Framework 16 and Deep Learning

Framework Laptop 16
46

Hi,

I have not tried pytorch, but have played with rocm on a FW16.

  1. rocm does not use vram. It uses GTT based memory. So set your vram to the smallest in the bios.
  2. no copy to/from gpu ram is needed.
  3. you need to do a gpu sync before reading the results.
  4. I found ROCM to be very buggy, as I wrote in this thread:
    Experiments with using ROCM on the FW16 AMD

So, unless pytorch is known to work on APUs, it might not be using the api correctly and instead trying to do copy to/from gpu ram.

At the FW 2nd Event AMD were also on stage:
Anush Elangovan - VP or AI software
Talked about ROCM.
2023 - Focused on day zero support of Models (LLAMA, DeepSeek)
2024 - Focused on performant day zero support.
2025 - Focus on accessibility of ROCM
We get pytorch to work on all of AMDs AI hardware.
Starting from the Laptops, to Desktops, to the Instinct GPUs.

So, if pytorch does not work, it sounds like better support might be coming in 2025.

2 Likes