and tuneable Higher perceptual quality, DLSS2-comparable
Customization Shader parameters Retrainable ML models (open format)
Use case Broad deployment, especially on lower-end devices Premium experiences, advanced visuals, energy savings

Getting started with NSS

When a new piece of technology lands, we believe that developers need to have an environment where they can experiment and explore their creativity. This is why we are launching the Arm Neural Graphics Software Development Kit (SDK). This SDK gives developers everything they need to start experimenting, testing and preparing for the next wave of mobile graphics. The SDK includes tools, sample code, and resources for mobile graphics development. It also includes:

  • A Vulkan ML emulation layer
  • Unreal Engine 5 plugin
  • Pretrained open-source models available via Hugging Face and GitHub
  • Comprehensive documentation and learning paths to help teams get started

The Arm Neural Graphics Development Kit also offers early access to ML extensions for Vulkan, VK_ARM_tensors and VK_ARM_data_graph. These extensions allow developers to use tensor resources and build inference pipelines directly into the render graph. This integration helps achieve the latency and efficiency targets required for real-time applications on mobile devices.

Together, these resources offer a practical head-start on integrating neural graphics before the first hardware arrives. Teams can prototype NSS or explore new use cases using a desktop environment that mirrors next-generation Arm GPU Architecture. The SDK enables deep integration and performance testing ahead of hardware availability.

Integration and testing in Unreal Engine

The Unreal Engine 5 plugin provides integration of NSS into Unreal Engine. Developers can quickly add the upscaler to their projects and preview how their games could look on next generation mobile devices. The plugin also connects to Unreal’s Neural Network Engine (NNE) framework. It manages ML resources and dispatching inference jobs in sync with the render graph.

These capabilities allow developers to integrate neural upscaling directly into existing rendering flows with minimal overhead. This helps preserve frame timing and maintain real-time responsiveness, even during AI-powered post-processing.

It is worth noting that the ML inference running in the Unreal Engine 5 plugin uses software emulation. Performance may not fully reflect what to expect on actual silicon. The primary goal of the plugin is to enable experimentation and integration.

The Vulkan ML emulation layer

2026 mobile devices with next generation Arm GPUs featuring Arm neural technology will support Vulkan ML extensions natively. To prepare, we need a way to expose new functionalities in a way that a developer writing code using these extensions could consume.

This is why we are providing a Vulkan ML emulation layer that exposes the software emulation for the required extensions. Developers can enable this layer and use the extended Vulkan headers to run content using Vulkan ML extensions. One great example is NNERuntimeRDGVulkanML, available through the Unreal Engine 5 plugin. It extends the NNE framework with a native backend using the Vulkan ML extensions.

A new era of neural graphics on mobile begins

NSS is one of the first neural graphics solutions designed for mobile gaming. It combines high-quality upscaling with a developer-ready SDK. It delivers high-quality results, low latency, and power efficiency through an open, developer-ready infrastructure. NSS is more than a neural upscaler. It gives developers full control over how they want to use the technology.

The Arm Neural Graphics Development Kit provides developers with the tools and capabilities to adapt and optimize NSS for their own unique content and artistic vision. Ensuring developers are ready when the first mobile devices built on Arm GPUs with Arm neural technology land in 2026.

Now is the time to experiment, test and prepare for the next wave of mobile graphics. Visit the NSS Developer Hub to explore the Arm Neural Graphics Development Kit and try out the pretrained models.

The tools are ready, so dive in and start exploring the future!

Get started (NSS Developer Hub)

Anonymous
Mobile, Graphics, and Gaming blog
  • Unlock the power of SVE and SME with SIMD Loops

    Vidya Praveen
    Vidya Praveen
    SIMD Loops is an open-source project designed to help developers learn SVE and SME through hands-on experimentation. It offers a clear, practical pathway to mastering Arm’s most advanced SIMD technologies…
    • September 19, 2025
  • What is Arm Performance Studio?

    Jai Schrem
    Jai Schrem
    Arm Performance Studio gives developers free tools to analyze performance, debug graphics, and optimize apps on Arm platforms.
    • August 27, 2025
  • How Neural Super Sampling works: Architecture, training, and inference

    Liam O'Neil
    Liam O'Neil
    A deep dive into a practical, ML-powered approach to temporal super sampling.
    • August 12, 2025