Quote from: Terror Byte on Today at 03:16:55Ngreedia will probably want 30% more than HALO products.

Sure, but good thing is that for inferencing (not to be confused with training an AI, where CUDA is the go-to solution), which this N1(X) is designed for (up to 128 GB slow[1] LPDDR5X RAM), is mostly hardware agnostic once vulkan works.

[1] 273 GB/s = 256 bits * 8533 MT/s / 1000 / 8.

AMD's Strix Halo: 256 GB/s = 256 bits * 8000 MT/s / 1000 / 8.

Skip N1, N2 is coming quickly in 2027. Let N1 be for beta testers. I hope it's good though and gives Qualcomm, AMD and Intel some good competition. But it will aslo come down to price. Ngreedia will probably want 30% more than HALO products.

As usual in this time frame, it will all come down to memory speed (width).