Question E Cores - Useful or useless? What does the data tell us?

[Hulk]

Here are some benches from my 13900K with E cores and P cores set to 4.3GHz.

Cinebench R23
MT
8P with Hyperthreading - 17,525 (94W package)
16E (1 P @0.8GHz) - 18,157 (120W package)

ST
P - 1614
E - 1169

Handbrake bench from our forums
8P with Hyperthreading - 230.47 sec/7.84fps
16E (1 P @0.8GHz) - 219.45 sec/8.23fps

CPUmark99 - Yes old and outdated but the result is curious
P - 683
E - 722
Perhaps something to do with Gracemont's 17 vs. Raptor Coves 12 execution ports on this old single threaded integer benchmark?

For highly threaded applications like Handbrake (x265) and Cinebench rendering, 16E's have about the throughput (IPC) of 8P's. The P's are more energy efficient and can of course clock higher. The E's are more area efficient.

CB ST shows Raptor to have 38% better IPC than Gracemont. Of course Raptor loses it's HT capability here.

CB MT shows Raptor having 93% better IPC than Gracemont with Raptors HT in operation.

As for compute/area, let's assume 8.08mm^2 for one P including L2 and 10.28mm^2 for 1 E cluster with L2
Using CB R23MT at ISO frequencies we find that Raptor Cove generates 271 Cinebench MT points per square mm while Gracemont generates 442 Cinebench MT points per square mm. At ISO frequencies Gracemont provides 63% more compute with Cinebench for a given area. Of course some of this advantage is reduced when Raptor Cove is run at full frequency but the case for Gracemont's area efficiency is significant.

In terms of E and P "balancing" the 13900K is tilted a bit toward the P's. When you take the higher clocks of the P's into account, when using well threaded applications an 8+20 (or perhaps 24) part would provide a pretty even balance of E and P computer. Not that that metric means anything, I just find it interesting.

 

[Hulk]

Why are big companies so dumb?

Because all of the big smart ones declared bankruptcy.

 

[Exist50]

That would be very niche use-case.

Personally I don't really get the whole P and E cores thing. Especially the need for a a gazillion e-cores. Shouldn't like 2-4 very low power cores (like much lower power than what they are now, of course with lower performance) be enough for background tasks? What is actually meant with background tasks? Windows updates? checking email/other apps and notifications? All of these are likley IO (network) limited.

Not fan of Apple but IMHO they got it right for consumer: very beefy, high single-threaded performance cores. But less of them in total. 32E cores won't help applications load faster but a single very beefy core does. So I would see it exactly the opposite. like 8+P cores and maybe 4e cores.
But it all really depends greatly on the OS and schedulers as well. What actually is the defintion of background task? As far as I understand all user-actions go through p-cores.

I don't think these top end chips (13900k, 7950x, etc.) are targeted towards consumers at all. The benefits are essentially negligible vs a tier down. But in professional workloads, all these extra core/threads can actually be put to use. Also, simply as halo products.

 

[burninatortech4]

I suspect that with some tweaking to the frequency there's a point where the E-cores would be more energy efficient as well as more area efficient. I think that Intel could release a chip for people who do have workloads that can utilize a lot of cores that is mainly (or even entirely) E-cores and it would be useful for that market segment.

Isn't there a Xeon roadmap that uses entirely "E cores" in a single monolithic design?

 

[Mopetar]

There are some consumers who have hobbies or workloads that do benefit from these top end CPUs, but most consumers probably don't even need a midrange CPU like a 12600K or a 7600X and would be fine with a lesser CPU.

It's conceivable that both AMD and Intel could release more capable chips at the low end, but they want to give people a reason to move up to a higher tier of product beyond just core count.

 

[Hulk]

I don't think these top end chips (13900k, 7950x, etc.) are targeted towards consumers at all. The benefits are essentially negligible vs a tier down. But in professional workloads, all these extra core/threads can actually be put to use. Also, simply as halo products.

I agree. Also they are simply halo products to pump of the rest of the stack and show how "We're the fastest!"

I remember with clarity when hex cores would absolute top of the line, crazy expensive, monster compute capable parts. Now 8 cores is pretty much considered the normal "lower" end for enthusiasts. I could easily get by with 8 cores. Shoot, I had a 4770K a year and a half ago and was getting stuff done.

This brings up an interesting decision I've been thinking about. As you might remember my 13900K suddenly started needing 120% power settings to run stably. I contacted Intel and they were going to get me a new CPU. After a month of it being out of stock they are going to reimburse me my $600 or so I paid with tax at my local Microcenter. I'm actually considering pocking few hundred dollars and going with a 13700K. The biggest thing stopping me is if Intel is going to make me whole I feel a responsibility to buy the same CPU.

 

[igor_kavinski]

The biggest thing stopping me is if Intel is going to make me whole I feel a responsibility to buy the same CPU.

Do you have to ship the CPU back to them? If not, keep it until you can afford the KS or even the Refresh coming later this year.

 

[coercitiv]

The biggest thing stopping me is if Intel is going to make me whole I feel a responsibility to buy the same CPU.

I would argue it's in Intel's best interest that you go with a lower tier SKU to ensure stability, because there's a non-zero chance the issue is motherboard related (or shared fault between the two). Just keep in mind the 13700K is in a different class when it comes to binning, as you might have already noticed when comparing your 12700K to other people's 12900K.

