Ryzen come with bundled gpu. It's called Vega.

However you have to pay the Nvidia tax to do anything.

So a nice AMD ultrabook (which reigns on current price/performance charts) has the firepower, but no access to ML libraries.

The AMD GPU acceleration layer is called ROCm . Zero support - not even with the brand new Google libraries like JAX.

https://github.com/RadeonOpenCompute

Unfortunately not on the newest consumer cards yet. Their CUDA competitor is ROCM which isnt available on the Navi 10 cards i.e the Radeon 5600-5700 series. I've been watching the space closely and it seems like support might land relatively soon though.

(Performance as compared to CUDA on cards that are supported is not great however)

You'd be looking for something with a discrete Nvidia graphics card with as many CUDA cores as possible.

Recommendations...depends on $$$...high end laptops are expensive as hell. Maybe something like a 1660 Ti?

Broadly I'd avoid ML work on a laptop though due to thermals

Get anything with a nvidia GPU. Without a nvidia GPU, it's pretty much dead

I wouldn’t go for a laptop for ML and I don’t care if CPU is Intel or AMD... parent commenter is mixing up CPU and GPU

Unless you're trying to do ML with real time video or audio feeds, I'd just get a random cheapo laptop and do all your stuff remotely with something like jupyter notebooks (eg. Google Colab).

Having a remote GPU which is 10x more powerful than you can afford to buy yourself, and you can scale up to 50 GPU's at the click of a button is a far better workflow than spinning fans on your laptop overnight, only to wake up in the morning and realise a code tweak is needed and then you have to wait another 12 hours...