Higher Core Clocks, Lower Benchmarks?

[iPossum]

I finished my new build last week and yesterday I began my deep dive into CPU overclocking. I am good to hook when it comes to GPUs, but since I haven't owned a computer with an overclockable CPU since the Pentium M750 was new, some heavy, self-inflicted educating was necessary. Like, what's an E-Core? Y'knowhatImsayin. So I watched a lot, read some more, dug up a 12-year-old Gamers Nexus video, then I hit the forums and got after it. I experimented with both all cores and turbo ratio with groups of P-Core methods. Played with different C-States and power plans and upped the voltage, trying both adaptive and fixed modes, all the why keeping my eye on a myriad of gauges. HWiNFO64, Afterburner, CPU and GPU Z, typical s***.

What I'm struggling with is, as I got higher stable core clocks, my benchmark scores were going down. The biggest drop was going from 5.1 GHz up to 5.2 GHz. I even throttled up the fan curve for the higher clock run. My i7-12700K's average temp at 5,088 MHz was 57 C, at 5,187 MHz, I got the average temp down to 55 C (see pic), and still, there was a noticeable drop off in my benchmark run. My ambient temp is 66 F and I'm running a 360mm Lian Li Galahad as well as 5 additional Lian Li 120mm case fans. Heat is definitely not an issue. I s*** you not, I'm pretty sure that CPU has never been above 60 C during stress tests, games, file transfers or anything else I've tossed at it.

So, can anyone tell me what I am messing up here? Or why a CPU that is running a higher clock at a lower temp would cause benchmarks to actually drop. I hope someone can help me out here because I sure as hell don't get it.
Thanks!

i7-12700K
MSI MPG Z690 Edge Wifi
RTX 3070 FE
Corsair Dominator 32GB DDR5 @ 5600 MHz
Corsair RM850
Samsung 980 Pro 1TB x2

*The run on the right had lower benchmarks than the run on the left.

 

[ThrashZone]

Hi,
You didn't post any benchmarks from even the basic ones like
Cinebench R23 or R20

12 series like 11 and 10 have a lot of power limits that need to be disabled.

 

[iPossum]

Hi,
You didn't post any benchmarks from even the basic ones like
Cinebench R23 or R20

12 series like 11 and 10 have a lot of power limits that need to be disabled.

Hey, thanks for the reply man! No, I'm such a cherry to modern CPU overclocking I don't have anything from Cinebench. I will remedy that, I used 3D Mark CPU Bench, Time Spy and the CPUID Benchmark button. I didn't really think the actual scores were super important for this, just the fact that they went down. I can run them all again and post them, adding Cinebench into the mix, with the same settings I used last night and see what happens. I did disable all of the C-State/EIST/Intel Step whatever/power management options that I am aware of. I disabled turbo boost and it let it just run pegged at the set clock speed. Voltage I had set to 1.45 and it ran pretty consistently around 1.39 so I don't think I needed more. I feel like I'm just missing something stupid, I don't know. It just didn't make sense to me. Cooler temps + higher clocks = lower benchmarks. That seems... off.

 

[RJARRRPCGP]

Likely a sign of being unstable. Are you seeing any non-fatal WHEA errors?

 

[iPossum]

Likely a sign of being unstable. Are you seeing any non-fatal WHEA errors?

Would I be finding that in Event Viewer? Like Event Viewer>Applications and Logs>Hardware Logs?

EDIT: Nevermind, I found it. Applications and services> Microsoft>Windows>Kernel-WHEA. Checking now

Likely a sign of being unstable. Are you seeing any non-fatal WHEA errors?

These are the only WHEA errors I have. Event ID 20. Quick search, looks like something to do with a driver, but all of my drivers should be good. I've gotten this error a few times a day each day that I've gotten it. I cannot remember what I was doing on those days specifically. I have messed with different OC control almost every day the last week and this specific error hasn't shown up every day so I kind of don't think its directly correlated with and overclocking, although I could be way wrong about that.

 

[Mr.Scott]

Fouled out OS will dump bench scores too.
You have too many variables and not enough info for anybody to diagnose this yet.
Be specific. Much more specific.
And for god's sake, use a decent benchmark.

EDIT- from your other thread, bad memclocks will foul things quite quickly.

 

[eidairaman1]

There is a point of no returns on overclocking.

Moar cooling

 

[iPossum]

Fouled out OS will dump bench scores too.
You have too many variables and not enough info for anybody to diagnose this yet.
Be specific. Much more specific.
And for god's sake, use a decent benchmark.

EDIT- from your other thread, bad memclocks will foul things quite quickly.

"from your other thread, bad memclocks will foul things quite quickly."

Can confirm.

And thanks for the tips. Now that I know more of what to watch for, VRM/PCH temp and power draw, stuff like that, I'm going to run them again (and yes, I will be acquiring better benchmarks lol) and see how it plays out. I'll update the thread tomorrow with more specific information.

