More valuable for gaming in 2024, E-cores or Hyperthreading?

xDiVolatilX

xDiVolatilX

2[H]4U
Joined
Jul 24, 2021
Messages
2,592
I have a question that I don't have time test myself so maybe someone knows or has a link to someone online that has tested this.

Curious about 13700k all the way up to 14900k.

What is more valuable using the hyperthreading on on the 8 CPU P cores?

Or using the 8-16 E cores available?

For gaming, which is better for more FPS?

If you had to chose one or the other not both so save some the heat generated by having everything just on at all times.

I can't find any good evidence of which is better online? Any help?
 
KazeoHin said:
But Performance will drop off if you disable HT.
Click to expand...
Depends on the game, it can be often margin of error difference once you have enough core.

Once you have 8 core or more (specially on a clean recent os install that do nothing), you can often see a stop in improvement by turning on HT for a game, dsogaming benchmark often have them
Dragons-Dogma-2-benchmarks-1.png


Elden-Ring-CPU-benchmarks.png


Call-of-Duty-Black-Ops-Cold-War-CPU-benchmarks-1.png


game engine are so well optimize that they can look like this:
The-Last-Of-Us-Part-I-CPU-scaling.jpg
 
Last edited:
xDiVolatilX said:
For gaming, which is better for more FPS?
Click to expand...
For a computer doing very little other than the game itself, it depends on the game (past enough core it should have a small impact) I think you need to look for the game in particular, if you have those 13700k-14900k type of cpu, can be something you do not think about without worrying if you want, never seen large gap for those.
 
Just to be clear.

I was being specific about

Hyperthreading VS E cores specifically

Because as some of you have mentioned after a certain amount of threads it's diminishing returns.

So if the sweet spot is 12 or 16 thread for example, is it better to have

8 P-cores and 8 E cores?

Or

8 P-cores with hyperthreading

Both 16 threads for example but which threads are more valuable for gaming? E cores threads or hyperthreading threads?

Because I feel like having 8 E cores or 16 E cores PLUS 8 hyperthreads is overkill like TOO many threads no game uses 32 threads and all the threads just generate more heat even if they are not fully active so if you had to pick and choose which threads to keep activated which threads are the fastest and worth keeping on?
 
xDiVolatilX said:
Just to be clear.

I was being specific about

Hyperthreading VS E cores specifically

Because as some of you have mentioned after a certain amount of threads it's diminishing returns.

So if the sweet spot is 12 or 16 thread for example, is it better to have

8 P-cores and 8 E cores?

Or

8 P-cores with hyperthreading

Both 16 threads for example but which threads are more valuable for gaming? E cores threads or hyperthreading threads?

Because I feel like having 8 E cores or 16 E cores PLUS 8 hyperthreads is overkill like TOO many threads no game uses 32 threads and all the threads just generate more heat even if they are not fully active so if you had to pick and choose which threads to keep activated which threads are the fastest and worth keeping on?
Click to expand...
8 P cores with HT will be faster than 8 cores without HT + 8 E-cores HOWEVER with 16 threads the difference is minor.

An E-Core is limited in speed: it will complete an operation more slowly and, without certain hardware instruction sets, will complete it using more processor cycles.

a core with HT has the full speed and capability of the core for each thread, and can complete a multiply cycle on one thread while doing the add cycle on another, and also work on one thread while the other is waiting for a block of memory etc.

That means in very rare but not unheard of instances, HT can essentially double a core's processing power, and both threads get to take advantage of the full IPC as if no other thread were being processed. worst case is that it evenly splits a core's horsepower so that two threads get half a core (so to speak)

Whereas an E-Core can ONLY be slow. there is no scenario where an E-core can operate as fast as a P-Core. So single-threaded power on HT is worse case half the speed of a P-core, best case 100% the speed of a P-core. E-cores are WAY less powerful than half a P-core, in most instances you're looking at one E-core being able to do the work of, say, 1/3 a full P-core? Which is actually really cool, because you can fit 8 e-cores into the space that would normally fit 1 P-core. So you can see why it's a space-efficient solution! but in terms of real-time, latency sensitive tasks, E-cores are not great. HT is MUCH more useful.
 
The niche scenario in that and something that may become more common as these types of CPU designs become more prevalent is the ability to assign simpler tasks to E-cores and free up the P-cores to do the heavy work. This is assuming that the E-cores do not throttle the P-cores due to increased heat/power usage, which appears to not be the case with current designs as well as not being smart enough to separate out the light from the heavy tasks. If 8 E-cores could become more power/heat efficient in addition to being more space efficient than a single P-core, this could open up the possibility of the P-core/E-core combo performing faster than HT.
 
Ecores the only time I really saw them used to potential when I had my old 12700K installed all 4 ecore were unpacking some really huge Steam game and it was done in minutes.
While other people were having a hard time with AMD chips or dated chips. Now I own a 13700K my PC is so stable I'm not sure if I'll upgrade to the latest CORE series chip.
I would like to but after upgrading like x15 times since 2002 for Everquest 2 with custom PCs I'm pretty sure I won't have to upgrade for a good 5 years.
I just upgrade last year to DDR5 6000mhz Ram it's easy to upgrade but will it come without hitches.
 
