Skip to content

Questions on GPUs, ARM translation, and multi-arch Android setups w Redroid #65

Description

@maddox05

I’ve been working on an Android cloud-gaming platform and have been using ReDroid extensively:

https://github.com/maddox05/android-cloud-gaming

I’ve been running into limitations getting ReDroid to work outside of ARM hosts, and I was hoping to get your perspective given your experience with redroid-script, libndk / Houdini experiments, and Android container tooling in general.


Background / What I’m Trying to Do

  • Use case: Run many Android “phones” concurrently for cloud gaming
  • Android versions: Mostly 12 and 13
  • Apps: Primarily ARM-only games
  • Hosts available:
    • ARM machines → works reliably, but no GPUs (cant get OSS nvidia drivers to work on g5g amazon instance, if you can help here LMK!!)
    • x86 machines (AMD CPUs + AMD GPUs) → where I’m currently blocked

My long-term goal is to run the same Android containers across mixed infrastructure (ARM + x86) while still having GPU acceleration available.


Questions I’d Love Your Insight On

1. GPUs with ReDroid

  • Have you personally had GPU acceleration working with ReDroid?
  • If so:
    • What host architecture (ARM vs x86)?
    • What GPU vendor (AMD / NVIDIA / integrated)?
    • What Android versions?

I’m especially curious whether AMD GPUs on x86 are fundamentally problematic, or if it’s more about current tooling maturity.


2. ARM translation layers (libndk / Houdini)

  • Were you ever able to get ARM translation working reliably with ReDroid?
  • If yes:
    • Which Android versions (11 vs 12/13)?
    • What setup or approach worked?
  • If not:
    • Do you see this as an AOSP 12+ limitation, or more of a container / native-bridge issue?

I noticed prior discussion around libndk / Houdini being difficult or broken on newer Android versions, so I’d love to know what your real-world conclusion was.


3. Single Docker image across x86 + ARM

  • In your opinion, is it realistically possible to:
    • Use one Android Docker image for both ARM and x86 hosts?
  • Or is the practical reality today:
    • Separate images per architecture?
    • Or ARM-only if you want full ARM app compatibility?

If you’ve tried multi-arch images (or decided they’re not worth it), I’d love to understand why.


Why I’m Asking

For cloud gaming, ARM-only apps are the norm, and GPU access is critical. I want to make sure I’m not fighting fundamental limitations of the Android runtime or container model before committing to a long-term architecture.

Even high-level guidance or confirmations like “this isn’t viable today” would be extremely helpful.

Thanks a ton for your work on redroid-script and for any insight you’re willing to share 🙏
Maddox

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions