[seanalltogether]

I'm really curious if unified cpu/gpu chips are the future for laptop/desktop hardware. Mac is now unified across its product line, consoles are unified, phones are unified. My limited understanding though is that unified memory means giving up either high speed (for the cpu), or high bandwidth (for the gpu). Is that correct?

[brucethemoose2]

You can have your cake and eat it with a wide LPDDR5X bus. This is what Apple does, and its fine. Other specialized chips (like Tenstorrent's accelerators) do this too.

I think an AMD 7900-like approach where the memory controllers/cache are on tiny external chips is particularly practical. Its efficient and economical. I hope AMD (and others) repeat this with laptop CPUS.

GDDRX is not a good fit for laptops anyway because its so power hungry. GPUs and the Xbox/Playstation use it just because its the absolute cheapest bandwidth/$, at the cost of everything else.

HBM is very expensive and being hoovered up by the AI market. I wouldn't count on seeing it in consumer stuff again.

[wmf]

It's more of a low latency (DDR/LPDDR) vs. high bandwidth (GDDR/HBM) tradeoff.