Running training is going to throttle the GPU quickly due to heat buildup. It’s just a fact that these systems have limited airflow and can’t sustain super high clock speeds / usage under load - esp ml training loads which really put systems under duress.

I don’t know enough about either build but something to be aware of.

I wouldn't recommend buying a laptop for deep learning. It will cost a ton, weigh a ton, run extremely hot, have almost no battery life, and throttle severely. I use a MacBook Air and connect to colab or cloud GPUs for deep learning and it's a great experience.

Have you considered other options?

I consider laptops of this class the worst of all worlds. They are too large and heavy to really be considered “portable”. Sure you could move them if you wanted to. At the same time, their form factor doesn’t dissipate heat well enough to realize the full potential of the specified hardware.

Other options:

1. Get a desktop. Maybe not practical if you aren’t always working from the same place. IE. Sometimes at home and sometimes in office. You could almost certainly build a desktop for less than equivalent advertised specs in a laptop form factor.

2. Spend ~$1.5k on a mid range more portable laptop and ~$1.5k for an eGPU on a desk. You still have a laptop to bring around places and you have more horsepower somewhere when you need it.

3. Spend <$1k on a laptop and connect to cloud services

Why are there daily threads on buying laptops for machine learning?? Let’s be real, you want to play games and want a good laptop for it, and u think it’s worth it if it can “also serve as a ML machine”. If u wanna do ML, don’t buy a laptop. Use colab or one of the many other cloud providers

What sort of loads are you planning on running on your laptop? If you’re running LLMs the answer’s gonna be totally different than if you’re trying to train RL agents in grid worlds.

I bought an MSI gaming laptop some years back (~$1500) and it worked good for basic stuff while I was learning Neural Nets. MSI was about the best price-wise and there was a slot for a 2nd SSD so it was easy to put Ubuntu on a 2nd disk and keep Windows as-is for other work.

I have to agree with other comments however. The laptop will run very hot under heavy training and in my experience, the mobile versions of the GPUs are not nearly as fast as the similarly named desktop versions but cost considerably more.

BTW... If you're just doing basic NN training for school you probably don't need a 14-core laptop and 64GB of RAM. You can probably get by with something smaller, maybe a 3070 or 3080 (no TI) and then rely on other resources like Colab for training larger more complex nets.

I ended up getting an Asus Zephyrus G15 with a Ryzen 9, 32GB Ram and a 3080 mobile. It's been just fine for getting baseline numbers and being able to run most things for debugging purposes. You wouldn't want to run long training sessions on any laptop as the other comments say, but when I bought mine it was ~$2k. It's absolutely not worth spending more than that for a machine that is going to be limited to approximately the same performance that lower-priced laptops will have because of thermals. Don't buy the $5k system (or frankly the $3k system either). Way overpriced and the extra $ won't buy you any performance or productivity over much cheaper systems. It's all flash and marketing.

Completely agree with this. A $2k laptop is nice for experimenting/debugging. Put your extra $$ into a service or maybe buy a desktop with a big GPU at some point in the future.

Buy a pc