 

[ThrashZone]

Hi,
High frequency memory isn't always going to equal more performance
Work on timings at rated speed not sure what works on ddr5 I use asrock memory config utility to read timings used in the os
Then adjust them in bios.

Basic adjustments are lowering tREFI = DRAM Refresh Interval
Rest is tightening timings if not same as Cas latency example 16-16-16-36 for c16 if they are already tight you should be good.

Going to a higher frequency you might want to loosen them a little example 16-17-16-38
But mostly test each adjustment if scores go up you're doing good
If scores go down, bad.

 

[iPossum]

Hi,
High frequency memory isn't always going to equal more performance
Work on timings at rated speed not sure what works on ddr5 I use asrock memory config utility to read timings used in the os
Then adjust them in bios.

Basic adjustments are lowering tREFI = DRAM Refresh Interval
Rest is tightening timings if not same as Cas latency example 16-16-16-36 for c16 if they are already tight you should be good.

Going to a higher frequency you might want to loosen them a little example 16-17-16-38
But mostly test each adjustment if scores go up you're doing good
If scores go down, bad.

That sounds super interesting and also way beyond what I know about RAM right now. I had to spend the last day and a half with my computer torn apart to swap out and add some fans and then get the cables squared away but I’m getting back on the OC horse today. I’ve been reading a lot about intel chip power management. Haven’t delved into the ram yet. My clocks are way higher than that. Like 40-40-40-38 I think. What do you use to check performance boosts in ram? I have Karhu but that’s just error detection I think. Do you run, just like a game benchmark like you would for a GPU or CPU?

 

[ThrashZone]

Hi,
Quick test I just use cmd as admin and enter winsat mem

Then move to R23/ R20 these are really the best/ fast performance and stability tests to use

Then use blender open data this one is the best there is for stability at realistic workloads
Blender Open Data — blender.org

Then if all is good I move to several 3dmark benchmarks for kicks.

 

[Assimilator]

The answer is simple, your CPU isn't completely stable at those higher clocks. What is happening during the benchmark (or any high load) is that the CPU's internal mechanisms are detecting this and dropping the clocks to a point where stability is restored - and these dropped clocks are often extremely conservative, often way below what you know your chip is capable of. Whatever clocks that benchmark software is reporting are likely the maximum clocks it saw, not the average. It's also why you need to watch the clocks with a monitoring program while the benchmark is running, not just at the beginning or end.

The fix for this is generally more voltage to the CPU, but of course if you have reached the limits of what your particular chip can do, more voltage won't help. Once you get to that point it's generally time to start playing with sub-voltages like VCCSA, VCCIO etc., but that's generally a lot of fiddling for not much in the way of returns.

 

[Mr.Scott]

XTU benchmark is a rough test benchmark and it tells you directly what fails, if anything.

 

[iPossum]

The answer is simple, your CPU isn't completely stable at those higher clocks. What is happening during the benchmark (or any high load) is that the CPU's internal mechanisms are detecting this and dropping the clocks to a point where stability is restored - and these dropped clocks are often extremely conservative, often way below what you know your chip is capable of. Whatever clocks that benchmark software is reporting are likely the maximum clocks it saw, not the average. It's also why you need to watch the clocks with a monitoring program while the benchmark is running, not just at the beginning or end.

The fix for this is generally more voltage to the CPU, but of course if you have reached the limits of what your particular chip can do, more voltage won't help. Once you get to that point it's generally time to start playing with sub-voltages like VCCSA, VCCIO etc., but that's generally a lot of fiddling for not much in the way of returns.

Right on. That is kind of what I was wondering. I'm curious what causes a CPU to be unstable at a certain clock speed. Like if you have plenty of power and your heat is no issue, at what point is the CPU just, literally done working at higher speeds. Shallow thought brain says heat and power draw should be the only limiting factors but that would be way too easy. Based on that logic I would be able to get this i7 to like, 6 GHz lmao.

 

[Blue4130]

I finished my new build last week and yesterday I began my deep dive into CPU overclocking. I am good to hook when it comes to GPUs, but since I haven't owned a computer with an overclockable CPU since the Pentium M750 was new, some heavy, self-inflicted educating was necessary. Like, what's an E-Core? Y'knowhatImsayin. So I watched a lot, read some more, dug up a 12-year-old Gamers Nexus video, then I hit the forums and got after it. I experimented with both all cores and turbo ratio with groups of P-Core methods. Played with different C-States and power plans and upped the voltage, trying both adaptive and fixed modes, all the why keeping my eye on a myriad of gauges. HWiNFO64, Afterburner, CPU and GPU Z, typical s***.

What I'm struggling with is, as I got higher stable core clocks, my benchmark scores were going down. The biggest drop was going from 5.1 GHz up to 5.2 GHz. I even throttled up the fan curve for the higher clock run. My i7-12700K's average temp at 5,088 MHz was 57 C, at 5,187 MHz, I got the average temp down to 55 C (see pic), and still, there was a noticeable drop off in my benchmark run. My ambient temp is 66 F and I'm running a 360mm Lian Li Galahad as well as 5 additional Lian Li 120mm case fans. Heat is definitely not an issue. I s*** you not, I'm pretty sure that CPU has never been above 60 C during stress tests, games, file transfers or anything else I've tossed at it.

