Expected default graphics acceleration for RK3588?

[gpupoor]

I understand that panthor is not supported for wayland but in that case I would expect libmali to fill in... but I ended up with lvmpipe.

 

The overlay along with another recent post led me to believe that panthor was the expected default and libmali was an artifact, but I'm now questioning that after finding a thread identifying use cases where libmali can be preferred and a comment suggesting panthor wouldn't be available until 6.13.

As a result, I am now questioning if there is a reason that software acceleration is the default and the user is expected to determine which path they want to go?

I know support for the hardware has been slow going so I understand that it's not well supported yet, but hoping to get a better handle on expectations.

 

Is there an expected default for a Rockchip RK3588 SoC with the Mali‑G610 GPU? Does it vary between 6.1/6.12 and the desktop/server builds?

Edited November 24 by gpupoor
clarificatin

[usual user]

1 hour ago, gpupoor said:

I understand that panthor is not supported for wayland

I have nothing to complain about.

glmark2-wayland-odroid-m2.log

vkmark-wayland-odroid-m2.log

[KhanhDTP]

@gpupoor

 

You want to set up GPU Acceleration, right? Have you read this post yet?

 

[Igor]

8 hours ago, gpupoor said:

Does it vary between 6.1/6.12 and the desktop/server builds?

Yes. It is only enabled / working with 6.1 and desktop (Gnome) builds. For mainline based, perhaps 6.18.y ...

[gpupoor]

Thank you @KhanhDTP. I think I have a pretty good handle on how (and your post on that thread was in particular very helpful, thank you). 

It looks like a lot of people here are desktop optimized. My plan is headless, so expectation is 25.11.1 / 6.1 and libMali route but with RKNN (XFCE only for occasional desktop, Gnome is too hungry, XFCE + X11 + software rendering is fine on rare occasions I'd need it). 

I think the question I was trying to ask is more along the lines of understanding what to expect for the default, and if it made sense to enable one of the drivers by default, but I'm realizing there are proprietary dependencies and enough obscure use cases that this gets more challenging.

Thanks to everyone for all of the thoughts, hard to imagine navigating all of this without your contributions.

[gpupoor]

Edit: I'm not sure how to format nicely here... so I updated the related PR. That said, this probably isn't the best reference location for this. In any case, posting in case helpful for anyone else. If it's useful for the project to better community (and potentially automate) some of this I'm happy to get involved, but here's what I've got for now:
 

There are fundamentally two driver paths for RK3588:

1. libMali path - Enables NPU for ML inference, basic GPU only
2. Panthor path - Enables full GPU acceleration, no NPU

Within each path, you can choose headless or desktop (XFCE/GNOME), but:

• libMali + desktop = slow rendering (software only)
• Panthor + desktop = smooth rendering (hardware accelerated)

Key insight: The NPU and GPU hardware are separate, but the software drivers conflict. You must choose which accelerator matters more for your use case.

Additional constraint:

Kernel matters - NPU only works on 6.1 vendor kernel. Mainline 6.12 has no NPU support but best GPU drivers.
 

This creates a matrix of ~6 viable configurations, each optimized for different workloads (production ML, ML development, desktop, latest kernel, etc).

More detailed breakdown: 
https://github.com/armbian/build/pull/8979#issuecomment-3587135022
 

Edited November 27 by gpupoor

[KhanhDTP]

@gpupoor

 

Quote

More detailed breakdown: 
https://github.com/armbian/build/pull/8979#issuecomment-3587135022

 

Why can't I see this option?

. Kernel: 6.1 vendor

. Driver: libMali

. Panthor Overlay: OFF

. Desktop: GNOME

. Display Server: Wayland

. NPU (RKNN): not tested (I am not using NPU)

. GPU Accel: Smoothest

. Desktop Perf: Smoothest

 

Noted: must install WSI package

https://github.com/ginkage/vulkan-wsi-layer

[gpupoor]

Thanks @KhanhDTP, this is great. I tried to stick with approaches that have made their way into official channels, but I'm happy to include because it's very cool to have a full working path identified (B2).

I think I understood your setup but please let me know if I missed anything.

Again, not sure where/how the best place to keep track of this is so in the absence of anything better I'll use the PR, but happy to migrate to another location.

[usual user]

10 hours ago, gpupoor said:

Kernel matters - NPU only works on 6.1 vendor kernel. Mainline 6.12 has no NPU support but best GPU drivers.

I have nothing to complain about and the VPUs are also working.

classification-3.11.log

fluster-run-odroid-m2-summary

[gpupoor]

Making some assumptions... you're 6.18 + Mesa 25.3 + Teflon delegate for the TensorFlow Lite tests?

[usual user]

1 hour ago, gpupoor said:

Making some assumptions... you're 6.18 + Mesa 25.3 + Teflon delegate for the TensorFlow Lite tests?

Right, I always use the same system with all my devices. Makes comparisons more meaningful. Only the loaded DT is the only difference.

[gpupoor]

Thanks for confirming. That makes 6.18 look very compelling.

I have some ideas for how I'd like to present this and I'm finding myself wanting to flesh out my overview... If anyone has preferences for a better place than my PR comment to publish please let me know.

 

[gpupoor]

Still working through things, but FYI I was able to get 6.18 going on Ubuntu with some small effort. GPU needs work but still beats CPU, NPU looks great.