The Intel 13th Generation Core Raptor Lake-S range leaks 4 to 24 cores on three separate dies.

Wow 16 e cores. What could anyone possibly want with that many? Certainly they won’t be useful for much.

 

What would anyone want 8 P cores for? Especially for consumers who spend all day gaming and consuming media, 2 P-cores is plenty. P cores are a waste of energy and die space, in fact it’s too bad that Intel doesn’t make an all E-core processor.

 

e-core could be great for running OS and background task, while p-cores run games and programs that need the power, in perfectly setup world... last check scheduling and all that is still mess, mutli core support is still some what iffy past 6 core? then again these e-core/p-core cpu seem to use MORE power at idle then traditional cores? atlest last i looked

i7 13700/f look interesting 65w tdp, as i rather go down in TDP then up from 6700k tdp stand point

 

Those 16 E cores will do wonders for my chess engines. I'm happy with that.

 

but wouldn't it be more useful to have say 4-8 E and 10 to 12P?

 

they are good for things like rendering and any multi-threaded application, It makes sense they would go heavy on the E cores because they can get more performance per mm^2 with a lower power consumption. If intel were to put 16 P cores on a die , You would be looking at a monsterous tdp under any heavily threaded workload avx workload, for instance at 5ghz 16 p cores would draw >400w package power ( this is extrapolated from the power consumption of a 12400), so it would be throttled frequency wise to the point where the multi-threaded performance is likely lower vs a design with the e-cores.

 

That configuration makes no sense for any workload including gaming. Lightly threaded workloads like gaming require 2-8 cores at most. Heavily threaded workloads like 3D rendering, video editing, audio processing, and so forth are well parallelized across many low-power cores.

For example, workstation and server processors (EPYC, Xeon Scalable) are composed of many "E-cores" that usually consume less than 5 watts per core. These cores are usually in the >2 base or <4 GHz boost clock range. For content creation and compute, more "E-cores" (meaning low clocked, low power cores that are more efficient for numerous reasons) is the best solution.

The phenomenon of 5 GHz+ processors is exclusively to cater to high-refresh rate gaming, and is frankly not useful for any other workload.

 

I don`t understand why Intel won`t release CPUs only with P-cores for enthusiasts, it would be much better.

 

No, what for? I agree 4P cores and 16-24E cores would be the best. Intel's E cores are just 20-25% slower than their P cores, which are already 10-15% faster than Zen4 cores and 25-30% faster than Zen3 cores. But 4x E-Cores take the same die space as 1x P-Core and consume less power than 2x P-Cores.

 

No they are not just 25% slower than a P-core.
More like 3 times slower.

 

I like 4 e-cores on 13400 actually. pair that with more cache and a better imc (4133 ddr4 in 1:1 mode at least ) and sub 200eur price and that's going to be a clear price/perfomance winner.

just lol,I have no other words for comparing zen3 cores to intel's e-cores

 

Too much power consumption, Its already creeping into HEDT territory. If we're gonna see high core count Pcore only parts, its going to be using their server chips on a proper HEDT platform, but so far intel has been so uncompetitive in this arena that they essentially haven't released any new Workstation or enthusiast HEDT parts to retail since the placeholder w-3175x and cascadelake refresh of lga 2066. lga 4189 workstation theoretically exists but you can't get them anywhere afaik.

 

That depends entirely on what you are doing. For the very homogenous workloads that people seem to like to benchmark (but rarely actually do in real life) - like rendering and ray tracing - they are pretty good.

Real life workloads don't mimic that though. However, for many years the consumers and these review sites have focused on those types of benchmarks (cinebench, pov-ray, blender, etc) so this is what we get.

 

People who dismiss the usefulness of E-cores remind me a lot of people who blindly follow political parties. There's such overwhelming evidence of how an all P-core mainstream CPU (from Intel, but eventually from AMD too) is a bad idea, yet people want an all P-core desktop CPU as if they actually need the extra instructions and clock speed on every single core for everyday use. The naysayers just see how the E-cores run worse in highly parallel benchmarks, but how often are you ever running those? If your CPU makes you money, you're going to want to get a legit workstation CPU or even a server CPU. Sure, initial benchmarks weren't looking so good because the Windows scheduler didn't know how to properly organize threads, but as far as I understand, that has mostly been resolved.

