I don't know why several people in this thread insist on giving me (well known) history lessons for something that was obviously said in a specific context.
i was clearly talking about the kind of custom builds meant to accommodate a mobile CPU in a desktop environment. A few posts up I even put the original Anandtech article talking about this.
Had any fresh ideas like 3 years ago so something like that. The output of these chips makers is like watching stars on the firmament, what you actually see is from way in the past.
It doesn't have to be a radical design change (as much as we would love to see something new), a gradual change in focus from clocks to PPC may be enough, and I believe @IntelUser2000 spoke about this as well, even in relation to Alder Lake and Golden Cove. The shift can be gradual, sacrificing 5-10% clocks at a time in favor of a more powerful core with lower max clocks.
Not in a typical sense. Remember it's BGA, so if you have the motherboard developed and you only need the CPU from Intel then yes you'll probably be able to buy it.
But unless you are a super hardcore DIY that goes through the process of fully custom designing a motherboard to work with modern processors, you are likely a small company that needs the CPU to ship in their products.
If you look at Zen 3, or even Apple M1, it does not deviate from the Core philosophy. They are just wider and bigger. So in that case they will not change.
Superscalar? Pentium.
Out of Order? Pentium Pro/Pentium II.
Micron Op Fusion? Pentium M
Macro Op Fusion? Core.
Physical registers? Done with Sandy Bridge.
uop Caches? Sandy Bridge again.
What needs to change is their focus on absurd, inefficient 5GHz+ frequency aim for something much less but much wider.
Pentium M might have it's ancestry based on Pentium II, but really it's the philosophy of going about it that changed. In a way Netburst is actually Pentium III philosophy continued. Because nothing really changed. Because in the Pentium III days you just needed to move to a fancy new process and voila! you get a much better CPU. Pentium M was when they decided "We need to be different in our approach, it's not about clocks. We might reach a point where the new process itself is not enough".
Netburst's 4-5GHz frequency target was reduced to 3GHz in Core right? So iterations of Core made it reach 4-5GHz again. So focus on going back down to the 3-4GHz range. Pretty much rinse and repeat at this point because as I've said it before and again, unless you are going exotic cooling, anything much higher than 5GHz is impossible.
Maybe this Hybrid thing will give them more flexibility in this regard but it doesn't change much if the big core portion is super large and inefficient.
Let's say if Core was efficient as Zen 3? Alderlake might have been 12+12.
While this may end up being true at 95W+ TDP, for mobile your statement is simply incorrect. And we do have good comparison data already.
Hardware Unboxed tested TGL H45 and looked at performance with different TDP settings and boost disabled. This is a missing piece of the puzzle which shows that TGL H really likes more power, as increasing TDP towards desktop levels helps TGL catch up in terms of efficiency.
I dont doubt Zen is more efficient, but 35 watt zen is very close in performance to 45 watt TL, so the difference does not appear to be as outrageous as some claim. Of course, if one wants to make Intel look bad, you can cite the 65w figures, but that is kind of misleading, because 45 w is almost equal in performance to 65 w. And again, I thought it was well established that TDP is not equal to power consumption. Actually, I have a feeling that Intel "fudges" TDP more than AMD, so if actual power consumption were measured, it would make the numbers worse for intel, but who knows until we see actual numbers of real power consumption.
35W Zen is close in performance to 45W TGL and you're unable to see a big efficiency difference there?! You do understand we're talking about a ~30% delta in power?
It has already been mentioned in this thread that the 65W figures are NOT representative for TGL performance on the Intel reference system, since the cooling could not sustain continuous 65W package power.
We're not discussing performance/TDP as defined by both companies for their desktop lines, we're discussing performance/package power consumption expressed in Watts. Take a look again at the CPU Power Scaling graph from Hardware Unboxed, the X axis is clearly defined as such. Both Intel and AMD products have sensor data that accurately reflect package power consumption in real time, accurate enough for this type of comparison anyway.
ark.intel.com
Facebook
Twitter