RTX 4090 & 53 Games: Core i9-13900K E-Cores Enabled vs Disabled

[Minus Infinity]

But what about power. If the E cores make bugger all difference, surely disabling 16 of them would lower temps and power usage a fair bit?
You'd be better just limiting the 13900K to say 180W for gaming.

 

[Lycanwolfen]

I have done some work on some of the newer core i7's with E cores and P cores. I find windows really gimps them. There is two settings I find that makes a hude difference in performance if you enable it. One is this Key HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Power\PowerSettings\54533251-82be-4824-96c1-47b60b740d00\93b8b6dc-0698-4d1c-9ee4-0644e900c85d change the attribs to value of 2 and then the next one is HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Power\PowerSettings\54533251-82be-4824-96c1-47b60b740d00\bae08b81-2d5e-4688-ad6a-13243356654b and also change attribs to value of 2 then in the power settings advance under processor you will see two new options. I put it too all processors and then the P cores and E cores are in sync reguardless of there slow or higher speeds. I saw a huge increase all all games and apps with these settings enabled.

 

[Dirt Chip]

General recommendation I have seen a lot towards people that purchase a 13600K for gaming is to turn them off if they stay on Win 10, and leave them on if they move to Win 11.

Is it based on something?

Maybe it is just the opposite? You can get a CPU without e-cores jsut for gaming and don't lose any performance. E-cores are made to allow Intel to stay competitive at multithreading and not lose very badly at efficiency.

Where did you lose pref outside of a spacific example that is counterd by other examples that show increase?

 

[swirl09]

But what about power. If the E cores make bugger all difference, surely disabling 16 of them would lower temps and power usage a fair bit?
You'd be better just limiting the 13900K to say 180W for gaming.

Things may have changed since I did any testing, but I found E-cores were hit and miss. So I just turned them off and instead bumped my P-core clocks and ring. Wasnt a big gain, but it was recordable and I dont have to wonder if X game is going to play nice with E-cores or work with certain DRM etc.

Again, this was a year ago so. But if on vs off is 0.1% better on a 13900K at 4K, then I would assume if you turned them off and just gave 100Mhz to the P cores, it would win.

 

[Dirt Chip]

Maybe it is just the opposite? You can get a CPU without e-cores jsut for gaming and don't lose any performance. E-cores are made to allow Intel to stay competitive at multithreading and not lose very badly at efficiency.

Also, if it doesn't hurt pref, why turn it off?
It can help in background process that unlike this sterile test banch, have some things running while you game.

If you reffer to 5800x3d, you are right- it's doing fine without e cors but it's quite often not the ideally cpu to chooses and as was stated in the article not the scoop the testing.

 

[nguyen]

Things may have changed since I did any testing, but I found E-cores were hit and miss. So I just turned them off and instead bumped my P-core clocks and ring. Wasnt a big gain, but it was recordable and I dont have to wonder if X game is going to play nice with E-cores or work with certain DRM etc.

Again, this was a year ago so. But if on vs off is 0.1% better on a 13900K at 4K, then I would assume if you turned them off and just gave 100Mhz to the P cores, it would win.

I tried disabling E-cores in Spiderman Remastered and not only the FPS are lower, frametime consistency is also bad.

Overall on my 13700KF, 5.3ghz Pcore + 4.2ghz Ecore give higher FPS, more consistent frametime, lower power draw compare to 5.5ghz Pcores only, though I only tested in Spiderman.

Would be nice if Intel provide software to park Ecores in windows, so far only Gigabyte has executable to do so but only work with their Z690 motherboard.

 

[Dirt Chip]

@W1zzard Unfortunately, intended or not, that is not the picture your graphs are showing (literally). Clipping graphs at +/-15% makes it look like there are significant differences. I really wish you didn't do that. In any of your comparison charts.

You always clip your graph after the top\max result. This way you clearly see the relativness which is basically percentage.
If not, where did you stop? 50%? 100%? 500? 1000?

What you shouldn't do is clip some parts of the graph below the max to emphasize the difference.

The max value is not +/-100%. It's +/-15%. That makes a tiny 5% variance go 1/3 of the way, which the brains really perceives as 33%. It's a means to emphasize small differences, but when used without a disclaimer is really just a way condition the mind towards a certain outcome.

Nop. As long as you clearly write the % on the y axis it is very much fine.

Only if you clip chunk's under the max value than you manipulate.

Things may have changed since I did any testing, but I found E-cores were hit and miss. So I just turned them off and instead bumped my P-core clocks and ring. Wasnt a big gain, but it was recordable and I dont have to wonder if X game is going to play nice with E-cores or work with certain DRM etc.

Again, this was a year ago so. But if on vs off is 0.1% better on a 13900K at 4K, then I would assume if you turned them off and just gave 100Mhz to the P cores, it would win.

