Intel i5 13600K - just as super-sweet when overclocking?

OpenSource Ghost

Limp Gawd
Feb 14, 2022
I was looking to upgrade my 8700K to i7 13700K for my gaming rig, but then looked at how well Intel i5 13600K was performing and began wondering whether it was worth spending extra $$$ on i7... Then I realized that most reviews weren't taking "overclockability" of i5 13600K vs i7 13700K into consideration.

Does 13600K overclock just as well as 13700K? What about 13600KF? Does lack of iGPU provide more overclocking headroom?
Overclockability of modern CPUs is somewhat overrated. The boost algorithms are already designed to virtually get all the extra performance that is out there to be had. Sure, you can buy a pre-binned CPU with guaranteed clocks or whatever if you overclock to remove the silicon lottery factor out of the equation. Most "overclocking" motherboards will remove the boost power limits to allow you to get the full boost as long as you want (vs. for a shorter time "stock" limit). The difference is going to be minimal though between the power limits removed and dialing in an all core OC for most tasks. You will likely notice the higher power draw and cooling requirements on a manual all core OC though. Overclocking headroom boils down to silicon lottery more than whether or not there's an iGPU or not.

As for the 13600k vs 13700k, you can probably boost the 13600k a little higher as it has two less performance cores, but you're getting a higher binned core with the 13700k as the temperatures and power draw needs to stay lower with the extra cores to stay within the boost range and power limit. It's probably a wash. Right now the 13600k and 13700k are priced pretty evenly in terms of price per performance core right around $55 per core give or take.
OC-wise 13600K(F) is very fun CPU to play with.
Got my 13600KF to 5.5GHz on all p-cores and 4.3GHz on all e-cores and ring to 4.7GHz. Ring I will OC further but for now I am focusing on testing stability of cores by actually using processor. This result with core VID at 1.235V as shown in HWiNFO64.
By default it ran at something like 5.1/3.9GHz with nonsense vcore of >1.3V and I got it faster and colder at the same time by lowering vcore and increasing clocks. Then of course to get to the absolute limit power consumption will increase and best/worst cores will have to be identified and their ratios configured. Performance-wise it is better to have different cores at different frequencies than trying to have them all have the same frequency and with it lower frequency to what worst core can do. For example I identified 2nd p-core consistently fails at 5.6GHz first and needs much higher voltage than rest to be stable. It can be configured to run at 5.5GHz while the rest (or at least those which can do it...) run at 5.6GHz. For now I didn't implement this strategy.

I run 13600KF without Hyper Threading. Imho for gaming HT is nonsense and especially on Alder/Raptor Lake where HT cores should really be used last, after all p-cores and all e-cores are saturated with work. Disabling HT lowers power consumption and temperatures by a lot and simplify monitoring how application/game behaver regarding threads. For me its single core performance for system snappiness (and system/programs really do feel faster on 13600K, even with default clocks coming from OC'd 9900K)

The only thing i7 has over i5 (other than default clocks no one cares about 🙃) is 2 p-cores which might be useful for the future case where game actually needs more than 6 cores.
If game does need more than 6 cores though it can always use e-cores and with them having IPC for the most part comparable* to Skylake it isn't even an issue. Less optimal than having full 8 beefed up p-cores but also should not cause any issues, especially if game engine was coded with e-cores in mind**... which given Meteor Lake supposedly having 6 p-cores in its fastest configuration might be even less of an issue than it might otherwise be.

Anyways, there is little reason to worry about thread count. Threads are plentiful on 13600K(F) and I treat it not as 6 core CPU but 14 core CPU it really is.

K vs KF: should not matter that much. KF might use less power because iGPU is literally cut out with laser. I got KF because on 9900K I didn't use iGPU and also because it kinda was visible in device manager even when disabled... not a big issue but why worry about drivers, etc. for something I won't ever use. Might be useful for debugging but then again I have PCI-E cards for retro PC's so I can always pick some 7800GTX to check if my main GPU fried and this causes PC to not boot should such thing ever take place.

Also keep in mind 13700K(F) should be binned higher than 13600K(F) thus might overclock better. In the end however its as always silicon lottery...

*) Eight Raptor Lake e-cores overclocked to something like 4.3GHz are quite comparable to Skylake CPU like Core i7 9700. In Cinebench (SIMD ALU - most typically used instruction types in heavy workloads eg. rendering, video compression/decompression, games, etc) I get scores between 9700 and 9700K. Pretty crazy to get almost the same single core performance on Arom-derived cores as I had on 9900K 🤯

INT/ALU (typical stuff used when running program logic) performance seems also comparable. These were instructions these cores were optimized for.

Where e-cores kinda suck is classic FPU workloads, so more like old programs and games... which because we already had CPU's with this particular characteristics we already know how impactful it is: None at all. I mean comparing Phenom II vs Core 2 the performance of the former in eg. SuperPi was quite a bit lower but this didn't translate at all to lower performance when dealing with floating point workloads and this is simply because even at Core2/Phenom times all programs and games used some form of SIMD. Funnily enough taking SuperPi scores ~4GHz e-cores are still considerably faster than Phenom II @ 4GHz... not faster than Core 2 @ 4GHz :)
Overclockability of modern CPUs is somewhat overrated. The boost algorithms are already designed to virtually get all the extra performance that is out there to be had.
True if true else false. For 13600K(F) it is false.
This CPU has more potential performance than factory overclocking shenanigans can squeeze out of it.
With some effort it can run much faster and and at the same time using less power.
13600K running at 5.4Ghz (all P-Cores, SpeedShift disabled) @ 1.3v, cooled by Noctua NH-D15. I think 13600K may be today's 2500K.
What coolers are using to do minor overclock or day to day normal operation.
5.5GHz p-core and 4.3GHz on e-core, 4.7GHz ring @ 1.25V (set in BIOS), 1.24v shown in HWIiNFO64 for all cores. Hyper Threading disabled. All power saving features disabled in BIOS and Windows (Speedstep, parking cores) so clocks are always at maximum.

Noctua NH-D15Swith one original fan and one smaller 120cm which can spin a little faster + 120cm fan directly in front of the case where drive bays are pulling air (case has them transparent with dust filters) + 120cm fan in the back pushing air out of the case. Additionally for GPU comfort one 120cm fan pulling air from underneath the case and two pulling from front of the case.
A lot of fans but CPU can still get >100C with Prime95 small TTFs. With Cinebench R23 temps are below 100C and fans do not spin up (set to get really loud at 100C and beyond, otherwise spin rather slowly and some do not spin at all without moderate CPU load) so its fine.

Temperature-wise the biggest improvement was when I disabling Hyper Threading. Also allows to get stable with lower voltages. Of course without HT I loose some multi-threaded performance but for games its the correct setting because HT allows background processes stealing performance from p-cores. Also didn't saw any benefit to HT in MT cases I actually run (other than Cinebench...). In Cinebench R23 with HT off and 5.5/4.3GHz I get almost 22K/2.2K points with almost exactly 10x MT performance multiplier. At 5.6GHz which is also doable, albeit with much higher vcore (~1.3v wherabouts) I get slightly above 22K/2.2K
My i5 13600 is running at 5.5/4.2 without adjusting anything but ratio. I think I need more voltage for 5.6 because it's shutting down during userbench CPU test. Voltage is about 1.2ish