The story about cpu cores

[MightyNerdy9]

Hey , I want to educate and learn more about it , 

 

Intel has E cores and P cores , are the E cores more slower so its not technically "helps" , although amd has only one kids of cores and all of that are matters , 

 

sorry if i mixed here or made a huge mistake , 

 

thanks guys

 

[emosun]

supposedly they save power or something but idk why they can't just be all the same either.

feels more like they just demand advancement regardless of the usefulness

[RONOTHAN##]

Intel figures that for the same die area, they can fit 4 E Cores in the space of 1 P core, and since the E Cores are about 30% the performance of the P cores you effectively get 120% the performance for the same speed, assuming the workload handles it properly. You still get a couple P cores for the single threaded workloads, though since most of those don't scale past 6-8 Cores you only get 6-8 of them. 

 

At least that's the theory behind it, in practice they do cause some issues because software compatibility is still very much hit or miss. The way it came about was more likely to copy ARM based CPUs to get better efficiency, but since the efficiency cores ended up not being that efficient, my guess is the only reason they kept with it was the above reasoning. 

 

AMD has decided that the drawbacks of them aren't worth the benefits, so they're sticking with homogeneous CPUs for the time being. They are rumored to be researching a hybrid CPU though, so both might be hybrids in the next year or two

[8tg]

Imagine instead of having your i7 with its bunch of cores, you have your i7 and alongside it is an intel atom which also has a bunch of cores.

So all the unimportant stuff can go be handled by the intel atom, and your i7 can do all the important stuff.

 

Thats basically what it is. Its a matter of operating systems and applications properly using CPU affinity and priority on specific cores, but E cores are more like an entire extra low power CPU on the primary CPU that handles all the little stuff to take load off the P cores.

[Alex Atkin UK]

9 minutes ago, emosun said:

supposedly they save power or something but idk why they can't just be all the same either.

feels more like they just demand advancement regardless of the usefulness

They are not as fast but they can fit more of them in the same space.  So the CPU overall ends up being faster in heavily threaded workloads.

 

Its a similar concept to how CPUs like the Threadripper work, they have lots of slower-clocked cores so the combined performance is better than having less higher clocked cores, but it depends on very specific workloads.

 

Games can swing both ways, which is what Intel count on with the mixed core types.  Some games will run better on P cores, some will benefit from the E cores too.  Its also partly that Intel P cores are insanely inefficient for power consumption (compared to AMD), so they are using E cores when your PC is idle to bring the power consumption down.

[MightyNerdy9]

I have learned from a friend that games for example , don't use more than 4 cores 

 

so if i want to compare for example : 14700kf which has as 20 cores hybrid and the 7900x which has one kind of 12 cores 

 

on which one should i target if i want to run multitask last games , streaming for example ? 

 

and also regarding power consumption , the 7900x takes 170w and the 14700kf takes 125w 

[emosun]

7 minutes ago, Alex Atkin UK said:

They are not as fast but they can fit more of them in the same space.  So the CPU overall ends up being faster in heavily threaded workloads.

 

Its a similar concept to how CPUs like the Threadripper work, they have lots of slower-clocked cores so the combined performance is better than having less higher clocked cores, but it depends on very specific workloads.

 

Games can swing both ways, which is what Intel count on with the mixed core types.  Some games will run better on P cores, some will benefit from the E cores too.  Its also partly that Intel P cores are insanely inefficient for power consumption (compared to AMD), so they are using E cores when your PC is idle to bring the power consumption down.

sounds great if it actually worked , reminds me of them trying cylinder shut of systems in the early 80s which also would be neat if it worked.

[emosun]

1 minute ago, MightyNerdy9 said:

I have learned from a friend that games for example , don't use more than 4 cores 

is your friend from 2011 becuase that used to be true a decade ago

[da na]

17 minutes ago, MightyNerdy9 said:

I have learned from a friend that games for example , don't use more than 4 cores 

 

so if i want to compare for example : 14700kf which has as 20 cores hybrid and the 7900x which has one kind of 12 cores 

 

on which one should i target if i want to run multitask last games , streaming for example ? 

 

and also regarding power consumption , the 7900x takes 170w and the 14700kf takes 125w 

That's true for some games, in fact some games don't use more than a single core. Luckily times are changing though and newer games are able to more fully utilize the capabilities of modern multicore and hyperthreaded processors.

