[sudosysgen]

Unified memory literally is RAM. It's just LPDDR5. It might have better timings, due to being located closer, that's it. The framework 16 also does DDR5-5600, so you get 100GB/s of bandwidth with a dual rank memory setup.

In theory you'd take a latency hit, but in reality (at least the M1) was limited by it's memory controller and had DRAM latencies of ~100ns, which is actually worse than Zen 4 at ~70ns. This is because before querying RAM, you first need to query every level of cache, and then you hit the latency of the DRAM itself, so the actual distance is not a dominant factor.

So no, unified memory doesn't give any actual performance advantage. It could in theory (and marginally), in practice it's just marketing.