Last edited:
xDiVolatilX said:
So how do we explain this?


View: https://youtu.be/I8DJITHWdaA?si=qpTth0y9orE7gXFz

Danny is saying out of a 40 game benchmark the majority of them run little bit faster especially 1% lows with hyperthreading OFF because the main cores physical and logical cores are not fighting for the same resources etc.

Watch this review and tell me what you think.
Click to expand...

Makes sense to me. The CPU used in that video has 8 P-cores and 16 E-cores, so 32 threads with HT on and 24 with it off. Most games don't need that many threads. Many can't even use that many. Games generally have a few hot threads and a bunch of less demanding ones. Turning off HT prevents the hot threads from having to share a core. Whether HT is on or off the hot threads still have to end up on a P-core for optimal performance, and ideally the other thread on each P-core should be idle if the CPU doesn't need to use all of it's threads and the game doesn't have more hot threads than the CPU has P-cores. Regardless of whether HT is on or not the ideal setup is the hot threads have a P-core all to themselves. I've quite a bit of performance testing on financial trading apps, and HT on or off hardly makes any difference if the other thread on a core is idle. I reckon a lot of the cases where HT on is a little slower is because the scheduler did something suboptimal and put 2 threads on a P-core when it would be better not to. A "perfect" scheduler would always be about as fast or faster with HT on as long as mitigations for HT security bugs are disabled. But making a perfect scheduler is really, really hard so I'm not surprised it often works out slightly better to disable HT.
 
13900k make HT on for playing video game on a fresh windows with nothing in the background a bit mute.

xDiVolatilX said:
are not fighting for the same resources etc.
Click to expand...
If you look at 11:35, the cpu scheduler-game engine-os (do not known enough who responsible for this), seem to be really good at having the second thread at 0% for all the main game core. Could it just be that a core that has HT on is always a tiny bit slower at single thread even if it does not fight off resource ?
 
LukeTbk said:
13900k make HT on for playing video game on a fresh windows with nothing in the background a bit mute.


If you look at 11:35, the cpu scheduler-game engine-os (do not known enough who responsible for this), seem to be really good at having the second thread at 0% for all the main game core. Could it just be that a core that has HT on is always a tiny bit slower at single thread even if it does not fight off resource ?
Click to expand...
That's been my experience, but it's just a tiny bit. Again, my experience is with financial trading apps rather than games, but I bet the results with games are similar.
 
xDiVolatilX said:
8 P-cores and 8 E cores?

Or

8 P-cores with hyperthreading

Both 16 threads for example but which threads are more valuable for gaming? E cores threads or hyperthreading threads?
Click to expand...

Before complications come in e-cores are more valuable when you keep all 16 threads busy. A hyperthreaded pair has about 118% of a single core, whereas an e-core and a p-core are about 150%.

It gets complicated when you have a lull in activity and drop from 16 busy threads down to -say- 2. If one or both of those was scheduled on an e-core it stays there a moment and its performance is 50% of a p-core. If you idle one thread in a pair that was on a hyperthreaded core then whichever thread is still busy will instantly get a full p-core worth of performance.
 
Well boys I tested the latest 3DMark CPU test and this is what I found interestingly enough,

13700k stock 6000mhz ram c32 4090

The point of this test is to test 16 threads in 2 ways. 8pcore+8hyperthreads vs 8pcores+8ecores. This way the only variable is 8hyperthreads vs 8ecores to determine which is faster for the 3DMark CPU game benchmark.

The first image is with

8 pcores + 8 ecores NO hyperthreading just 16 physical cores


8 pcores 8 ecores no hyperthreading.png


This second image is of,

8 pcores + hyperthreading NO ecores 8 physical cores 8 logical cores



8 pcores with hyperthreading no ecores.png


So looking at these scores clearly the 8 physical ecores are faster than the 8 logical p-hyprethreads. It clearly wins by a thousand points or so. As far as 3DMark CPU test goes, it is obvious that E cores are more powerful than p- hyper threads for a CPU intensive game or CPU game benchmark.

If anyone wants to chime on with more conclusive evidence please do I'm interested to see if this test is accurate enough. Feel free to run the same test on cinibench if you like this test was a 3DMark game engine CPU test.
 
Last edited:
xDiVolatilX said:
If anyone wants to chime on with more conclusive evidence please do I'm interested to see if this test is accurate enough. Feel free to run the same test on cinibench if you like this test was a 3DMark game engine CPU test.
Click to expand...

Your tests are valid, but they do not represent games. As I said above, there is no question that an e-core has more computing power than a hyperthreaded sibling - when all cores are in use.

When on the other hand the application drives only a fraction of the cores (such as current games) then threads that could use a full P-core can be stuck on an E-core for a moment (which in computing terms can be an eternity).
 
You must log in or register to reply here.
Back
Top