Enma45, Strix Halo already runs at 8000 MT/s. LPCAM2 would make the memory being swappable/upgradable, which would, of course, be very good. But the memory bandwidth might remain the same, unless they up it to LPDDR5X-9600 MT/s, or switch to LPDDR6.
And now the major Strix Halo limitation: Only up to 128 GB RAM/unified memory. This does not offset the switch from a dense LLM to a MoE LLM: Qwen3.5-27B (27 GB at q8 quant) performs slightly better than Qwen3.5-122B-A10B (122 GB at q8 quant). The 122B runs 27B/10B = 2.7 times faster, fine, but it performs slightly worse than the 27B dense AND the RAM prices have since increased by like 4 times. So, Strix Halo's 128 GB RAM, at 256 GB/s, are neither here, not there. What is really needed is at least a 192 GB Strix Halo, then it could run a quant of the better than 27B dense performing MoE models (at least according to AA), like DeepSeek V4 Flash, MiMo-V2.5 and Minimax-M2.7 and a 256 GB RAM (yes, RAM prices..) config would fit Minimax-M3 and probably some others that are not in this:
artificialanalysis.ai/?models=gpt-oss-120b%2Cgpt-oss-120b-low%2Cgemma-4-31b-non-reasoning%2Cgemma-4-31b%2Cmistral-medium-3-5%2Cdeepseek-v4-flash-non-reasoning%2Cdeepseek-v4-flash-high%2Cdeepseek-v4-flash%2Cminimax-m3%2Cminimax-m2-7%2Cstep-3-7-flash%2Cmimo-v2-5-0424%2Cqwen3-6-27b-non-reasoning%2Cqwen3-6-35b-a3b-non-reasoning%2Cqwen3-6-35b-a3b%2Cqwen3-5-122b-a10b%2Cqwen3-6-27b%2Cqwen3-5-122b-a10b-non-reasoning&intelligence=artificial-analysis-intelligence-index
Strix Halo's 4060 Laptop performance also means not very fast prompt processing.
Basically, Strix Halo needs to double and everything, both, the memory bandwidth and the out of it resulting FPS/prompt processing speed and in memory size (to 256 GB).
PS: I see that Enma45's comment has disappeared..