[condiment]

It's GPU performance.

Spin up ollama and run some inference on your 5-year-old intel macbook. You won't see 4000x performance improvement (because performance is bottlenecked outside of the GPU), but you might be in the right order of magnitude.

[blihp]

Not possible given the anemic memory bandwidth [1]... you can scale up the compute all you want but if the memory doesn't scale up as well you're not going to see anywhere near those numbers.

[1] The memory bandwidth is fine for CPU workloads, but not for GPU / NN workloads.

[jandrese]

Comparing GPU performance to some half decade old Intel IGP seems like lying with statistics.

"Look how many times faster our car is![1]"

[1] Compared to a paraplegic octogenarian in a broken wheelchair!"

[umanwizard]

Well, Apple isn’t making that comparison, the OP was.