enhanced turbo, speed shift, C-state ....... on or off? 14900KS

Scuzzmuffin

New member
Joined
Oct 24, 2024
Messages
14
Ive read a bunch of guides around overlocking/undervolting the 14900ks, but there are 3 settings I'm still confused if they should be on or off for best performance, and how they effect overclocking.

1) Enhanced Turbo - should this be on or off?
- Default appears to be on. Ive seen several people leave it alone and a few turn it off,
2) Intel Speed Shift - should this be on or off
- Default appears to be on. Ive only seen it discussed like once, but didnt say whether to keep it on or off
3) C-State - should this be on or off
- Default appears to be on. Ive seen about half say turn off, and half say leave it alone.
4) CPU Switching Freq - Keep on auto or set value, 500, 600, 700, 800?
- Ive seen a few comment places that a value between 600-800 can improve performance, but with no real support on what it does.
5) Turbo Boost Max Tech 3.0 - does disabling this improve performance with overclocking?
- i understand it try to push certain workloads to specific cores, which is suppose to improve performance, but if we are running all core at 57 or 58x should this be turned off?

My goal is to get max performance out of the 14900ks at 58x.
* im currently at 57x with the following results
-----Intel XTU benchmark : 14200
-----Cinebench R23 : 41200
-----Time Spy CPU Test : 23500
***100% load temp of 89c***

Thanks in advance
 
Last edited:
Enhanced Turbo - should this be on or off?
Best to leave it off unless you have a really good cooling solution that can keep temps below 70C. And even then you don't want to drive up CPU voltages any more than necessary with Raptor Lake, which the higher clock frequency jump of Enhanced Turbo (I think 1x Ratio) is likely to do. Just not worth the risk for a tiny undetectable reward while gaming, etc.
Intel Speed Shift - should this be on or off
Absolutely leave it on otherwise your CPU will never downclock (all the way to an idle state of 800Hz - 1000Hz) when cores are inactive or lightly in use.
C-State - should this be on or off
- Default appears to be on. Ive seen about half say turn off, and half say leave it alone.
Again, Default is fine. People who promote deactivating C-States are usually hunting for that very last bit of perceived latency reduction. But, again, the wear-and-tear risks outweigh the rewards. And the rewards are heavily debated, anyway.

CPU Switching Freq - Keep on auto or set value, 500, 600, 700, 800?
Definitely don't touch that Auto default - I believe it is 500Hz - and especially not on MSI motherboards. Trust me when I say there is very little chance of an upside, but there are several known issues to the downside when increasing the switching frequency, including instability and crazy transient spikes in voltage that can increase CPU degradation (wear-and-tear).

Turbo Boost Max Tech 3.0 - does disabling this improve performance with overclocking?
- i understand it try to push certain workloads to specific cores, which is suppose to improve performance, but if we are running all core at 57 or 58x should this be turned off?
The answer is a little more complicated when it comes to the Intel Turbo 3.0. It will depend on whether you have done the following: [1] Have you installed the latest BIOS that includes microcode 0x12B? [2] What steps have you taken to reduce CPU Voltage via undervolting? [3] Why do you want to deactivate the factory boosting behaviour in favor of manual core ratios? [4] What is your target application, e.g. games?
 
@FlyingScot thank you for your answers

My answers to your questions on my turbo max 3.0 question.

I do have the latest microcode via the bios update on my MSI z790 godlike max motherboard. A60 bios, microcode 12B

I have LLC on mode 6
Ac at 35
Dc at 50
Vcore on adaptive with a -0.1 offset

Just seeking best performance. I’m a min-maxer. I find enjoyment in knowing I pushed things to the edge
 
Last edited:
@FlyingScot thank you for your answers

My answers to your questions on my turbo max 3.0 question.

I do have the latest microcode via the bios update on my MSI z790 godlike max motherboard. A60 bios, microcode 12B

I have LLC on mode 5
Ac at 35
Dc at 50
Vcore on adaptive with a -0.1 offset

Just seeking best performance. I’m a min-maxer. I find enjoyment in knowing I pushed things to the edge
Wow! I like your setup. Those are basically my favorite loadline settings at present. And the UV looks healthy, too. Are you using CEP? You certainly could with those loadline settings. No real reason not to.

