Question Which is better? More E-cores or more P-cores ?

[FlameTail]

Apple is betting on it's Big cores. The M1 Pro doubles the number of firestorm cores but halves the number of icestorm cores compared to the M1. A future 'M1 Quadra' is rumoured to have 32 Big cores and just 8 small cores. So, yes, Apple is focusing on it's big cores.

Intel on the other hand seems to be betting on it's E-cores( small cores ). Alder Lake has upto 8P+8E. Raptor Lake is rumoured to bump that upto 8P+16E. Arrow Lake will supposedly have 8P+32E ! So intel will be putting in more small cores, which is a complete opposite to Apple's approach.

Which approach do you think is better ? Why ?

Who will win ( performance, efficiency, cost wise ) in the end ?

 

[Hulk]

anyone here a BIG fan of overclocking?

AMD and Intel, especially overclock the crap out of their processors right out of the box. All of the Turbo, Turbo 2.0, Turbo 3.0, etc... is just finding any times that it's possible to ramp up clocks and then they do it automatically.

Back in the day the point of overclocking was to get more value from your processor. The classic example was the Celeron 300A. It came stock at 300MHz via a 66MHz FSB. Move the FSB to 100MHz and everything scaled very nicely as the processor ran at 450MHz. Some even went to 504MHz if you got a good one. At the time the PII 450 cost 5 times a much as the 300A so you can guess why the 300A was so popular. In addition while the 300A only had 128kb L3 it ran at full speed on the CPU unlike the larger slower off-CPU cache of the PII 450 so the 300A>450 was actually faster in many applications.

Anyway, back to my point. Overclocking is really only useful when there is headroom in the CPU, meaning you bought a part that could have been binned higher but wasn't to preserve the pricing structure of the CPU stack. Now that the non K parts are not overclockable we can't get at the additional frequency headroom, and the K parts are pretty much automatically overclocked there isn't as much talk about overclocking these days. That's my take on it anyway.

I could run my 12700K a few hundred MHz higher but I prefer not to have an oven under my desk for a 3-5% performance improvement.

 

[Geddagod]

Both top dogs from Intel and AMD, the i9 12900K and 9 5950X have 16 cores. But half of the 12900K's cores are small cores.

DESPITE THAT, the 12900K beats the 5950X in benchmarks.

Sure, you can argue that the 5950X is last gen and the real rival to the 12900K is the 7950X ( or whatever it is called ).

But then, intel is also at a node disadvantage. That makes me wonder- if both of them were on the same node... the 12900K would absolutely roast the 5950X.

The small cores are no joke.

Very interesting.

Don't think Intel 7 is at a node disadvantage with TSMC 7....

 

[Geddagod]

Well maybe, remains to be seen. Regarding current Golden Cove having more work to done regarding their watt performance, thats why, if i were to buy new CPU right now, i would lean toward AMD.

Merry Christmas to everyone!

Golden Cove might be ineffecienct, but alder lake as a whole (12700k and 12600k) have MT efficiency comparable to zen 3 (except like the 5950x), tho for the 12900k it is definitely a power hog.

 

[BorisTheBlade82]

Exactly. Gracemont efficiency is area efficiency not power efficiency. Virtually no one needs more than eight strong threads (are there even people who need more than eight strong threads at a time).

Nearly always, if not always, using more than eight threads means multi-threaded. With the E cores getting half the performance in one quarter of the space - put otherwise double the performance in the same space as a Golden Cove core - I’m curious if anyone can define a scenario where a ninth or higher number P core is better than adding a four pack of E cores based on pure performance alone. I’ll be waiting to see if someone can pull that off.

Just as I wrote in my post above: It is a question of the general distribution of thread usage for common workloads. Of course there are workloads that would benefit from exactly 9 big cores. But generally it seems that Intel considered 8 + 8 being the sweet spot for their area and power target.