But that is the same problem from the opposite side: if you turn off you can wonder at what games you loose pref, so this 'issue' basically cancel itself out.

 

[DarthJedi]

> We wish Intel made it a touch easier to toggle E-cores (maybe through XTU), so gamers wouldn't have to pull up the system BIOS or set core affinity

On Asus boards at least, this works with a simple scroll lock key press.

Also, you are not testing correctly - Windows 10 should not be used with hybrid CPUs. Use Windows 11 - there is a critical difference with the kernel, thread scheduler.

 

[Dirt Chip]

> We wish Intel made it a touch easier to toggle E-cores (maybe through XTU), so gamers wouldn't have to pull up the system BIOS or set core affinity

On Asus boards at least, this works with a simple scroll lock key press.

Also, you are not testing correctly - Windows 10 should not be used with hybrid CPUs. Use Windows 11 - there is a critical difference with the kernel, thread scheduler.

If any change, and according to you, win11 will show only better resoult in favor of e-cors on.
So win 10 is the "worst case" for e-cors and it`s still show zero change on/off => no real reason to bother yourself by turn it off unless at a very spacific game.

 

[DarthJedi]

If any change, and according to you, win11 will show only better resoult in favor of e-cors on.
So win 10 is the "worst case" for e-cors and it`s still show zero change on/off => no real reason to bother yourself by turn it off unless at a very spacific game.

Basically, yes.

 

[Wirko]

Awesome job with the thorough testing. Handy to know for those wanting to get the absolute most out of the cpu for certain games.

I am curious if there's any noticeable difference in frame times with e cores on or off. Would be an interesting follow up. Maybe mix in hyper threading on or off as well. All core and threads on. All cores on no HT, p cores only and no HT, and for fun e cores only lol. Maybe even test with all cores locked at max clock vs allowing them to down clock based on use. Test fps and 1% and .1% lows as well.

Just an idea for a slow day I suppose. HT and turbo boosting might be negligible, but I do remember arguments back in the day about possible impact on latency with HT or turbo speeds or frequency changes, etc. Maybe figure out which games or game engines prefer which environments. This could be a lot of work though so limiting to a couple of games/engines would be understandable. Especially if running just one batch of tests shows its basically a pointless task.

Very much agreed, it would be an interesting experiment to test a subset of games running on E-cores only. The few outliers in this review, both positive and negative, would be the best candidates. 1080p would suffice. Also, as you say, with and without HT.

Hah, reminds me of a video that I just watched today (or rather listened to)

Well, "statistics is a lie", and it's bigger lie if you don't read the fine print.

 

[MarsM4N]

@W1zzard The Core i9-13900K "Raptor Lake" really is the best processor you can buy for gaming, if money is no bar. It can actually save you a few bucks over the AMD Ryzen 9 7950X, money saved that you can spend on a faster GPU.

Anyway, 13900k is a bad investment for gaming and should be avoided if no other tasks are in use. Put your money on the GPU.
But it shows nicely that you can peacefully grab 13700 or better, 13600, for gaming only with e cors "on".

Exactly. I still find it puzzling that some who clearly should know better trying to sell a CPU that is designed for productive works as a "best processor you can buy for gaming". You can save way more bucks if you go for an 13600K or 13700K (which will give you give/take the same FPS in games) and put the money you saved towards a better GPU. That's something I would put in my "Conclusion" & no disinfo for overselling.

Results don't make sense to me. The games that run faster with e cores enabled are more likely just random variance? Far Cry being faster really makes no sense to me.

It's either a DRM issue or Windows is juggling around some processes between the P-Cores & E-Cores. Who knows, but something doesn't play right.

Still remember the headlines "Intel 12th Gen Alder Lake CPUs aren’t playing nice with DRM in some games". Intel claims they've fixed it by now, but who knows. I wouldn't also be surprised if E-Core people still getting kicked out of games or even collecting bans. Apparently the thing is that anti cheat engines recognize E-Cores as VM's (Virtual Machines) or a 2nd PC, which is what the more advanced cheats are today.

Since overall performance is virtually the same, with a few outliers in either direction, did it have any positive, negative, or game dependant affect on power comsumption, or maybe 1(5) % lows?
Since the average user will have more background processes running, I would venture a guess that in the real world usage, it would tilt toward e-cores on rather than off if it even has an impact.

Yea, power consumption with E-cores on & off in gaming would be very interresting.

FPS performance & 1% lows with loads of background processes (online streaming, rendering, cloud uploads, etc.) running would also be something interresting to test. However, if there is no benchmark tool out there (f.e. like PCMark) that includes such an benchmark, it would be kinda hard to set something up & get reliable results, I guess.