Well, given where you're at with your tweaking and twiddling, and assuming you're willing to accept my answers to your other questions, then what I would do is push on with your manual all-core tuning, without enabling enhanced turbo (just do it manually by upping the peak ratios yourself) and keep an eye on peak voltages/temps, etc. while gaming - assuming that you don't have a need to use a heavily multithreaded app on a regular basis. I think we have to be careful not to become too distracted by the benchmark/stress test apps that are nearly all all-core workloads to the detriment of good gaming performance. I take the approach that for those moments where bragging rights is important then have a dedicated OC Profile that is heavily tuned for the all-core workloads. But for the rest of the time, use a game-optimized set of settings.

So with this in mind, I would save your best efforts with your current approach (once finalized) to one of the available BIOS OC Profiles and then I would see how much I could extract out of the factory boosting algorithms. In gaming, for example, you might get better performance due to the preferred cores boosting higher. A lot of games are still very dependent on one or two cores to keep everything in sync. I now keep a carefully eye on the HWInfo64 sensor "Max CPU/Thread Usage [%]" to give me an idea of any potential CPU bottlenecks because with modern CPUs, and their many cores, CPU Usage%s are pretty much useless.

For a factory boosting profile, if you like tuning, I would suggest trying positive/negative turbo offsets to control/enhance the behaviour of Intel Turbo 3.0 and perhaps take a deep dive into OCTVB to control/enhance the behaviour of TVB and eTVB. See my little chart below for the 14900K. You might be surprised at how much fun you can have. I'd then compare your two approaches and then pick a winner for your particular usage case.

Factory Boot 14900K.JPG

As far as the sequencing of events, you could check out my undervolting guide in the link in my signature. I'm sure most of it you already know. But the sequencing, etc. might keep you out of trouble, or at least give you some addition ideas. If you want more info on OCTVB then Skatter is your man. I can point you in his direction if you're not aware of his OC videos and articles.

EDIT: I forgot to mention that you could start out with your current loadline settings for the factory approach. You would only need to tweak them if you need more voltage at the higher frequencies.
 
Last edited:
Wow! I like your setup. Those are basically my favorite loadline settings at present. And the UV looks healthy, too. Are you using CEP? You certainly could with those loadline settings. No real reason not to.
Oops! My eyesight is not what it use to be. I thought you were using LLC5 instead of 6. It's a relatively small difference, but CEP might be on the edge of stability with LLC6 and AC_LL 35. I kind of still like LLC5 for some reason. I guess it's because you can drive AC_LL even lower...and you get to enable CEP.
 
@FlyingScot
Thanks! yeah i went all in on my setup

CPU - 14900KS
MB - MSI Z790 Godlike Max
PSU - EVGA 1600watt
Ram - TeamGroup T-Force Xtreem 48GB (2 x 24GB) DDR5-7600 PC5-60800 CL36
GPU - Asus ROG Strix 4090
Cooler - ASUS ROG Ryujin 3 360 w/ Grizzly Contact Plate
Case - MSI Prospect 700r

Up to this point ive had CEP disabled. Does enabling CEP provide a advantage in performance if it isn't limiting you? i guess im still confused on what CEP does and how it benefits overclocking.

would you recommend moving LLC from 6 to 5? how should i adjust my current ac/dc of 35/50? change to 15/20?
 
