Question Question about Intel CPUs with P Cores and E Cores

Mar 24, 2023
1
2
15
Hello everyone!
Frist time posting in this forum. So, here’s the question I have about Intel CPUs that heterogenous cores aka bigLITTLE, aka Efficient and Performance cores.

1st I don’t understand the point of having E cores and P cores on a desktop CPU. I understand why you would have that setup on a laptop, mobile device or any device that may have heat or power issues, but one the desktop it isn’t necessary.

2nd let’s say that it does make sense on a desktop as well; and we want to be responsible citizens and conserve energy. That’s great, and I applaud that; but why so many Efficient cores?
The whole idea is the efficient cores will get you by on low impact or no activity, and when you need more power the Performance cores kick in and do all the hard work. So in my view you should only have at most 4 efficient cores (I actually would prefer 2 E cores) and the rest all performance cores.

Now I have a theory on why Intel is doing this, and in my opinion it’s very nefarious and shady. Intel (well most computer hardware and software manufactures) always hated users that used consumer grade equipment and used it for business class work. Intel especially hated that user would use an i5 or i7 for a Server application and not buy the over priced (yet same product) Xeon CPUs. So now they are gimping the consumer grade CPUs with a bunch of E cores (and few P cores) to force users to purchase a Server or Workstation class CPU. One of the top-of-the-line Raptor Lake processors (Intel® Core™ i9-13900KS) has 8 P cores and 16 E cores! Why do you need 16 E cores? And because of this they sell it as a 24 Core CPU.

Remember Intel always thought that the average consumer doesn’t need more that 4 Cores. (By the way I’m not looking at threads, just physical cores.) One last thing; I wonder that how a Raptor Lake CPU would hold up to a true homogenous core CPU like the Ryzen 7950X, in really heavy workloads like a Workstation or Server environment.

Please let me know what you think; I am open to being completely wrong about this.
 

jnjnilson6

Distinguished
Hello everyone!
Frist time posting in this forum. So, here’s the question I have about Intel CPUs that heterogenous cores aka bigLITTLE, aka Efficient and Performance cores.

1st I don’t understand the point of having E cores and P cores on a desktop CPU. I understand why you would have that setup on a laptop, mobile device or any device that may have heat or power issues, but one the desktop it isn’t necessary.

2nd let’s say that it does make sense on a desktop as well; and we want to be responsible citizens and conserve energy. That’s great, and I applaud that; but why so many Efficient cores?
The whole idea is the efficient cores will get you by on low impact or no activity, and when you need more power the Performance cores kick in and do all the hard work. So in my view you should only have at most 4 efficient cores (I actually would prefer 2 E cores) and the rest all performance cores.

Now I have a theory on why Intel is doing this, and in my opinion it’s very nefarious and shady. Intel (well most computer hardware and software manufactures) always hated users that used consumer grade equipment and used it for business class work. Intel especially hated that user would use an i5 or i7 for a Server application and not buy the over priced (yet same product) Xeon CPUs. So now they are gimping the consumer grade CPUs with a bunch of E cores (and few P cores) to force users to purchase a Server or Workstation class CPU. One of the top-of-the-line Raptor Lake processors (Intel® Core™ i9-13900KS) has 8 P cores and 16 E cores! Why do you need 16 E cores? And because of this they sell it as a 24 Core CPU.

Remember Intel always thought that the average consumer doesn’t need more that 4 Cores. (By the way I’m not looking at threads, just physical cores.) One last thing; I wonder that how a Raptor Lake CPU would hold up to a true homogenous core CPU like the Ryzen 7950X, in really heavy workloads like a Workstation or Server environment.

Please let me know what you think; I am open to being completely wrong about this.
E Cores are actually quite powerful by themselves. Even if the processors only had E Cores, they would still perform nicely and powerfully. Intel are adding up E Cores so that the total Power Usage and heat dispersion of said generations of CPUs are lowered down, while at the same time the number of Cores is high. If we had 24 P Cores instead of 8 P and 16 E Cores, the power draw and temperatures would increase dramatically, so the P cores would have to be downclocked in order that the processor outputs the same heat and intakes the same amount of electricity as in the scenario with 8 P Cores at a much higher frequency and 16 E Cores at a lower one.

E Cores are quite powerful and give out the possibility for more cores with an overall lesser effect on power draw and heat dispersion. If there were no E Cores and Intel wanted there to be P Cores only, they'd have to decrease the clocks of the P Cores significantly in order to obtain the same specifications (in regards to TDP and temperature) they would with the addition of E Cores (differing from P Cores in the margins they leave out for heat and power consumption by running at lower speeds and harboring a pertinent architecture).
 
To add, the E-cores are also tiny in comparison. 4 E-cores take up the same space as 1 P-Core, at least in Alder Lake and Raptor Lake. If we went by Tech Power Up's testing results with E-cores only on the i9-12900K, 8 P-cores on average outperforms the 8 E-cores by a relative factor of 152%. If you replaced those 8 E-cores with 2 P-cores, you'd actually end up with a theoretical performance uplift of 190% over 8 E-cores, versus the 222% uplift from 8+8. Keep in mind in long sustained tasks, the P-cores are going to drop down in frequency quite far from their maximum turbo boost of 5+ GHz by default.

For gaming, while there's a much larger gap between P-cores only and E-cores only, vs. 8+8, I'd argue most of that is because 1. no game can saturate the processor's thread count and 2. games are more sensitive to clock speed in terms of frame rate. Considering the i5-12600K can easily keep up with the i9-12900K (there's about a 10% margin), this suggests that games don't really need as many P-cores as one might think.