[Alex Atkin UK]

8 hours ago, emosun said:

sounds great if it actually worked , reminds me of them trying cylinder shut of systems in the early 80s which also would be neat if it worked.

The fact they are planning to release a custom scheduler that will work with a dozen games are so just makes it all the more hilarious.

 

This is why I avoided the 7950X3D too.  I don't trust any company to get this right for gaming and with Intel especially its obviously a desperate attempt to squeeze the last drop out of their terribly power inefficient cores. 

 

Even the E cores aren't great, look how the Celeron N100 (4 efficiency cores, no P cores) compares to a Steam Deck (4 full fat Zen 2 cores). https://browser.geekbench.com/v6/cpu/compare/2803629?baseline=3890185

 

The N100 is using about twice as much power to achieve this.

[starsmine]

9 hours ago, emosun said:

supposedly they save power or something but idk why they can't just be all the same either.

feels more like they just demand advancement regardless of the usefulness

4 cores fit in the same footprint as one P core and per core use much less wattage.

For anything that benifits from N threads, 4 e cores > 1 SMT P core at the same wattage.
for a select amount of 1 thread tasks, the e-core's total watt hour to complete a task is less on the e-core, even if it takes longer due to using much less wattage per thread. 

So select 1 thread tasks that are not latency dependent and tasks that are N threaded, e cores wins. 

It does work. 

 

8 hours ago, emosun said:

sounds great if it actually worked , reminds me of them trying cylinder shut of systems in the early 80s which also would be neat if it worked.

It does work. First attempts at it in the 80s did not get great results but you see it getting amazing results in modern LS engines. 

Crysler, Mazda, GM, Honda, and Ford all do it with new cars you buy today. 

 

  

46 minutes ago, Alex Atkin UK said:

Even the E cores aren't great, look how the Celeron N100 (4 efficiency cores, no P cores) compares to a Steam Deck (4 full fat Zen 2 cores). https://browser.geekbench.com/v6/cpu/compare/2803629?baseline=3890185

 

The N100 is using about twice as much power to achieve this.

Where are you seeing total watts used on that bench?

[emosun]

32 minutes ago, starsmine said:

First attempts at it in the 80s did not get great results but you see it getting amazing results in modern LS engines. 

darn now if only that was what i typed huh

 

33 minutes ago, starsmine said:

4 cores fit in the same footprint as one P core and per core use much less wattage.

For anything that benifits from N threads, 4 e cores > 1 SMT P core at the same wattage.
for a select amount of 1 thread tasks, the e-core's total watt hour to complete a task is less on the e-core, even if it takes longer due to using much less wattage per thread. 

So select 1 thread tasks that are not latency dependent and tasks that are N threaded, e cores wins. 

It does work. 

yeah sounds super simple can't wait for somebody to make a program for that

[MightyNerdy9]

so this discussion made me go mental lol 

 

for multitasking , i should go with intel or amd  ? 

[starsmine]

1 hour ago, MightyNerdy9 said:

so this discussion made me go mental lol 

 

for multitasking , i should go with intel or amd  ? 

look at the benchmark for your multitasking task

6 of one half a dozen of the other, AMD is probably the better buy at the current moment for most things, but nothing wrong with intel. 

None of what I said should have influenced you one way or the other.

8 hours ago, emosun said:

darn now if only that was what i typed huh

It kinda was. 

[Alex Atkin UK]

11 hours ago, starsmine said:

 Where are you seeing total watts used on that bench?

They're my benches though I admit I didn't check at the time what the power draw was when I tested the N100 and can't now as its my router.  However it did trip my 12V 3A PoE adapter when I tried to use it, suggesting that when it boosts it may be over 36W (or my PoE splitter is faulty, but it never tripped on the N5105 with its TDP unlocked).

The Steam Deck on the other hand I just re-ran the test and it peaked at 15W single-thread, 20-25W multi-threaded (from the wall).

 

So I think its fair to guesstimate the Steam Deck is at least 33% more efficient (granted its the OLED model and slightly undervolted).  It idles at 5-6W on the desktop and the one time I got the N100 to boot with the PoE adapter I think it was more like 11W idle, though not a fair comparison as pfSense is not really designed to idle as hard for obvious reasons.