 

[Dirt Chip]

Exactly. I still find it puzzling that some who clearly should know better trying to sell a CPU that is designed for productive works as a "best processor you can buy for gaming". You can save way more bucks if you go for an 13600K or 13700K (which will give you give/take the same FPS in games) and put the money you saved towards a better GPU. That's something I would put in my "Conclusion" & no disinfo for overselling.

Yes, but you didn't complete the quote "...if money is no bar". And it`s the truth.
No need to or it will boomerang to you

Very much agreed, it would be an interesting experiment to test a subset of games running on E-cores only. The few outliers in this review, both positive and negative, would be the best candidates. 1080p would suffice. Also, as you say, with and without HT.

I`m pretty sure you can`t do only e-cores. at least 1 p-core must be activated, no way to turn them all off.

 

[MarsM4N]

Yes, but you didn't complete the quote "...if money is no bar". And it`s the truth.
No need to or it will boomerang to you

To be fair, I did quote the whole sentence, right on the top. Opinion is opinion, and a bad recommendation should be allowed to be critiqued. At least in a free country, lol.

The comparison is very interesting, but why it is done in Win10...
E-core scheduling support is available only in Win11 afaik.

False. According to Intel the task scheduler also works in Win10. But it's more "optimized" for Win11.
Is Windows® 10 Task Scheduler Optimized for 12th Generation Intel® Core™ Processors?

 

[DarthJedi]

To be fair, I did quote the whole sentence, right on the top. Opinion is opinion, and a bad recommendation should be allowed to be critiqued. At least in a free country, lol.



False. According to Intel the task scheduler also works in Win10. But it's more "optimized" for Win11.
Is Windows® 10 Task Scheduler Optimized for 12th Generation Intel® Core™ Processors?

They will never say "no" to you, but no. Thread Director works, the CPU will work, but Windows 10 does not know how to manage threads by priority and load correctly in context of P and E cores performance-wise.

 

[Jism]

I wasn't aware that anybody was worried about the e cors in games...

This is TPU at it's best: debunking myths pretty much.

 

[zlobby]

Hey, @W1zzard ! Can we please get some 1% or even P5 when talking about FPS? I know you want to push the reviews out quickly, but even so you could add these numbers afterwards.

I have done some work on some of the newer core i7's with E cores and P cores. I find windows really gimps them. There is two settings I find that makes a hude difference in performance if you enable it. One is this Key HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Power\PowerSettings\54533251-82be-4824-96c1-47b60b740d00\93b8b6dc-0698-4d1c-9ee4-0644e900c85d change the attribs to value of 2 and then the next one is HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Power\PowerSettings\54533251-82be-4824-96c1-47b60b740d00\bae08b81-2d5e-4688-ad6a-13243356654b and also change attribs to value of 2 then in the power settings advance under processor you will see two new options. I put it too all processors and then the P cores and E cores are in sync reguardless of there slow or higher speeds. I saw a huge increase all all games and apps with these settings enabled.

If Linux is good for anything, it would be to show us how bad Windows is in scheduling and thread management. Sure, average Joe won't notice it much in Word and YouTube.
Let's not even start talking about DAW's and near real-time applications...

Out of the games with performance penalty I have Metro Exodus so I've done some testing. This is on 13900K 5.9 core 5.2 cache 4.5 ecore 4x8 ddr4-4100 cl14 timings tuned.

(8 threads) 8P/0E/HT0 - 165.26
(16 threads) 8P/0E/HT1 - 160.31
(16 threads) 8P/8E/HT0 - 157.58
(24 threads) 8P/8E/HT1 - 148.42
(32 threads) 8P/16E/HT1 - 139.06

So the penalty is not from E-cores, it is from the number of threads. At 16 threads the penalty for running 8+8 instead of 8c/16t is just 1.5% and disabling hyperthreading increases overclocking potential by 200MHz at the same voltage, so performance will equalize.

This is quite in line with some other tinkering I've been doing since getting the CPU, this kind of situation happens from time to time, that's why the maximum number of threads you want to have in gaming setup is 16, going above causes issues that are not so easy to detect, because it is not necessarily games just refusing to launch or crashing (this does happen at 32 or more threads in some games though), but underperforming in either framerate or frametime consistency, and you have no way of knowing until you test it for each game.

I've also found some single threaded games like CSGO or Witcher1 that get 10% boost from E-cores even though they obviously sit idle, so either the CPU can access E-core resources somehow or it doesn't always function properly with E-cores off, so disabling all of them is not a good idea.

So overall 8P/8E/HT0 seems a way to go for gaming. 16 threads do not have any compatibility issues or meaningful performance penalty, you need E-cores for games that get a boost from them and disabling HT increases overclocking potential by 200 MHz at the same voltage.

Law of diminishing returns in action.

 

[samum]

These matchups are great at addressing the pros and cons potential upgrade paths for gaming. I do agree with some other posters that the 1% minimum would be helpful, since that's really what determines how smooth the game feels.

