Intel MTL power usage is confusing.

[Haswellx86]

Looks like Intel finally paid Linus for their regular shill subscription, and now they actually put out a video after ignoring any reference to MTL before. But this got me checking for some stuff, and what I found is weird. They were showing how there is an IPC regression in MTL, but something that clicked me in my mind is that is the 155H even the competitor for the 1370P? It is hard to decide this as they both now have different naming schemes, and a good idea I thought would be to compare their power usage (yes I know, and i7 could be better than an i5 at the same power, but power throttling is a thing which diminishes this effect). But what If found is that the max turbo power for the 155H is 115 watts! Compare that to the 1370P which has 64 maximum turbo power. They both have the same amount of cores (6P+8E) and yes those 2 LPE cores which nobody cares about, but upon that, the 155H only clocks to 4.8 max, whereas the 1370P clocks to 5.2! Even the Ultra 9 can only clock to 5.1. Where is this extra power coming from? Are they taking in account of the much more beefier IGPU and the NPU? 

 

Also, looking at the 165H specs, it has 5 GHz turbo, which is closer to the 1370P than the 155H, so why not use a laptop to benchmark that uses a 165H? It has the same power requirement if the specs are to be believed (but yes, power throttling will likely get both of them to 28 watts only and from here, only their efficiency will depend for the clock speed). They really should have shown a clock speed and power usage graph. Could have given us a better idea about the efficiency difference.

 

Also, am I the only one hating the tile-based modular approach Intel is looking forward to? The only reason I liked Intel is because of the monolithic design they made which let cores talk directly to each other, minimizing latency. Compare that to AMD, who always keep having some problems between CCD inter-communication. I guess this really did hurt AMD's performance but they still found ways to make really fast cores which out way this problem and still be competitive. At least Intel is not running their connections through the PCB. Using a silicon interposer, the latency can be much more minimized, but there is still some latency compared to monolithic design. If you do don't think that this latency is that big of a problem, watch multiple videos of this guy - https://www.youtube.com/watch?v=6O5XGVaPDZo (although I don't know if this is relevant here as it seems to be a CPU scheduling issue).

 

If you notice thoroughly, all this modular approach is going to do is let the processor shut down more of itself which a) only benefits laptops and we better not even have this on desktop, and b) this still doesn't help the actual power efficiency of the processor. Most of the time, you would be actually doing stuff rather than keeping your laptop on standby the whole time for the processor to shut more of itself down and save power. The only reason Intel doesn't release these new tech on desktop early is because they fail to reach the performance target of desktops, as they did now.

 

@Gat Pelsinger indeed. No I am not an Intel shill, at least not any more. But hey, the new IGPU and NPU is cool tbh.