Linux?

https://www.phoronix.com/news/AMD-Ryzen-AI-Open-Source-Demo
AMD Has Open-Source Ryzen AI Demo Code - But Only For Windows

It's not fully open-source, even for windows - many libs are binary only. They basically open-source UI and API to the model.

It's not fully open-source, even for windows - many libs are binary only. They basically open-source UI and API to the model.

This is how a story gets off to a bad start.
Providing this support would ensure seamless integration with the Linux ecosystem, promoting open interoperability and enabling users to maximize the potential of AMD's AI technologies. Embracing Linux would not only expand your user base but also foster a more inclusive and collaborative technological environment.

Not being jailed at Microsoft shouldn't even be a debate.

AMD is losing the Linux guys https://youtu.be/Mr0rWJhv9jU

AMD is losing the Linux guys https://youtu.be/Mr0rWJhv9jU

Dunno about anyone else, but I don't have 5.5 hours to burn on a random YouTube video.

Not sure why this is closed as there still is no Linux support.

Dunno about anyone else, but I don't have 5.5 hours to burn on a random YouTube video.

The video has timestamps with AMD issues split into roughly 4-minute sections which one can skip to.

But TL:DW: it won't happen, not even for proper Linux drivers

@andyluo7

@andyluo7 can you reopen this ticket? Maybe change the title and etc. As I understand, it's a Xilinx device, and we already have a driver, but maybe we should extend it a little. I understand that this specific demo is designed for Windows, but we can create the same thing for Gwenview for example (from KDE Desktop).

Linux upvoted by chaudhariatul here

Kindly suggest which is an appropriate option from below, which will help with buying a mini pc:

Linux will never be supported on 7040
Linux support for 7040 after 5.6.2 or maybe 5.6.X
Linux support with 8000 series

Re-opening issue so that customers can log their Linux requests.

+1

New release, still no Linux? Dissapointing!

No, it doesn't use the NPU on Linux. This thread is about NPU support on Linux.

…

-------- Original Message --------

On 9/24/25 14:08, mihairaduonofrei wrote: mihairaduonofrei left a comment [(amd/RyzenAI-SW#2)](#2 (comment)) > NO. It doesn't. It does support GPU. > I see, AMD Ryzen AI software uses both the NPU and the integrated GPU for different functionality: - CNN models and power-sensitive AI inference tasks run on the NPU - large, iterative generative AI models run on the integrated GPU as a separate GPU resource. — Reply to this email directly, [view it on GitHub](#2 (comment)), or [unsubscribe](https://github.com/notifications/unsubscribe-auth/AA5FWNUPWTMLRCHP2NBKWZD3ULMYZAVCNFSM6AAAAAAYZSNTHKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGMZQGA4TIOJTGI). You are receiving this because you commented.Message ID: ***@***.***>

Need it soon.

When to use what?
RyzenAI with Lemonade
or
RocM with vLLM and Pytorch

There is now Linux support in Ryzen AI v1.6.1: https://ryzenai.docs.amd.com/en/latest/relnotes.html#version-1-6-1
It is available only via Early Access Lounge. At least it is rolling!

Please keep moving forward with Ryzen AI support on Linux. I'm a Linux only guy and I want to utilize the NPU. Without it the AI chip is just a waste.

+1 for linux NPU support

it's happening? https://canonical.com/blog/canonical-amd-rocm-ai-ml-hpc-libraries

it's happening? canonical.com/blog/canonical-amd-rocm-ai-ml-hpc-libraries

Please, please, READ THE ARTICLE YOU LINK before linking. Especially the first paragraph.

AMD ROCm is an open software ecosystem to enable hardware-accelerated AI/ML and HPC workloads on AMD Instinct™ and AMD Radeon™ GPUs, simplifying the deployment of AI infrastructure with long term support from Canonical.

It's for GPUs. NOT NPUs.

it's happening? canonical.com/blog/canonical-amd-rocm-ai-ml-hpc-libraries

Please, please, READ THE ARTICLE YOU LINK before linking. Especially the first paragraph.

AMD ROCm is an open software ecosystem to enable hardware-accelerated AI/ML and HPC workloads on AMD Instinct™ and AMD Radeon™ GPUs, simplifying the deployment of AI infrastructure with long term support from Canonical.

It's for GPUs. NOT NPUs.

Hm, apologies, I thought ROCm included XDNA integration. The terminology is quite confusing.

It's not fully open-source, even for windows - many libs are binary only. They basically open-source UI and API to the model.

This is how a story gets off to a bad start. Providing this support would ensure seamless integration with the Linux ecosystem, promoting open interoperability and enabling users to maximize the potential of AMD's AI technologies. Embracing Linux would not only expand your user base but also foster a more inclusive and collaborative technological environment.

Not being jailed at Microsoft shouldn't even be a debate.

