4K Gaming on Raspberry Pi [video] | Hacker News Reader

4K Gaming on Raspberry Pi [video]

youtube.com

84 pointsmdp202119 comments

Re: high performance ARM desktops. Found this[1] the other day:

> Radxa has something coming soon with many core ARM v9, up to 64GB LPDDR5, 45Tops NPU, 5Gbps ethernets and PCIe 4.0 at your price range. Stay tuned.

"Your price range" is 300-500 USD. Interesting ARM days ahead...

[1] https://old.reddit.com/r/SBCs/comments/1ft0o4i/beyond_the_sb...

Blog post from Jeff Geerling about the setup: https://www.jeffgeerling.com/blog/2024/use-external-gpu-on-r...

Didn’t he do one of these a few years ago? Feel like I’m having Deja vu.

He's been experimenting with GPUs on the Pi for a long time, only now it's getting to a usable state

He tested "EVERY graphics card on a Raspberry Pi" (Apr 13, 2023): https://www.youtube.com/watch?v=l9dItRUjQ0k

This is great. A big drawback of arm chips is that none of their GPUs seem to support OpenGL, just GLES. Being able to pair a Raspberry Pi with a dedicated GPU and shave hundreds of dollars off the price of assembling a machine is going to go far towards getting AI, or even just desktop research programs like ChimeraX, into the hands of researchers in low purchasing power markets.

Snapdragon Adreno's freedreno GPU drivers have full OpenGL support: https://mesamatrix.net/#OpenGL , but definitely the exception. I think you can get some support through Zink on top of Vulkan, and the Pi's Vulkan drivers have been improving.

Thanks for the pointer! I'll be trying to set up Zink and see if I can write some documentation around using it for ChimeraX.

Crazy how big his channel has become. I used to install bare metal servers with Jeff's Ansible scripts almost a decade ago!

I am looking for a low power at idle (5-10 watts) server that could activate/deactivate gpu at will as a GPU takes a lot of extra watts at idle. The use case is that the server keeps running for NAS/home server stuff and can activate a gui environment at will (Wayland seat) for gaming. Bonus points for the software able to handle multiple people (seats) for small desktop use cases. A small ARM SBC makes sense. Give me hardware encoding for AV1 and take my money.

I want to experiment with graphics programming without using APIs like Vulkan and write GPU code at the lowest level possible. I believe a lot of complexity can be shaved off with added advantage of a custom rendering toolchain. I can do native voxel primitives and ideas from Ken Silverman's old voxel engine, all implemented in the GPU. The only problem will be cross platform support.

I would really love a low cost board with open GPUs for that. Custom software renderers are the future I am looking forward to. Its the holy grail to ultimate developer freedom, without the complexity and verbosity of current day situation. Cross platform will need work from hardware manufacturers though!

Ignoring supporting multiple GPU vendors for a minute, the backwards and forwards compatibility would likely be bordering on non-existent.

Booo, I wanted to see the big GPU nestled on top of the Pi 5 in the Pineboards uPCIty Lite HAT :D

In the live stream version of this video, he does strap it directly on top of the HAT: https://www.youtube.com/watch?v=EAlrCFJZlnI

>[through external GPU card]

So not interesting at all then.

Getting an external GPU to work on a raspberry pi is a huge feat... PCIe support, functional drivers, these things don't just appear out of thin air. Jeff has been working on getting to this point for multiple years now.

I would recommend reading his blog posts on the matter to see how much progress there have been, as well as the issues along the way.

I like his content, but these sort of unbelievable thing on a raspberry pi articles with clickbait titles aren't particularly interesting when they essentially involve offloading all the work to an external component that is more powerful than the pi itself.

The point is that he made it possible to do the offloading.

Did he? I started to skim the video and it looked like he just used a lot of commercially available adapters.

Reply on news.ycombinator.com