Hey all!
I just upgraded my Z690-based system from a 12600K to a 14700KF.
Seeing how this chip is listed with a max turbo power-draw of 253 W, and I do know from reviews and tests that my 240mm AiO can handle 250 Watt (and my case has pretty great airflow), I wasn’t too worried about the added wattage.
Until I ran Cinebench 24 to test the new CPU.
Cores were maxed out all the time, naturally, but according to HWMonitor, none of the P-cores ever boosted to their stock maximum of 5.6 GHz. All of them had a recorded peak of 5.5 GHz.
Thing is that after a few minutes, multiple P-cores started to show some red (100°C), so I got scared and aborted the run after… ~3 - 4 minutes. What really weirded me out though is the fact that power-draw peaked way above what I would’ve expected. It went up to nearly 300 W (299.5). Which is quite a lot more than the advertised 253 W - especially since the CPU never reached its maximum frequencies. I should probably add that it wasn’t just a spike in power - I was keeping a close eye on power-draw the whole time and it never went below the high 270s. Which is still more than 20 W higher (most of the time more than that) than what I would’ve expected without overclocking.
Needless to say this was a stock run (other than having XMP enabled).
Some specs:
Board: MSI Z690 Tomahawk DDR4 (latest BIOS with 14th gen support is installed)
Memory: 2x16 GB G-Skill Trident Z 3600 CL16
Case: Fractal Design Pop XL Air RGB
Fans: 2x 140 top exhausts, 1x 120 rear exhaust. Front has the 2 120 for the AiO and the rad plus one of the crappy 120s that came with the case, all as intakes. Temps were great with the 12600K installed (and that was without the additional front intake), so I’m pretty sure I have good enough airflow.
Anyone know what might be going on here?
Should I perhaps activate the power-limit in the BIOS? I have it on “unlimited” right now … well, technically speaking it’s not unlimited but some ridiculous value like 4000+ Watt on this board, but still … :D
S.
Needless to say this was a stock run
try LiteLoad 1 in bios
I also recently updated from a 12700k to a 14900k on a aorus z690 board. I only noticed it when going into XTU that power limits were too high even in stock bios settings. Using 330w at stock clocks of 5.7ghz all core in cinebench. I found that along with disabling MCE, I also needed to set the turbo power limits to Intel POR. This now correctly does power limiting to 253w but I still get 5.7ghz all core.
I think the default auto settings on the z series boards seem to ignore the limits in favor of having stability/compatibility with all performance levels of cpu.
Thanks, everyone!
Believe it or not, but in over 20 years of PC-gaming, this is my first venture into higher end CPUs (never went above i5-levels before this 14700KF), so I need to do some reading up, I think.
Re the AiO: I’m pretty sure it was one of the more popular HW-channels (GN, perhaps?) who had tested both 360s and 240s using a load of 250W. The Lian Li 240 I have was up there with larger ones in their ranking, so I thought I’d be good.
Having just checked real world performance in Baldur’s Gate 3, Act III (which is pretty CPU-intensive), it seems to have worked, so far. The CPU (well, one or two P-cores) did boost up to 5.6 GHz plus performance in the city was markedly improved. Load on individual cores was a lot lower whereas with the 12600 I was on the mid 90%s at times on one or two cores.
Plus the GPU now boosted higher (I have it OCed, but before it never went all the way to the maximum frequency I set) and my FPS (once I took the limiter off) were quite a bit higher and more stable, too. Before it was still very smooth, mind you, but they were dipping below 60 at times, now they were always above 60 and went up into triple digit territory, even while walking around the city with “dynamic crowds” enabled. CPU drew a little more power than the 12600, but that’s to be expected.
Temps while gaming were totally fine and not any higher than with the 12600 - in fact, I’m pretty sure they were actually a tad lower.
This is still with the 288/285W limit set in the BIOS, BTW… I’ll reboot and change to the lowest setting once I’m done typing here.
Besides that, I should probably also check the Turbo Boost/Duration settings in the advanced CPU config screen and make sure they match Intel specs.
This board doesn’t have “detailed” settings for the cooling solution. Like I said earlier: There are just three “generic” ones (“stock cooler”/~250W, “tower cooler”/~285W and "liquid cooler/~4096W).
From what I just read on the MSI-forum, these motherboards generally seem to be set pretty aggressively out of the box. On mine for example, it’s set to the liquid cooler option by default - not sure about individual PL1/2 settings, though.
Not a huge issue with the 12600K, I suppose, since that thing at stock speeds can’t draw near enough power to cause super high temps. But it seems the i7 is a bit more greedy when it comes to juice … :)
S.
Did you noticed much of a performance change from the 12700k to the 14900k?
Massive difference when playing at 1440p with up to 60fps increase in some games.
