Will the Ryzen AI 300 do AI?

Framework Laptop 13 Linux
1

So this mainboard is supposed to support ROCm, right? From the blog post:

image
image2018×454 99 KB

But it appears there’s no ROCm support, and none in progress:

Framework’s statement about this laptop’s suitability for ROCm and AI seems pretty clear … what am I missing?

Thanks!

Which Linux distro are you using? Bluefin
Which release version? :latest, updated last night
Which kernel are you using? 6.15.7-200
Which BIOS version are you using? 3.03
Which Framework Laptop 13 model are you using? AMD Ryzen™ AI 300 Series

2 Likes
2

These are currently the ways to use. Ramalama probably one of the easiest as it uses the GPU and pulls the needed stuff.

3

Bigger picture, Framework claimed this laptop would do ROCm, but AMD never came through. I see no ROCm support, no AI features working, and nothing on AMD’s roadmap. It seems like this mainboard was abandoned before it was released.

I’m disappointed that my AI 300 Framework is not good at AI, and Ramalama’s plodding performance seems to confirm this.

Just trying to see if I’m missing anything? Anyone have an alternate experience?

4

This is true for most “AI” PCs, its not just a framework problem. It was industry-wide misleading marketing for based on vaporware.

Framework screwed up by mentioning ROCm, they shouldnt have mentioned a feature that didnt exist yet. But ill cut them some slack here because ultimately it’s AMD who’s making “AI” chips that arent even supported by their own software.

1 Like
5

Agreed, except I didn’t buy this laptop from AMD. I bought it from Framework.

If AMD lied to Framework about ROCm, then I’d hope Framework would at least acknowledge it. If Framework is doing the lying, though, then I guess the current situation makes sense.

1 Like
6

@bron looks like support is coming in rocm 7. While there is no official info what’ll be supported, there are files in there for all FW’s current AMD platforms and for the desktop. Looking at pre-release timing, they’ll probably get 7.0 out in a month, and i’ll take another 2-4 weeks to integrate it into the main ai packages, and then like another month to get to 7.1 with all the bugfixes.

RIght now you can try rebuilding code on top of beta which is PITA. I also won’t be surprized if there already are docker images that bundle beta libs. If someone finds them and shares here, that’d be great!

But yeah, AMD fucked up on this one with marketing selling stuff 2-3 quaters before the software is ready. And I doubt we’ll see reasonable NPU support withing half a year; I don’t count “install this obscure stack, windows only” as reasonable support. FW also messed up by proxying; they might be able to exert a small amount of pressure on AMD, but AMD won’t put consumer hw on linux before their MI300s, that’s not where the money is.

Good thing that AMD is catching up and support seems to be coming, and they aren’t abandoning recent hw.

4 Likes
7

@Alex_Morozov agreed with everything you say. And thank you for the potential good news. Fingers crossed.

Framework needs to be more careful repeating AMD’s empty promises.

8

A previous thread covering this disaster: AMD ROCm does not support the AMD Ryzen AI 300 Series GPUs

1 Like
9

Why did you only have a look on AMD support, it is something “special”.
If you use fedora 42, SIGs/HC - Fedora Project Wiki
the gfx1150 gfx1151 gfx1152 is activated for there rocm distribution on fedora 42.

In most, case the GPU is fater than the NPU. To have the full perf of the NPU you need model that use the “block float” there is realy not much that exist.

1 Like