 

[Godrilla]

These matchups are great at addressing the pros and cons potential upgrade paths for gaming. I do agree with some other posters that the 1% minimum would be helpful, since that's really what determines how smooth the game feels.

Same here I care more about the 1% lows more than the average. I am holding off on upgrading my 9900ks with 4090 suprim liquid until the zen4 3d to improve the 1% and 0.1% lows.

 

[RogueSix]

Wow, thanks for the extensive testing. I was originally considering a Zen 5 system but I'm now leaning more towards Intel and the 13900K. Good to know that the E-cores are a non-issue (I'm a 4K gamer). Seems like Intel have definitely improved their hardware scheduler a.k.a 'Thread Director' and successfully taken the reins away from any Mickeysoft scheduler shenanigans. This should be high priority on AMD's agenda as well imo. They also need to put a stronger focus on hardware scheduling. Microsoft is too slow and sluggish with their adaptations.

 

[moobear]

The e cores to Intel is what the chiplet is for AMD. A very good investment that the opposite should adopt in order to stay competitive.

I wasn't aware that anybody was worried about the e cors in games...

Anyway, 13900k is a bad investment for gaming and should be avoided if no other tasks are in use. Put your money on the GPU.

But it shows nicely that you can peacefully grab 13700 or better, 13600, for gaming only with e cors "on".

I disagree intel forced the e cores up on us tbh noone asked for the e cores without them intel would sell better because they probably would be cheaper

 

[zlobby]

Same here I care more about the 1% lows more than the average. I am holding off on upgrading my 9900ks with 4090 suprim liquid until the zen4 3d to improve the 1% and 0.1% lows.

Zen4 X3D is the thing every gamer waits for! Not to mention intel...

I disagree intel forced the e cores up on us tbh noone asked for the e cores without them intel would sell better because they probably would be cheaper

I too don't see the point in E-cores in HEDT CPU's.

They will never say "no" to you, but no. Thread Director works, the CPU will work, but Windows 10 does not know how to manage threads by priority and load correctly in context of P and E cores performance-wise.

It's a scam lead by intel and supported by M$. Both want to push product by any means neccessary.

 

[bug]

You always clip your graph after the top\max result. This way you clearly see the relativness which is basically percentage.
If not, where did you stop? 50%? 100%? 500? 1000?

You've kind of answered your own question: for percentage graphs, you should always go to 100%.

What you shouldn't do is clip some parts of the graph below the max to emphasize the difference.

That, too.

 

[Max(IT)]

But what about power. If the E cores make bugger all difference, surely disabling 16 of them would lower temps and power usage a fair bit?
You'd be better just limiting the 13900K to say 180W for gaming.

the 13900K doesn't even come close to 180W while gaming.
Mine mostly stands between 60 and 80W, with just a few spikes over 120/130W (it depends on the game).

I disagree intel forced the e cores up on us tbh noone asked for the e cores without them intel would sell better because they probably would be cheaper

actually not. E-cores allow for an higher core counting without being crazy expensive and power limited.

To be fair, I did quote the whole sentence, right on the top. Opinion is opinion, and a bad recommendation should be allowed to be critiqued. At least in a free country, lol.



False. According to Intel the task scheduler also works in Win10. But it's more "optimized" for Win11.
Is Windows® 10 Task Scheduler Optimized for 12th Generation Intel® Core™ Processors?

Actually it is not "more optimized" for Windows 11. It is completely NOT OPTIMIZED in Windows 10, according to their words

So it is not very smart to use Windows 10 with 12th or 13th Gen Intel CPUs.

I have done some work on some of the newer core i7's with E cores and P cores. I find windows really gimps them. There is two settings I find that makes a hude difference in performance if you enable it. One is this Key HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Power\PowerSettings\54533251-82be-4824-96c1-47b60b740d00\93b8b6dc-0698-4d1c-9ee4-0644e900c85d change the attribs to value of 2 and then the next one is HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Power\PowerSettings\54533251-82be-4824-96c1-47b60b740d00\bae08b81-2d5e-4688-ad6a-13243356654b and also change attribs to value of 2 then in the power settings advance under processor you will see two new options. I put it too all processors and then the P cores and E cores are in sync reguardless of there slow or higher speeds. I saw a huge increase all all games and apps with these settings enabled.

what about power consumption with those settings ?

 

[EatingDirt]

You've kind of answered your own question: for percentage graphs, you should always go to 100%.

No. That's nonsense and sorry, but you don't know what you're talking about. In the case of this graph, and many percentage graphs where there is a baseline number involved, in this case low vs high FPS, this graph can theoretically go to near infinity %. If we were comparing a Bulldozer FX-4200 to the 13900k this graph would go well over 100%, for example.

Not to mention that this graph would be practically useless with the amount of 0-2% bars.