About two and a half years later. Finally, we're here!
Very happy that now we have a NPU inference working demo!
For anyone interested I wrote a little guide here for NPU inference on Linux with FastFlowLM: https://community.frame.work/t/guide-use-npu-xdna2-with-arch-linux-and-fastflowlm/80879/3

Ryzen AI 1.6.1 for Linux still requires access to the Ryzen AI Software Early Access Lounge. However, the request form has now been removed, and AMD stated that all critical Ryzen AI software features, architectural updates, and engineering patches are publicly available and accessible. They also mentioned that the Early Access page will be removed since all alpha deliverables have reached GA status.

When considered together with the discussion in ROCm/legacy-rocm-build#331, does this suggest that a public release of Ryzen AI Software for Linux is coming soon? I hope that official packages will be provided at least for Ubuntu 26.04 LTS.

It's working! Claude Opus 4.6 figured the build steps to get this running in Linux. Here is a repo with a Dockerfile:

https://github.com/hpenedones/fastflowlm-docker

I'm getting about 60 tokens per second on the NPU: I added some benchmarks results to the README.

@hpenedones

It's working! Claude Opus 4.6 figured the build steps to get this running in Linux. Here is a repo with a Dockerfile:

https://github.com/hpenedones/fastflowlm-docker

I'm getting about 60 tokens per second on the NPU: I added some benchmarks results to the README.

It took me some tries to get to work, I had to switch to kernel 6.17 which gave upgrade errors so I manually compiled and upgraded the xdna-driver, but I have success!
Added my steps to a working NPU in an issue on your repo @hpenedones : hpenedones/fastflowlm-docker#1

❯ docker run --rm \
  --device=/dev/accel/accel0 \
  -v /opt/xilinx/xrt:/opt/xilinx/xrt:ro \
  -v /usr/lib/firmware/amdnpu:/usr/lib/firmware/amdnpu:ro \
  -e XRT_ALLOW_0_VER=1 \
  --ulimit memlock=-1:-1 \
  fastflowlm validate
[Linux]  Kernel: 6.17.0-14-generic
[Linux]  NPU: /dev/accel/accel0 with 8 columns
[Linux]  NPU FW Version: 255.0.11.69
[Linux]  amdxdna version: 1.0
[Linux]  Memlock Limit: infinity

Verbose: 
  Total tokens:        591 (0.45%)
  TTFT:                598.16 ms
  Prefill speed:       73.30 tokens/s
  Decoding speed:      59.84 tokens/s

❯ xrt-smi examine
System Configuration
  OS Name              : Linux
  Release              : 6.17.0-14-generic
  Machine              : x86_64
  CPU Cores            : 16
  Memory               : 63434 MB
  Distribution         : Linux Mint 22.3
  GLIBC                : 2.39
  Model                : 21M1CTO1WW
  BIOS Vendor          : LENOVO
  BIOS Version         : R2NET39W (1.13 )
  Processor            : AMD Ryzen AI 7 PRO 360 w/ Radeon 880M

XRT
  Version              : 2.23.0
  Branch               : HEAD
  Hash                 : f8e923a4168dd552e02810de81288c34b416a0b4
  Hash Date            : Tue, 3 Mar 2026 10:59:04 -0800
  amdxdna Version      : 2.23.0_20260309, a8f1ade32b324527821f8165ed05a61baedcb456
  virtio-pci Version   : 6.17.0-14-generic
  NPU Firmware Version : 255.0.11.69

Device(s) Present
|BDF             |Name       |Version  |Topology  |
|----------------|-----------|---------|----------|
|[0000:c4:00.1]  |NPU Strix  |AIE2P    |6x8       |

+1

Navigation Menu

Repository navigation

Linux? #2

1107 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Linux? #2

Description

Activity

stalkerg commented on Jun 4, 2023

GreyXor commented on Jun 4, 2023

grigio commented on Jun 5, 2023

MrDrMcCoy commented on Jun 5, 2023

nospace-here commented on Aug 27, 2023

stalkerg commented on Aug 27, 2023

quentonh commented on Aug 31, 2023

chaudhariatul commented on Aug 31, 2023

quentonh commented on Oct 20, 2023

acwn1976 commented on Oct 20, 2023

Alexander-LIT commented on Oct 20, 2023

gel-crabs commented on Oct 20, 2023

1107 remaining items

LouDnl commented on Oct 10, 2025

padthaitofuhot commented on Oct 17, 2025

Zer0XoL commented on Oct 28, 2025

onofreiciuc commented on Oct 30, 2025

VladimirVLF commented on Nov 8, 2025

cpreisler commented on Nov 10, 2025

elhigu commented on Dec 6, 2025

dmjef commented on Dec 10, 2025

ThatOneCalculator commented on Dec 10, 2025

dmjef commented on Dec 11, 2025

GreyXor commented on Feb 25, 2026

lego37yoon commented on Feb 27, 2026

hpenedones commented on Mar 8, 2026

LouDnl commented on Mar 9, 2026

atleast4IQ commented on Apr 26, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions