Key takeaways
This playbook is a snapshot of where things are right now. It walks through what has been built, what has been tested, and what has been learned along the way. Some of this is ready to pick up and try today, other parts are a bit further out. They give a clearer picture of where things are heading, but they also come with more complexity and more open questions.
This isn’t presented as a finished solution. It’s shared as it is - early, evolving, and grounded in real experiments. Some things worked exactly as expected. Others didn’t, or needed more iteration than anticipated. That’s part of the process.
If there are a few takeaways to highlight, they would be these:
- Arm Neural Technology isn’t a one-size-fits-all upgrade. It’s something you apply where it makes sense.
- NFRU is a good entry point. It’s relatively easy to evaluate and can give you immediate value.
- NSSD is a bigger shift. It opens up new possibilities - especially around lighting - but it requires investment and a willingness to explore new workflows.
And maybe most importantly: the kind of game you’re building matters. Some content fits naturally with these techniques today, and some doesn’t (yet). This is still early in the journey. These are some of the first real attempts at combining ray tracing, dynamic lighting, and neural reconstruction on mobile in a practical way. There isn’t a fixed playbook yet - that’s what’s being built together with the community.
More teams are encouraged to get involved. Try things out, see what works for your game, and share what you learn.
Get involved
Start with Enable Neural Frame Rate Upscaling in Unreal Engine to configure the plugin and evaluate NFRU with your own content. Continue with Analyze Neural Frame Rate Upscaling using Project Moku to examine visual quality, frame pacing, and performance in repeatable test scenes. Arm publishes the Neural Frame Rate Upscaling model on Hugging Face , and the Neural Graphics Development Kit provides the SDK and supporting tools. When you need to customize the model, continue with Train and evaluate Neural Frame Rate Upscaling models using Model Gym .