AMD Ryzen AI 9 365

AMD has produced new processor, which has NPU acceleration. That doen’t give advantages if you use AI with your browser. But if you use Windows and have Copilot in Office, it works. I don’t know if it works also in Windows without Office.

I think that the development of operating systems goes towards AI. And then I think that KDE might take AI advantages to use.

Keep AI out of ANYTHING KDE including the code development.

Unfortunately, this hardware has an active neural processing unit (NPU) –

 # inxi -xxx --filter --cpu
CPU:
  Info: 6-core model: AMD Ryzen 5 8600G w/ Radeon 760M Graphics bits: 64
    type: MT MCP smt: enabled arch: Zen 4 rev: 2 cache: L1: 384 KiB L2: 6 MiB
    L3: 16 MiB
  Speed (MHz): avg: 2994 min/max: 414/5076 boost: enabled volts: 1.2 V
    ext-clock: 100 MHz cores: 1: 2994 2: 2994 3: 2994 4: 2994 5: 2994 6: 2994
    7: 2994 8: 2994 9: 2994 10: 2994 11: 2994 12: 2994 bogomips: 104202
  Flags-basic: avx avx2 ht lm nx pae sse sse2 sse3 sse4_1 sse4_2 sse4a
    ssse3 svm
 # 
 # journalctl --no-hostname --dmesg --grep=xdna
Feb 17 10:37:39 kernel: amdxdna 0000:0b:00.1: enabling device (0000 -> 0002)
Feb 17 10:37:39 kernel: [drm] Initialized amdxdna_accel_driver 0.1.0 for 0000:0b:00.1 on minor 0
 #

More information here: <Neural processing unit>

rocminfo” shows that, Agent 2 – gfx1103, AMD Radeon 760M Graphics – has 2 entries in the “ISA Info:” –
both entries contain amdgcn-amd-amdhsa–gfx11 information.


To ignore AI or, not to ignore AI – that’s the question …

  • Given that, for the CPU I’ve shown above, the hardware capabilities are possibly intended to be used by applications which use a small already trained AI model, I suspect that the Desktop Environment as such doesn’t need to be involved.

For example, the KDE Plasma application digiKam has the following capability:

Metadata Enrichment: Elevate your photo management with intelligent AI-driven tagging and rating. Automatically enrich your images with detailed metadata, making it easier than ever to organize and find images.