Last edited:
Thanks! yeah i went all in on my setup
CPU - 14900KS
MB - MSI Z790 Godlike Max
PSU - EVGA 1600watt
Ram - TeamGroup T-Force Xtreem 48GB (2 x 24GB) DDR5-7600 PC5-60800 CL36
GPU - Asus ROG Strix 4090
Cooler - ASUS ROG Ryujin 3 360 w/ Grizzly Contact Plate
Case - MSI Prospect 700r
Good heavens! You have a beastie of a system!! Do the lights dim when you fire it up??? LOL

Up to this point ive had CEP disabled. Does enabling CEP provide a advantage in performance if it isn't limiting you? i guess im still confused on what CEP does and how it benefits overclocking.
If you have the time, read my LEVEL THREE Knowledge in my signature "Guide..." link. It's the best view I have to-date of the reasons for enabling CEP. In the Level FOUR Knowledge, you will see some testing that Charonme did with a basic oscilloscope. I sat on the fence for a long time watching the CEP debate in action. I finally got the nudge I needed when I watched BuildZoid's video of the history of the Loadline mechanism. Since then, I have come to see CEP as the future. I also think that CEP might have some benefits under certain conditions. Intel also recommends CEP. So for those reasons, I lean towards trying to enable it. But I'm no CEP zealot. If it gets in the way of your tuning goals, and you can do much better without it (especially with lower voltages) then I'd be fine with disabling it.

In regards to your question, "Does CEP provide an advantage in performance..." I think Charonme's testing hinted that it could. But while the tests were repeatable, we all recognized that any perceived advantage was within the margin of error. So CEP still remains an emotional decision.

would you recommend moving LLC from 6 to 5? how would that effect my current ac/dc of 35/50?
The droopier loadline does seem to do well with the all-core workloads, but the LLC5 has the potential to behave in other ways that may or may not be beneficial to you, like when it comes to gaming. If I were you, I'd definitely try it both ways, i.e. build around both LLC5 and LLC6 and then pick a winner based upon your priorities.

If you visit my Raptor Lake Survey (link in signature) you will see an entry by xxOvrKillxx. I briefly worked with him to dial in his 14900KS. He tried both LLC5 and LLC6 and liked LLC5 for it's overall characteristics and rock-solid stability. He did likely sacrifice some all-core benchmark performance but he liked the better gaming experience.
He's still playing with his system if you want to try and reach out to him. I think he even has some fresh experiences with LLC6. You'll note that he is using the factory setup for frequencies, so his settings might be a good starting point for the factory boosting approach. Btw, he also has an RTX4090 and a good cooling solution/case. You might be able to make direct comparisons.

In regards to latency reduction (i.e. your earlier questions) if that becomes an issue, check out the user comment left by Vassil_V in that same spreadsheet. I haven't tried his suggestions but he is an accomplished tuner.

Hey, and don't forget to submit your settings to my survey (multiple entries if you want) when you finalize your settings.
Keep me posted.
Cheers!
 
It has definitely tripped the breaker once.

I dont see in your guide any suggestions on AVX, do you leave it alone
i think it has 3 items in the MSI Bios
* AVX - enable/Disable
* AVX Offset - (0 - -20)
* AVX Guardband - (0-256)

Right now i have all 3 on auto.
 
It has definitely tripped the breaker once.

I dont see in your guide any suggestions on AVX, do you leave it alone
i think it has 3 items in the MSI Bios
* AVX - enable/Disable
* AVX Offset - (0 - -20)
* AVX Guardband - (0-256)

Right now i have all 3 on auto.
AVX is a bit of a developing story for me. I do tend to leave it alone because in Comet Lake (and earlier days) AVX offsets could actually lead to instability due to it cutting in an out and causing transient spikes. I still prefer not to use AVX offsets for that reason. However, what's interesting is that Intel has changed the logic (at least with Raptor Lake) where light AVX workloads (maybe in DirectX for an example) actually ignore the offset now. Only heavy AVX workloads will downclock the CPU if you use negative offsets. That does make the utility of offsets potentially more useful.

In regards to AVX Guardband, I've never messed with it. I also saw someone recently complain that it had no measurable effect in their testing. The Auto setting appears to be pretty good.

Of course, we could also discuss Ring offsets. But hey, I have no idea where to start with that one when it comes to Raptor Lake. I won't touch it unless there was a really good reason to downclock the ring to reduce degradation. Plus, the Ring is very very sensitivity to even the slightest reduction in voltage.

BTW, keep an eye on which VID is the highest. Sometimes the E-Cores are dominant and drive up the voltages for the P-Cores. That might be a reason to downclock the E-Cores to get rid of the overlap. Fine tuning, indeed. But it's worth keeping it in mind.

