There are separate chips.

But just like Strix Halo, they have to be soldered. There’s no way to reach the signal integrity required with connectors.

▲

zero_bias 6 days ago | parent | next [-]

No, M series is a system on chip (SoC), that’s why it’s able to run local LLM models in a range impossible for other laptop brands: VRAM == RAM, unified shared memory at max speed for both CPU and GPU

	▲	dontlaugh 6 days ago \| parent [-]
		Strix Halo has the same unified RAM with no separation. Sadly it’s not in many laptops, probably the easiest way to obtain it is in the Framework Desktop or a mini pc.

▲

benoau 6 days ago | parent | prev [-]

I've heard many people saying CAMM2 solves this.

▲

wtallis 6 days ago | parent [-]

We're still waiting to see any CAMM-style memory module show up in a mass market product at any speed, instead of merely getting press coverage where the number of articles written seems to outnumber the number of laptops actually built and shipped. But even if you are willing to take the examples thus far seriously as real products, they haven't come close to matching the speed of soldered LPDDR.

▲

WesolyKubeczek 6 days ago | parent [-]

I was considering an LPCAMM2-fitted Thinkpad. I was eyeing to buy one with less memory and then buy a 96GB module to upgrade it. However, the module was nowhere to be found in stock, and where it was found, it was priced almost like the whole laptop.

▲

zozbot234 6 days ago | parent [-]

CAMM is still less effective than in-package RAM bundled with your CPU. The Framework folks looked into using CAMM for their recent AMD APU-based desktop and it was a no go.

	▲	WesolyKubeczek 4 days ago \| parent [-]
		It could be a compromise that’s better than DIMMs, though. Not every task needs this insane memory bandwidth and (as small as it gets) latency, although I admit it’s so damn nice, but it’s also nice that if memory goes bad, I can just replace it.