If you just want to get things done, the 13700K will be great. If you still want to mess around with OC and UV, dicussing results on the forum, then you might want to accept the fact that 13900K is your toy and not your tool, hence buy it again for what it offers you as a hobbyist.

 

[Hulk]

I would argue it's in Intel's best interest that you go with a lower tier SKU to ensure stability, because there's a non-zero chance the issue is motherboard related (or shared fault between the two). Just keep in mind the 13700K is in a different class when it comes to binning, as you might have already noticed when comparing your 12700K to other people's 12900K.

If you just want to get things done, the 13700K will be great. If you still want to mess around with OC and UV, dicussing results on the forum, then you might want to accept the fact that 13900K is your toy and not your tool, hence buy it again for what it offers you as a hobbyist.

Good point. It's a toy and a tool. I'm going to get another 13900K.

Do you have to ship the CPU back to them? If not, keep it until you can afford the KS or even the Refresh coming later this year.

I have to return the CPU before they'll refund the money. Either I'm just going to go with the 13900K or I'll get something really cheap to hold me over until I figure out if I want a bigger toy (ie. refresh or KS).

 

[igor_kavinski]

There's this: https://forums.anandtech.com/threads/intel-12100f-90-shipped.2610603/

 

[Timur Born]

E cores are useful, even if just for putting threads there to keep off P cores and allow better P core boosting. E cores are much better at Integer workload than floating-point. At 43x an E core runs 7-Zip at the *same* power and performance as a P core at 40x.

Curiously when all 4 E cores of a cluster are loaded then frequencies drop slightly for SSE load, drop more for AVX2 load, but do *not* drop for AVX1 load.

 

[Timur Born]

What actually is the defintion of background task? As far as I understand all user-actions go through p-cores.

An "background" process is one that has no window or has its window not in focus (especially minimized). It may be necessary for another process to cover the out-of-focus background window (like being maximized) for this to happen.

All Integer background load is pushed to E cores, while Floating-Point load is *not* pushed to E cores. BUT *all* background processes that set their process priority to low/idle *themself* are pushed to E cores, even when they use floating-point calculations.

This can be disabled for all-core load for all processes by using the "Best Performance" power MODE (in combination with the "Balanced" power PROFILE). Single/low core threads might still be pushed to E cores occasionally even after the power MODE change.

For low/idle FP processes this can be disabled by manually setting the process priority to "Above Normal" or higher. But the latter only works if the process set its own priority to low/idle before.

Someone posted a list of which instructions are handled how (like AVX vs. SSE) for E core pushing, but the above general rules is what I see in practice.

 

[Hulk]

E cores are useful, even if just for putting threads there to keep off P cores and allow better P core boosting. E cores are much better at Integer workload than floating-point. At 43x an E core runs 7-Zip at the *same* power and performance as a P core at 40x.

Curiously when all 4 E cores of a cluster are loaded then frequencies drop slightly for SSE load, drop more for AVX2 load, but do *not* drop for AVX1 load.

I think those 17 execution ports in Gracemont make it extremely capable for certain integer code. Perhaps when this isn't a lot of complex scheduling/cache misses

 

[hemedans]

E core also usefull for lowend, now we have i3 with 10 cores and Pentium with 5 cores.

Pc with cpu like i3 1220P can be found around $300 with perfomance to Rival Apple M1.

 

[Timur Born]

13900K E-Core power & frequency: 16x P95 SFFT on 16 E-cores

SSE: 121 W / 4257 - 4282 MHz
AVX1: 117 W / 4301 MHz
AVX2: 129 W / 4222 - 4232 MHz

So while E cores are generally bad at floating-point operations, they seem to prefer AVX1 load over SSE (suprising) and AVX2 (unsurprising).

 

[itsmydamnation]

13900K E-Core power & frequency: 16x P95 SFFT on 16 E-cores

SSE: 121 W / 4257 - 4282 MHz
AVX1: 117 W / 4301 MHz
AVX2: 129 W / 4222 - 4232 MHz

So while E cores are generally bad at floating-point operations, they seem to prefer AVX1 load over SSE (suprising) and AVX2 (unsurprising).

what do you mean by avx1 , avx2 because avx2 mainly added integer SIMD instructions, i would not expect any real difference in power consumption between the two.

 

[Timur Born]

what do you mean by avx1 , avx2 because avx2 mainly added integer SIMD instructions, i would not expect any real difference in power consumption between the two.

 

[itsmydamnation]

View attachment 77037

ok makes sense , FMA isnt apart of AVX2 , was just added to the intel core at the same time ( haswell)
for example AMD had FMA on bulldozer but not AVX2

 

[Timur Born]

OCCT v12 benchmark results for single-core:

P core SSE 116
P core AVX 216

E core SSE 65
E core AVX 78

 

[Roland00Address]

E core also usefull for lowend, now we have i3 with 10 cores and Pentium with 5 cores.

Pc with cpu like i3 1220P can be found around $300 with perfomance to Rival Apple M1.

This can not be true…[looks it up]. Wow I am impressed. I will retract those words from 1 sentence earlier.

and it is curious the i3-1220p 28w vs the i5-1230U 09w, same amount of cores, same amount of maximum ghz on the efficiency and performance cores, yet the 28w chip is only 30 to 50% faster than the 09w Intel. The M1 is still faster than both chips but like hemedans said it is very close in synthetic stuff that compares cross platforms.

Interesting 🤔