A 240 does NOT handle 250Watt… Where’d you read that? Maybe with some push-pull fan combination at jet-power noise levels.
YES you should enable powerlimits. Jeez. Just set some sane settings like PL1=220W, PL2=200W and tau=30 seconds. This is what the motherboard does anyway if you just set ‘cooler’ to 240mm (most motherboard have these options). They’ll only enable 250 or unlimited once you set cooler to 360 or more. You might lose 5% performance OMG (and 0% in actual real-world gaming). You regain the performance and then some once you apply an undervolt of 50mV or whatever you can do, and then it’ll run cool and quiet.
What is weird about this? When you run cinebench you want two things to happen.
- get the highest score. (Most important)
And
- use the most power (it is a benchmark tool after all, useful for troubleshooting and other things)
The power values you get from software aren’t accurate.
At all core load, you may not see peak single core burst. The CPU itself knows when it needs to hit peak frequency. But because your load doesn’t require it, it won’t hit it.
Instead it knows you demand all core loading at high power so it shifts to the optimal frequency and power usage for your power virus load.
Run a lighter higher frequency load like gaming with unlocked FPS and good memory and you should see it hit its advertised frequency.
MCE on?
Why are you ‘scared’? Once hitting 100 the CPU will just throttle and reduce turbo clocks to a sustainable level, if you want a score, leave it for 10 minutes.
As for 240mm, they’re usually not enough for such CPUs, I would consider 280mm the bear minimum if you are constantly under heavy load, a top notch thermal paste like Kryonaut also helps but if you plan to game, you’ll be fine with a 240 or an air cooler.
MCE on?
Yup.
Why are you ‘scared’? Once hitting 100 the CPU will just throttle and reduce turbo clocks to a sustainable level, if you want a score, leave it for 10 minutes.
Well… I wasn’t quite sure about the thermal limits of these things and you gotta admit… this does look a bit scary, doesn’t it? :D
It also weirded me out that the CPU would hit that high of a power-draw without even boosting to its stock max - when Intel claim ~50 W less peak turbo-draw.
S.
If you turn on MCE it ignores the Intel stock power limits and sets them to 4095 Watts.
If you want to adhere to the Intel defaults you’ll have to keep MCE disabled. But also be aware the 253W limit typically won’t allow to hit all-core boost at 5.6Ghz.
I had to undervolt my 13700K to keep it below the 253W limit, which is possible but probably harder for your chip with the extra E-cores.
Thanks!
It’s not really called MCE in an MSI BIOS, I think - not sure what it’s called, but the three options/labels you can set are, basically, three different generic cooling solutions (think they call them “CPU Cooler”, “Tower Cooler” and “Liquid Cooler” … or something along those lines.
I did go back into the BIOS and changed the setting from “Liquid” (which is 4000-something W) to “Tower” which is… err… 288? 285W? Did another run in Cinebench 24 and let it finish this time. CPU never drew more than the value I had set in the BIOS. I’d say with that setting, my AiO is at the upper limit of what it can reliably cool … like 250 to 270W. Beyond that, it gets a bit overwhelmed.
CPU didn’t boost any higher than 5.5 GHz this time, either and three of the P-cores did hit 100°C again, but only for fractions of a second. Since I won’t be running this thing at constant “all-core 100% loads”, I *think* I should be good - but perhaps I’ll just set it to the lowest settings (255W IIRC), just to be safe. At least until I can scrounge up the cash to get a 360mm AiO… :)
Although I do wonder whether even a 360mm could keep temps truly in check in Cinebench24 - especially with the “no limits” MCE-setting. I hadn’t used that latest version of it before today and it seems to be quite a bit more demanding than the version I had used before.
Neither Time Spy (go figure :D ) nor 3DMark’s CPU Profile pushed the CPU as hard as CB24 did - not even by a long shot.
I did see slightly higher peak power during some of the CPU-tests in Passmark’s performance test. Think it went up to 340 W for a second or so… yikes!
Still: None of this explains why the CPU will hit these high power-draws when it’s not even boosting to its stock max turbo frequency. I mean… 300+W and the P-cores didn’t even go up to 5.6 GHz for even a second.
*shrug*
:)
S.
13th gen and 14th gen out of the box are like this and the 240mm usually cannot cool over ~200w
check if lite load is set to 9 and not 12 and you can tune it down (I found 3 to be the best)
you can also turn mce off and tune pl1 and pl2
5.6 should be the single/two core boost. 5.5 is the all core boost so thats correct. Mobos have power limits off by default so its boosting to 300. Nothing to worry about its safe to hit 100c 300w in cinebench and nornal behavior. My chip does the same but can draw up to 340 watts. No issues running it like this since i got it in January . Cpus arent as fragile as people think theybare
