Bzuco@londiste
Yeah, in mainstream AL(if it had only P cores) will beat ryzen 5800x. But that probably will be not enough for 5900x and as we saw from leaks, it will definitely not beat 5950x. Therefor we see E-cores as addition.
Beat in my words mean sum of (what perf. + at what consumption + for what price).
So for the intel future chiplet design will be necessary to beat AMD 2/3/4 chiplet designs on one die, otherwise one of the metrics in the sum I mentioned will definitelly fail.

That I do not argue with, Intel will need chiplets to compete with bigger AMD CPUs.
At the same time I do not see 12/16 core CPUs really mainstream for now (and not really counting ADL E-cores into that). And because of that, mainstream desktop segment could and should be monolithic, at least in comparison to current chiplet designs.

blanarahulFinally someone who speaks sense instead of just parroting that E cores are there for saving power while idling.

Guys, if the improvements were that significant, Intel and AMD would have done it long before ARM. They are engineering firms. Why would they not use an idea if it made engineering sense? Do you think AMD and Intel like that their laptops don't have better battery life?

My 10500H consumes about 2 watt at idle according to ThrottleStop. If it had 8 E cores I won't be saving more than a watt of power at best.

You guys seem to misunderstand why Apple is kicking so much butt with M1. Their big cores are just that ridiculously power efficient. They consume 50+% less power for the same level of performance.

Think about it, M1 Pro/Max have only 2 E cores, yet the new Macbooks claim 17/21 hour battery life.

Frankly this is why I wished Apple that M1 Pro/Max had only 8 P cores instead of 8 P + 2 E cores. Have a fair fight with the 11980HKs and 5980HSs of the world and let the reviewers/journalists decide who is more power efficient.

My 5600X consumes 28W when idle. Idle meaning Windows with background stuff and some always on apps like Steam, Skype, etc running but their windows closed. Once I leave a browser open it will be more. 5600G in the same situation consumes maybe 16W (and my old i5-8400 consumed 5W but that is besides the point).

Apple M1 cores are power efficient but not THAT power efficient. These are running at low clocks, low voltage. Underclock and undervolt a Cezanne to same range and it will be quite competitive, even with the node disadvantage (Cezanne at 7nm vs M1 at 5nm). Also, M1 is huge. 5700G/5800U have 60% less transistors. M1 16B transistors vs Cezanne 10B transistors and that is with M1's 4P+4E (to use Intel's designation) against Cezanne's full 8P cores. In addition to smaller manufacturing process Apple also has the advantage of vertical integration so they can use whatever ISA they want however they want and optimize for it.

As for why Intel or AMD has not done this yet - software. You can read about Windows 11 and scheduler optimization for ADL and probably will be able to read about that for months or year(s) to come. This is not a trivial problem especially on higher performance levels of desktop or server market.

blanarahulGuys, if the improvements were that significant, Intel and AMD would have done it long before ARM. They are engineering firms. Why would they not use an idea if it made engineering sense? Do you think AMD and Intel like that their laptops don't have better battery life?

Sorry, but it's FAR more complicated than that.

lexluthermiesterSorry, but it's FAR more complicated than that.

Such as... M1 cannot run the (legacy or not) x86 applications out there in the wild. Recompilation may be trivial if the source code still exists, but optimizing it for modern ARM is going to be another challenge.

mouacykSuch as... M1 cannot run the (legacy or not) x86 applications out there in the wild. Recompilation may be trivial if the source code still exists, but optimizing it for modern ARM is going to be another challenge.

For example...

napataE-cores are necessary for Intel to compete with AMD on multicore performance. The problem is there's also a ton of workloads that don't scale well with cores so you also need P-cores. If Intel was only interested in pure multicore CPUs then they'd probably go E-core all the way as they achieve much better results than P-cores both in terms of performance/W as performance/die. Most people don't seem to understand the Big.little approach and why it exists. The idea that marketing determines hardware and not the other way around is laughable. Of course for gaming E-cores are mostly worthless.

Atleast from the leaks it seems like a succesful approach. Intel increased multicore performance by 60-70% in a single generation. I doubt they could've reached 5950x MC performance wihout E-cores. There's no way 2 P-cores would outperform 8 E-cores. Most likely AMD will do the same thing in the future as it's the most efficient way to increase multicore performance. AMD can just delay it because their cores are much better in terms of performance/W than Intel's cores.

Compete with AMD on multicore performance.... where? For what market segment or use case? Server?

For the MSDT stack, 8 physical cores was already enough, maybe 10-12. Intel has them. They perform just fine, albeit at high TDP. As someone else stated, you do get more cores for the same TDP at AMD, and its also high. But really, the market for Ryzen 9 on MSDT? This is not where Intel gets the revenue.

You said it right. There is no E-core use case for gaming. So let's summarize the real world advantage for Intel here.
- The vast majority of MSDT-aimed use cases don't have any advantage.
- The core count on Intel products apparently goes up.
- The larger half of the ADL stack doesn't even use E cores.
- ONLY the SKUs that have E cores on desktop, benefit from a marginally lower idle usage and Intel can divert some margin of power headroom to the P core. Benchmark win. Yay. Real world use case? There is none. If you truly have a heavy multicore load you'd also need instructions to go with it. And if its simpler but still parallel, the thread count matters as much as frequency. So maybe you can transcode something a little bit quicker, by virtue of a double core count :p Worth...
- Any basic home PC doesn't even remotely know or care about all these developments and is fine with a 5 year older CPU, 2015 quad core would do the trick fine even. And here's the kicker... those already had 65W TDP too and idled well.

- On mobile, the same use cases exist, minus the heaviest desktop ones. Added use of more E cores alongside P cores? I don't see it, apart from, again, marginal power usage advantages. But laptops barely even idle more than desktops, if at all, because they just hibernate.

The competition here happens on IPC and mostly on benchmarks. The benefit of E cores is pure marketing: core count wars and synthetic wins. Its a bit like an engine that can rev in the red to show super high RPM, but it really can't drive continuously like that. And in that sense, its just Intel cranking out even more bursty CPUs for consumer, what's new? The real victory long term isn't big little. Its IPC increases, as it always has been, because the single core perf is pivotal for everything else. Not the way cores are set up, or split up... AMD has experienced a page of that black history too with FX.


Sure, its nice to believe in the idea that Intel is 'back' and 'competition is really on now', but I strongly doubt Intel has a design win here with the big little concept on x86. They can eek out a few % but the biggest win is the marketing win, much like the Megapixel, the Ghz, the Watts and all those other numbers Joe Simple won't understand, but higher is better. All I see is more CPUs nobody really needs, and exaggerated price points to go with it.

Vayra86There is no E-core use case for gaming.

Why not? Taking any OS activity or background stuff off of P cores can easily be a measurable bonus.

londisteWhy not? Taking any OS activity or background stuff off of P cores can easily be a measurable bonus.

Because you don't need top end CPU performance for gaming, and core count wasn't an issue prior to E cores either, or related to it.

Newer APIs have enabled the use of more threads as well, but they'll still run on P cores anyway. Enough is just simply enough and more won't do a thing.