Performance way down with lower voltage, while running a stable 4.9 GHz non-oc turbo. 12900K / Z690 Tomahawk Wifi DDR5

[Sjaakie]

Hi all,

 

I'm running a system with a 12900K / MSI Tomahawk Z690 Wifi DDR5 / 64 GB kingston Fury 4800 mts / Noctua NH-D15.

 

Its running pretty hot as usual, so im looking into tuning it down a tiny bit. Ive put the 'cpu lite load' setting on mode 8 instead of 9 (auto). This setting tunes down the voltage at higher loads. It runs stable and a bit cooler, as expected, but the strange thing is that performance is way down, all while the cores are running at a stable 4.9 all core boost.

 

Cinebench R20 goes from around 10500 to 8500. Turning the setting to mode 7 results in even lower score of 6500 (also while it keeps running at 4.9 Ghz).

 

I have done the same on a comparable system with another motherboard, but this one had no problem hitting the full 10500 cinebench score. It just crashed when putting the setting / voltage too low, as expected.

 

 

 

It seems a bit like a bug to be honest. I mean 4.9 is 4.9 Ghz, if it has a problem with a lower voltage id expect it to crash, not to run at a seemingly same speed but show way lower performance.

 

Ive tried it with the latest stable non-beta bios, and 2 versions earlier that it came with outside the box. Manually setting the voltage to a -0.025 offset crashes immediately / wont post.

 

 

If anyone has any guidance on why performance is this much worse while reporting a stable all-core 4.9 boost, or has any other thoughts on this I would be glad to know!

[Somerandomtechyboi]

You can try changing the cpu freq manually to 5g allcore with volt ~1.25v, only explanation i have is maybe the board makes the cpu run fast but it nukes the ring freq so it runs the same core but with nuked ring performance is also nuked

[unclewebb]

7 hours ago, Sjaakie said:

I mean 4.9 is 4.9 Ghz

Yes and no. I think 12th Gen CPUs have a new feature where they can be throttled internally. HWiNFO and most monitoring apps might report that the CPU is running at 4.9 GHz but when throttling internally, what the CPU is actually accomplishing is much less. Your Cinebench scores hint at what is going on.

 

One motherboard BIOS calls this new feature, Base Frequency Boost.

 

 

Intel CPUs have a feature called Clock Modulation throttling that they used to use. I think a new similar feature has been added for 12th Gen. I have not seen any mention of this in the public documentation but secrets like this are usually not released to the public.  

 

When Cinebench is in progress, does HWiNFO report both the clock speed and the Effective Clock speed at 4.9 GHz for the performance cores?

 

Try running ThrottleStop 9.4.6

https://www.techpowerup.com/download/techpowerup-throttlestop/

 

Post a screenshot of the main ThrottleStop window and the TPL window while Cinebench is in progress and your computer is running slow. It might show something that HWiNFO is not showing. 

[Sjaakie]

@SomerandomtechyboiThanks for chipping in! Ive checked the Ring freq but it seemed to be equal in both scenario's. Though something strange is certainly going on (at least that I dont understand). I've gone a different route which seems to work for now, leaving the CPU lite load alone and working with advanced adaptive voltage. The Vcore runs on auto until it hits 4.2 ghz and then a negative offset kicks in. It's churning along at 1.125 V / 200w / high 70's C and getting the Full Cinebench R20 / R23 score. I think ill put it a bit higher for stability (1.2 Im thinking) but im surprised its finishing the Cinebench Runs without issue at this voltage.

 

@unclewebbYou're certainly onto something here! Ive lookes into Effective clockspeeds and the E-cores are way down to around 2.1 Ghz (while 'normal' clocks stays at 3.6). Effective P-core stays at 4.9 though. Throttlestop is showing the same results as the effective cores in HWinfo:

 

 

Still something is off here though, turning off E cores entirely (and keeping everything else on auto) scores 7500, while turning cpu lite load on mode 7 like in the throttleshot screenshot scores 6500. Looks like this internal throttling mechanism is still at play here. Good to know that this can be an issue.

 

[unclewebb]

@Sjaakie

When ThrottleStop shows a multiplier of 22.26, the E cores are definitely throttling. This new limit might not show up in the Limit Reasons data that ThrottleStop and HWiNFO report. Is the ThrottleStop Limit Reasons window all blacked out when this is happening? 

 

When the E cores are being throttled like this, it is possible that the P cores are throttling too, just not being detected as throttling. 

 

17 hours ago, Sjaakie said:

'normal' clocks stays at 3.6

At least you found a way to figure out when there is a throttling problem. Changing the voltage does not usually change performance any significant amount. It does not cause the big slow downs that you are seeing. This is typically a power or current limit of some sort, throttling the CPU.