So, can anyone tell me what I am messing up here? Or why a CPU that is running a higher clock at a lower temp would cause benchmarks to actually drop. I hope someone can help me out here because I sure as hell don't get it.
Thanks!

i7-12700K
MSI MPG Z690 Edge Wifi
RTX 3070 FE
Corsair Dominator 32GB DDR5 @ 5600 MHz
Corsair RM850
Samsung 980 Pro 1TB x2

*The run on the right had lower benchmarks than the run on the left.
View attachment 239369

5.1ghz and under 60c should give you the idea that it's unstable. Surely with that OC you'd be getting higher temps if it actually spent any of the benchmark run at 5.1ghz.

 

[iPossum]

5.1ghz and under 60c should give you the idea that it's unstable. Surely with that OC you'd be getting higher temps if it actually spent any of the benchmark run at 5.1ghz.

I also thought it was running really cool. I assumed I just had a really good cooling set up lol. The average clock freq was, allegedly, pretty close to 5.1. I had to spend the last day and a half swapping cables and fans and rewiring, so I'll be jumping back on the benchmarks horse with several additions to my monitoring software.

 

[rcarlos243]

Under 60c on stress test at 5.1GHz looks way too low.

I have 12700K overclocked at 4.9GHz with 280MM AIO and I get under 80c on OCCT Large AVX2.

It is highly likely your OC is not actually stable.

Download Cinebench R23 (it is free) and do a multicore run to get a baseline score

Download MAXON Cinebench R23

Download Cinebench Release 23. Cinebench is now based on the latest Release 23 code using updated compilers, and has a minimum runtime activated by default (previously hidden in preferences). Cinebench is a real-world cross-platform test suite that evaluates your computer's hardware capabilities.

www.guru3d.com

You should get around 22K score.

If the score is way below, your OC is definitely not stable and you are probably clock stretching.

 

[iPossum]

Under 60c on stress test at 5.1GHz looks way too low.

I have 12700K overclocked at 4.9GHz with 280MM AIO and I get under 80c on OCCT Large AVX2.

It is highly likely your OC is not actually stable.

Download Cinebench R23 (it is free) and do a multicore run to get a baseline score

Download MAXON Cinebench R23

Download Cinebench Release 23. Cinebench is now based on the latest Release 23 code using updated compilers, and has a minimum runtime activated by default (previously hidden in preferences). Cinebench is a real-world cross-platform test suite that evaluates your computer's hardware capabilities.

www.guru3d.com

You should get around 22K score.

If the score is way below, your OC is definitely not stable and you are probably clock stretching.

Yea, when I actually got around to running R23 it was nowhere near 60° C lol. I got it up 24139. Using turbo ratios. 3 Cores at 5.3GHz, 6 Cores at 5.2GHz and 8 Cores at 5.1GHz. +3 e-core ratio offset, ring was 3.6GHz (up from 3.3). PL1 and PL2 both set to 300 W, max package power was 241 W. CoreVIDs and max Vcore were both around 1.31. Max package temp was 89° C. I disabled Enhanced Turbo and all of the TVB settings. Seems like those were the ones holding me back.

 

[unclewebb]

Here are some 12700K Cinebench R23 numbers you can compare to.

Intel Core i7-12700K Review - Almost as Fast as the i9-12900K

With the Core i7-12700K, Intel has released a formidable competitor to AMD's Ryzen 5800X and even 5900X. Thanks to eight powerful Golden Cove cores, the processor handles all workloads very well, including gaming. Compared to the i9-12900K, it runs almost as fast, but much cooler, with better...

www.techpowerup.com

Make sure your CPU maintains the full 5100 MHz while Cinebench is in progress. Set your BCLK to 100.000 MHz in the BIOS.

 

[Mr.Scott]

12700K vs. 12900K R23

Overclocking, overclocking, and much more! Like overclocking.

HWBOT is a site dedicated to overclocking. We promote overclocking achievements and competitions for professionals as well as enthousiasts with rankings and a huge hardware database.

hwbot.org

 

[iPossum]

Here are some 12700K Cinebench R23 numbers you can compare to.

Intel Core i7-12700K Review - Almost as Fast as the i9-12900K

With the Core i7-12700K, Intel has released a formidable competitor to AMD's Ryzen 5800X and even 5900X. Thanks to eight powerful Golden Cove cores, the processor handles all workloads very well, including gaming. Compared to the i9-12900K, it runs almost as fast, but much cooler, with better...

www.techpowerup.com

Make sure your CPU maintains the full 5100 MHz while Cinebench is in progress. Set your BCLK to 100.000 MHz in the BIOS.

Ill give it a shot. What should I be doing with the ring clock? I set it 3.6 just to match the i7 stock frequency. I wasn't sure if I went higher than that if things would get sideways when the ring clock was running faster than the CPU when it wasn't under load.