 

Point is that Zen 2 is far more efficient than what Intel are peddling, and we have Zen 3 now which is even better.  Desktop Intel right now has to push insane watts to beat the 7800X3D for example, and in many games that's still not enough to outperform it.

 

I was also a little disappointed to find replacing my 8600k with a 12400 consumed about the same amount of the power at idle, guess I should have gotten a model with E cores after all.

[starsmine]

34 minutes ago, Alex Atkin UK said:

They're my benches though I admit I didn't check at the time what the power draw was when I tested the N100 and can't now as its my router.  However it did trip my 12V 3A PoE adapter when I tried to use it, suggesting that when it boosts it may be over 36W (or my PoE splitter is faulty, but it never tripped on the N5105 with its TDP unlocked).

The Steam Deck on the other hand I just re-ran the test and it peaked at 15W single-thread, 20-25W multi-threaded (from the wall).

 

So I think its fair to guesstimate the Steam Deck is at least 33% more efficient (granted its the OLED model and slightly undervolted).  It idles at 5-6W on the desktop and the one time I got the N100 to boot with the PoE adapter I think it was more like 11W idle, though not a fair comparison as pfSense is not really designed to idle as hard for obvious reasons.

 

Point is that Zen 2 is far more efficient than what Intel are peddling, and we have Zen 3 now which is even better.  Desktop Intel right now has to push insane watts to beat the 7800X3D for example, and in many games that's still not enough to outperform it.

 

I was also a little disappointed to find replacing my 8600k with a 12400 consumed about the same amount of the power at idle, guess I should have gotten a model with E cores after all.

I want to caution. Maximum wattage is a poor measure of power consumption of a part when look at efficiency unless you are power limited in some way. 

For gaming or video consumption its 100% valid because you are pulling that wattage for the entire period of time. But for a task like say, encoding video, the total time matters. Pulling 300w for 30 seconds is more efficient than pulling 40w for 6 minutes.
But if that 40W system does it in 3 minuits, its more efficient. 

I also caution for the specific conversation that this thread is addressing comparing E cores with zen at all, if the question is Why E cores vs P cores. 

I wont say intel at the moment more efficient then zen. Although you can force configurations where that is true, its not really the topic. 

But I want to bring your attention to the the 8 P core intel parts at the bottom of the chart. Every time you added E cores, the Wh to complete a task went down. 

If someone can benchmark a Single SMT P core vs 4 E cores that would take away all the noise, but I don't know of anyone who has done such a benchmark. The fact that there are 8 P cores and those chips are all clocked at their most inefficient frequency/voltage ranges just muddies the comparisons.  Your N100 might be something interesting to compare with a bios locked to single core i3 if you want to do it. 

[Alex Atkin UK]

I certainly wasn't arguing that the split cores doesn't work, it clearly does.  But when you compare it to AMD, as you see in the chart you posted, its a bit of a bloodbath.

 

There's zero chance of doing any testing on the N100, its my router.  I just always try to give a pass of Geekbench on any PC I get my hands on before putting it into use, just for comparisons sake.  It just always seems to coincide with me losing my power meter.

[Recharged]

It's clearly not a great CPU strategy, as it consumes more power than 2 XBOX Series X consoles combined.

 

 

[starsmine]

21 minutes ago, Recharged said:

-snip-

 

its almost like you ignored everything above you. 

300W for 30 seconds is more efficient than 100W for 2 minutes for one. (150 watt minutes vs 200 watt minutes)

Running a processor outside its efficiency band will also mean you are... running out of your efficiency band. 

[Recharged]

Just now, starsmine said:

its almost like you ignored everything above you. 

300W for 30 seconds is more efficient than 40W for 5 minutes.

Running a processor outside its efficiency band will also mean you are... running out of your efficiency band. 

First the 9600k, now the 13900k, it's as if I hurt your baby?

[starsmine]

18 minutes ago, Recharged said:

First the 9600k, now the 13900k, it's as if I hurt your baby?

I kindly ask you to read the threads you post in and stop spreading misinformation in this and in other threads. 
This thread isn't even a page long, you will find all the context necessary in here that already disputes or discredits what you have said here and in the post you made about the 9600k. 

The misinformation isn't the total wattage draw of the parts at a full load that part is correct, its your framing. It's the relevance of that point is questionable when discussing e-cores. 
Sure, intel could have made a 12 P-core/24 thread chip instead if the i9 with its 8P/16E 32threaded version, but it would perform significantly worse at the same wattage in n-threaded tasks. And the Wh to complete a task would skyrocket with minimal changes in gaming performance. 

