"RTX 2000 Ada Generation Laptop GPU" vs. 3000 version

Hello Forum!

What performance difference might I expect between the “RTX 2000 Ada
Generation Laptop GPU” and the “RTX 3000 Ada Generation Laptop GPU”
when training garden-variety pytorch models?

I am considering getting a laptop* and these two gpus are among the choices.
(The rtx 500 and rtx 1000 are also in the mix, if people have thoughts on those.)

The 2000 and 3000 both have the same 8GB of memory as well as the same
base clock speed. The main relevant difference seems to be that the 3000
has about 50% more “CUDA Processing Cores” (4608 vs. 3072).

More cores sounds good, but would they really get used? How about thermal
considerations? Would one be able to saturate most of the cores on a laptop
gpu for a long training run without getting thermally throttled?

I haven’t found any comparisons that address machine learning or pytorch.
The comparisons I’ve seen concern rendering and gaming frame rates, a
rather different use case.

Thanks for any information or advice.

*) Specifically, I’m looking at the Lenovo ThinkPad P16v Gen 1 (AMD) or
the Gen 2 (Intel). Aside from the gen1 not supporting the rtx 3000, the main
difference seems to be the gen 1 is based on the AMD “Ryzen 5 / 7 / 9 Pro”
series of cpus, while the gen 2 is based on the Intel “Core Ultra 7 / 9” cpus.
Regarding that, my thinking is that the specific cpu won’t have much effect
on the speed of gpu-accelerated pytorch computations. Is that fair? Any
more general input on the cpu differences would be appreciated.

Best.

K. Frank