QuoteLenovo would probably reduce the IdeaPads to a single SO-DIMM slot.

Token generation (output) solely depends on memory bandwidth (and prompt processing (input) solely depends on GPU's FPS performance / compute), so halving the SO-DIMM RAM channels will halve the token generation. Not only in an self-hosting-AI-models age is this the wrong decision.

I think we all understand that a purchase with AMD is possible because having dual-channel RAM is essential; it greatly increases performance and improves the use of artificial intelligence in offline mode by 100% without having to send data to servers. Now we just need to know the price, or wait for AMD Zen 6, which will be a leap from 4nm with TSMC to 2nm and a new microarchitecture.