To make intel chips more efficient, you dont turbo clock them out of their efficiency range, remember to get those last few hundreds of mhz you have to increase voltage a lot, and power is voltage^2. 

[Recharged]

Just now, starsmine said:

I kindly ask you to read the threads you post in and stop spreading misinformation in this and in other threads. 
This thread isn't even a page long, you will find all the context necessary to that already disputes or discredits what you have said here and in the post you made about the 9600k. 

The misinformation isn't the total wattage draw of the parts at a full load, its more so that the relevance of that point is questionable when discussing e-cores. 
Sure, intel could have made a 12 P-core/24 thread chip instead if the i9 with its 8P/16E 32threaded version, but it would perform significantly worse at the same wattage in n-threaded tasks. And the Wh to complete a task would skyrocket with minimal changes in gaming performance. 

To make intel chips more efficient, you dont turbo clock them out of their efficiency range, remember to get those last few hundreds of mhz you have to increase voltage a lot, and power is voltage^2. 

It uses more power and the 9600 K stutters like ass in many games.

 

I am not joining your religion.

[problemsolver]

These two Intel articles explain it decently well. Both are echoed in previous posters' comments.

• How Intel® Core™ Processors Work
• https://www.intel.com/content/www/us/en/gaming/resources/how-hybrid-design-works.html

 

Fundamentally, it's a different design approach to a processor. The concept/principles are not new. Almost everyone using a smart phone is using a processor using similar design principles based on ARM's big.LITTLE architecture.

 

Intel's processor design has more flexible/advanced software for determing where work should go (Thread Director), than ARM's big.LITTLE architecture. But ARM has also adapted with DynamicIQ.

 

All that being said, there are advantages and disadvantages to any architecture, which is why we have so many different ones. The P & E core Intel processors likely generally have advantages when using them for mixed workloads because the E cores can keep background tasks out of the way of the P cores, and higher efficiency means less power which means less heat, which means higher clock speeds for a given cooler. If you're only doing one specific workload, then having a purpose-built processor is the way to go (hence ASIC crypto miners etc.) But, as others have pointed out, the software-side can be more complex for processors with P & E cores. Overall, I think we'll see these hybrid architectures become the standard processor design in the coming years, and as such the supply chain and toolchains, software, compilers, operating systems etc. will continue to be further optimized for such hardware.

 

Global climate change government regulations may also lead to more complicated CPU designs to ensure maximum efficiency on desktop computers, like we've seen in the Internal Combustion Engine (ICE). (Although the server-space is where performance/per watt has the most impact when it comes to power conservation efforts.)

[Alex Atkin UK]

Good timing:

 

Interesting results, as his conclusion suggests that using Intel for my NAS/Server and router was probably the better option. 

 

I should have known that actually as I noticed my 5950X box the SoC never drops under 20W according to Ryzen Master so like he said resulting in the IO die consuming more than the CPU chiplets at idle.

[starsmine]

2 hours ago, Alex Atkin UK said:

Good timing:

 

Uncanny timing
My only complaint is the focus on gaming, but that was the question he was trying to answer so thats not that fair of a complaint. 

once you get out of gaming the whole efficiency curve things shows up that I talked about. (VF curve, voltage/frequency)
put the 14700k in its efficiency curve and its amazing, But as Steve said, this isnt a use case anyone who buys a K chip would use. This is a usecase Businesses may use if they buy a fleet of T suffixed chips. 

like these https://www.intel.com/content/www/us/en/products/sku/230492/intel-core-i713700t-processor-30m-cache-up-to-4-90-ghz/specifications.html

What muddies the water with Intel K parts is that they automatically do what we used to do to CPUs in the past with turbos. We would take a 95W sandy bridge running at 3.8ghz and send it to 150W+ at 4.7ghz. Socket 1700 CPUs with the k suffix just do that with their infinite turbos on their own.

AMD parts do the same with PBO2 on sure, and for the use case of a desktop at home not in a fleet, It generally is the correct thing to do. 


It just makes it harder to talk about how efficient any part of an intel chip is because it overclocks for you off to the deep end of the vf curve as those parts are built to do just that and if you want efficiency you look at a different part of the product stack.