People should be mindful that whatever marketing person put those slides together got some stuff wrong. Good old marketing.
Anyways, the backside power delivery is by far the most interesting thing from the presentation, so of course nobody's talking about it. Let's walk through it.
1 - There's potential for immediate power savings. CPU's can lose 15% of their voltage due to IR drop. Let's say that backside power delivery cuts it by 1/3 (possible, maybe not likely), that's an immediate 10% power reduction (power scales with the square of voltage, .95^2=.9025).
2 - It gives additional routing resources at the same node. While logic has scaled very well, metal pitch has been moving at a snail's pace. In a node transition where logic density doubles, metal pitch may only scale by 10% on some layers, 5% on others, and not at all on others. As a result, the number of wires available per cell decreases. Ripping out the power grid can free up about 10% of the tracks on most layers and up to 50% on others. Also, it would allow for much less restrictive via placements on lower layers, which could dramatically improve routability.
3 - It trivializes legalization during design. Currently, cells need to be placed around the power grid to avoid low level connections. With backside supply, those connections don't exist, so you can place cells wherever you want. It gives a lot of freedom to standard cell designers to sacrifice metal for improved performance or power of the individual cells, and it gives the chip designers more freedom when optimizing cell placement.
Obviously there are big "If's" and caveats, but this has the potential to improve power efficiency and greatly streamline the design process. Also, most of the metal would be freed up on the highest (aka fastest) metal layers (and it's not even close), so chip architects should be able to increase IPC and reduce power by putting more signals on the fastest layers and reducing their latency.
Meanwhile, we are going to have, on desktop, the biggest IGP gap in history the second RMB shows up.
Yes, because if there's one thing AMD's known for, it's releasing its APU's on desktop platforms in a timely manner.
A bit concerned for AMD in the mobile space, TBH. They seemingly have nothing to go up against Golden Cove as Rembrandt is "Zen 3+" (unless something is missing from the leaks). I guess we will see how it plays out. All I know is that in another 1-2 generations, many people won't need anything beyond a laptop if this progress keeps up.
IMO Golden Cove's not the one to worry about. Sure it'll be faster than Sunny Cove, but also even larger and more power hungry. Gracemont's the one to pay attention to. If Gracemont's good, Alder Lake is good, and the reverse.