Also note, that due to silicon variance between P-Cores, sometimes the Single/Dual "preferred core" factory boosting algorithms do not result in peak voltages. Sometimes you have a runt-of-the-litter P-Core that is responsible for peak Vcore. So, you really need to get to know your CPU and all its cores and needs if you're trying to keep voltages down. But hopefully high voltages are far less of a concern with the new microcode.
 
14900k user here with my 2 cents:

I locked all p cores to 57 and e cores to 42. This keeps the voltage really stable and makes the enhanced turbo unnecessary. Only with the enhanced turbo and p cores at default I can reach the 6ghz on my 2 favorable cores. In multicore tasks, this makes no difference at all.
My e-cores were, as Scott said, causing high VID's. If I now go to P core 58 it will become unstable, so this seems to be my sweetspot.

As you have a KS model, I suspect you can get those speeds with a less agressive undervolt and higher power limit, but its not really worth the extra performance IMO, at least not now when its pretty overkill for almost anything besides benchmarking.

Ive got nearly the same settings as you have
AC DC 40
LLC6
-0.100 undervolt
During gaming and editing its so insanely fast and stable, I dont see any benefit of boosting it further, it might give me a frame or 2 extra in games, maybe 10 if its heavy on single core, and its just not worth the extra voltage or heat. Currently, I use about 1,32 volt during gaming, a far cry better than 1,47 it had out of the box.

For latency issues: try to disable core parking with QuickCPU or ParkControl, it solved A TON of latency issues for me. Keep C state on 10, for me that will lower the power usage when idle to 9 to 11 watt.
 
For latency issues: try to disable core parking with QuickCPU or ParkControl, it solved A TON of latency issues for me. Keep C state on 10, for me that will lower the power usage when idle to 9 to 11 watt.
Yep. QuickCPU is what Vassil_V described in his user comment in the survey. C-States on 10 is a good idea, but for some reason some people will have stability issues when changing the defaults. Just a heads up.

Lots of good info in MigraineFilm post. There are definitely many ways to tune Raptor Lake. My survey database is starting to illustrate that conclusion. If you like tinkering and pushing things to the limit then my suggestion to test different approaches can be fun even if you end up at about the same place. However, if you‘re trying to get to a conclusion fairly quickly then any of the options can work well, e.g. LLC5 + AC15ish +CEP, LLC6 + AC35ish [both becoming very popular], LLC=Auto(8) + AC80ish +CEP [popular for beginners]. All can work with manual or factory core ratios/boosting.
 
Last edited:
AC/DC =15
LLC = 5
Adaptive with a -0.085

100% utilization , my vcore was 1.307

10 run back to back, 42,500 - 42,870
 

Attachments

  • 42871.png
    42871.png
    36.6 KB · Views: 13
Last edited:
I do rather like that LLC5 approach. For some reason, the KS chips seem to like it. Did you enabled CEP yet? You should have no issues.
Now the real test will be to play some extended gaming sessions with a few different titles, including installing new Nvidia drivers to force Shader compiles.
If you run into stability issues, try stabilizing things by backing off the manual UV (maybe 15mV less) rather than touching the other settings.
Keep us posted!
 
spent the day swapping between windows versions.

21h2 saw a top score of 43150
24h2 saw a top score of 42550.
- so 600 point diff between current build of win 11 and the glory performance version.

All other settings the same.(57x)

I did a quick few on 24h2 at 58x instead of 57, and saw 42,880.

Still haven’t had a chance to swap the thermal paste yet.

43k might be a hair out of reach for me on the latest windows build. But I at least know what it could do on 21h2 before I swapped back to latest win11 build
 
Last edited:
Arctucas, I think me might have a potential future winner of the Raptor Lake Survey. ;)

I believe you are correct.

I have never pushed my 14900K past 6000MHz all-core with HT enabled. As I recall, I broke 41000 in CB R23.

This winter, I can set my rads out of the window when the air temps are around freezing and give it good run or two.
 
Back
Top