However... I do think it's a little odd for them to include 16 E-cores - that's excessive and a point of diminishing returns. E-cores are best for background processes, light foreground tasks, and child threads that aren't used for parallelization (for example, the thread transmitting instructions to the GPU). E-cores are ideal for anything that doesn't need to run as fast as possible, which is the vast majority of processes. Unless there's some major paradigm shift in how computers work, I suspect 8 E-cores (in a desktop CPU) is all we will ever need. You want enough cores where there's no chance that any thread will be waiting for availability, but who is ever going to need 16?

So yeah, I don't really see this particular CPU selling well because it seems to have an identity crisis. The 13700F seems to be the most sensible of the high end chips - it will be significantly cheaper, there won't be much overclocking headroom without splurging on a hefty PSU and cooler, you probably will never use the iGPU, and the extra 8 E-cores are will either be underwhelming with highly-parallel workloads or otherwise sitting idly the whole time.
Though for me personally if I end up going with Intel (not super likely), I'd rather get a 13500F or 13600F.

 

More like 50% slower on average, never slower than Sandy Bridge and in some cases on par with ZEN2.
If you look at the bottom of the page 7 games average, E cores are 50% slower than P cores

Intel 12th-Gen Core Alder Lake Architectural Benchmark | TechSpot

 

An E-core is half the speed of a P-core by any metric, and is the same speed or faster than a Skylake core (1 E-core = 1 i7 6700K core). No “real life consumer workload” needs more horsepower than an i7 6700K.

This is why E-cores are such an appealing proposition: consumers think they need a lot of CPU power for mundane tasks like web browsing, word processing, and so forth when in reality much lower power cores do these tasks just as well.

You’re correct that E-cores are great for background tasks, but you miss the fact that they are even better for highly parallel tasks. For tasks like rendering, audio processing, video encoding and so forth there is no “point of diminishing returns” because the task is divided and distributed across all cores.

E-cores make a lot of sense in this area because Intel has largely abandoned the bespoke HEDT segment, and even if they hadn’t, processors like the Xeon W 3175X were prohibitively expensive. Motherboards, ECC memory, etc for high end Xeon W ran into the thousands for dollars. If you wanted 28 cores then, it would run you at least $5,000. Yet, now you will be able to get 24 cores on a 13900K which will perform well in highly parallel tasks due to the high number of E-cores.

 

If you only look at averages, and in games that don't push the CPU a lot.
But in games that hit the CPU a bit harder, the difference in averages can be double Like in BFV.
But look at 1% low. That is when those CPUs are stressed and pushed to do their best.
Again, look at BFV. With only 4 P-Cores we have from 130 fps. But with the E-cores, we go down to only 48 fps.



Or with Hitman, where it goes from 116 fps, down to 38.

 

That is 100% wrong. Didn't need to read anything else there you posted as it is all founded on that fallacy.

Overall a P-core is 52% faster. But within specific sub-tests, the e-cores can be very potent. One of those use cases as I mentioned is rendering.

Notice that clock for clock, the e-cores are only about 17% slower than a p-core, and in fact 8 E-cores is faster than a 12 thread 3600X.

 

I see you didn't read my comment, or the benchmark you cited, I'm pro-E-core . You've cherry-picked a single benchmark that intentionally crippled the P-cores to make the same point I was making. Of course, if you read more than a single cherry-picked benchmark, you'd see that E-cores are about half the speed of P-cores, and that's a good thing!

You say that "an E-core is half the speed of a P-core by any metric" is a fallacy then say "overall a P-core is 52% faster" EXCEPT for in benchmarks that intentionally cripple the P-cores .



Of course, in the real world where people don't run their processor intentionally crippled, the performance discrepancy in that exact same benchmark nears 50% .

Aside from your literal fallacy I am still all for E-cores, so you don't have to try to convince me.