I would argue that for basic web browsing and office work, you won't get much more out of a 4 core processor anyway, especially if it's clocked fast enough. Heck, I have a laptop from 2021 with a 4C/8T CPU and it does all the basics just fine without leaving me wanting for more. Although it probably helps I threw Linux on there. However, even the Core i3-12300 which is a 4C/8T processor, it can hold up pretty well in gaming, with the i9-12900K only outperforming it by a relative difference of 129% (although this only lists averages. The 0.1% is likely going to be worse).

Another thing to point out are the Sapphire Rapids Xeon processors aren't going to win any single core performance awards. The maximum turbo boost speed I can see on Wikipedia's list is 4.8GHz for the workstation model. The server models top out at 4.0GHz. But none of these are going to remain there under all-core sustained workloads (most of them drop down to below 3.0GHz). This isn't good for games, which most consumers care about anyway.
 
Last edited:

jnjnilson6

Distinguished
I think I read that the current E cores are about as strong as a core in an i5-6600K.

Is that about right?
Yeah. I think you are right. My Core i7-12700H has got E Cores which range up to 3.5 GHz and the i5-6600K goes up to 3.9 GHz. Keeping in mind the improvements of newer architecture it should be reasonable that a synonymous performance could be inferred.
 
Another thing of note too, heterogenous architectures are basically trying to get the best of both worlds. That is, efficient sustained workloads that peg all the cores and having the high performance to meet low-latency, low time-on-CPU tasks. But like a hybrid car, you can get even more performance letting them both run at the same time.

Something to note about the first point: why would efficient sustained workloads matter? Because you might save more power than you might think when you do a task over time. As an example, I noticed if I let my Ryzen 5600X run a Handbrake job while letting it go to an all-core boost of about 4.5 GHz, it can crunch a video in roughly real time (so 1 hour of video is 1 hour of conversion) while using 75W doing so. Disallowing boost drops the speed down to 3.6Ghz, or about 80% the performance. While this increases the time to about 72 minutes per 60 minutes of video, it does this at 45W. So over time, the faster method consumes 75WHr, while the slower method consumes 54WHr, or about 72% as much.

When you live in a place with expensive electricity, that actually adds up pretty quick. And also crucially, if this was done in the hot summer months, less energy used means less heat dumped in my room, which means less time I need to run the AC.

I think I read that the current E cores are about as strong as a core in an i5-6600K.

Is that about right?
Most places have pegged the E-cores as having the same performance as Skylake.
 
Hello everyone!
Frist time posting in this forum. So, here’s the question I have about Intel CPUs that heterogenous cores aka bigLITTLE, aka Efficient and Performance cores.

1st I don’t understand the point of having E cores and P cores on a desktop CPU. I understand why you would have that setup on a laptop, mobile device or any device that may have heat or power issues, but one the desktop it isn’t necessary.

2nd let’s say that it does make sense on a desktop as well; and we want to be responsible citizens and conserve energy. That’s great, and I applaud that; but why so many Efficient cores?
The whole idea is the efficient cores will get you by on low impact or no activity, and when you need more power the Performance cores kick in and do all the hard work. So in my view you should only have at most 4 efficient cores (I actually would prefer 2 E cores) and the rest all performance cores.

Now I have a theory on why Intel is doing this, and in my opinion it’s very nefarious and shady. Intel (well most computer hardware and software manufactures) always hated users that used consumer grade equipment and used it for business class work. Intel especially hated that user would use an i5 or i7 for a Server application and not buy the over priced (yet same product) Xeon CPUs. So now they are gimping the consumer grade CPUs with a bunch of E cores (and few P cores) to force users to purchase a Server or Workstation class CPU. One of the top-of-the-line Raptor Lake processors (Intel® Core™ i9-13900KS) has 8 P cores and 16 E cores! Why do you need 16 E cores? And because of this they sell it as a 24 Core CPU.

Remember Intel always thought that the average consumer doesn’t need more that 4 Cores. (By the way I’m not looking at threads, just physical cores.) One last thing; I wonder that how a Raptor Lake CPU would hold up to a true homogenous core CPU like the Ryzen 7950X, in really heavy workloads like a Workstation or Server environment.

Please let me know what you think; I am open to being completely wrong about this.
It's not done for power alone.

You get your full 8 core CPU you are used to with zero compromises on clocks when many cores are active, like you get with ryzen, but then if you need to run something CPU heavy alongside your normal workload the e-cores take care of it so again there is no compromise on the performance you are used to from your old 8 core, again unlike Ryzen unless you cut the CPU in half and only use one ccx for each thing you run.

Bottom line you get 8 full clock cores and still get enough multithread grunt for anything you like to do, but both at once.

Thread-Director-comparison-for-Gaming-678x356.jpg.webp


Another thing of note too, heterogenous architectures are basically trying to get the best of both worlds. That is, efficient sustained workloads that peg all the cores and having the high performance to meet low-latency, low time-on-CPU tasks. But like a hybrid car, you can get even more performance letting them both run at the same time.

Something to note about the first point: why would efficient sustained workloads matter? Because you might save more power than you might think when you do a task over time. As an example, I noticed if I let my Ryzen 5600X run a Handbrake job while letting it go to an all-core boost of about 4.5 GHz, it can crunch a video in roughly real time (so 1 hour of video is 1 hour of conversion) while using 75W doing so. Disallowing boost drops the speed down to 3.6Ghz, or about 80% the performance. While this increases the time to about 72 minutes per 60 minutes of video, it does this at 45W. So over time, the faster method consumes 75WHr, while the slower method consumes 54WHr, or about 72% as much.

When you live in a place with expensive electricity, that actually adds up pretty quick. And also crucially, if this was done in the hot summer months, less energy used means less heat dumped in my room, which means less time I need to run the AC.
When electricity meters so much you are going to use your GPU to do it with hardware acceleration in minutes instead of hours and at a fraction of the power.