13700kf Overclocking, is it supposed to be this hot?!?

[grendel0501]

I recently finished putting together a new rig.

Intel 13700kf
Gigabyte z790 Arous Elite
32gb G-Skill
2tb m.2
Gigabyte 4090 Gaming OC

The 4090 is a HUGE graphics card compared to my previous 2080ti, it barely fits in the case. Also I used a piece of PVC as a 'anti-sag' brace.

I've spent the last few days working on mounting fans and finishing up the install.
I've been running CoreTemp randomly throughout the reboots to see if my Noctua NH-D15 is sat properly and from the idle temps, I thought I had a pretty good seating.

But as soon as I put on CineBench or Prime95 the temperatures jump almost instantly into the 85-100 degree range.
Also the temperatures are not even at all, there is alot of variation between them.

I've never had a CPU got this hot, so I resat the heatsink, thinking the thermal grease was spread in a bad way, but after the 2 reseats the temperatures are still crazy high, and the speed at which they get there is almost instant, no build up like past processors. Its like the heatsink isn't there at all.

And this is WITHOUT a overclock. Just stock settings....

I think I am hitting Tjunction for the first time in my life.




Whats going on here?

Are the Raptor Lakes just this hot?

The only difference between this and my other rigs, is I tried mounting the heatsink horizontal this time (blowing air up instead of at the back of the case).
But the seating seems good, I don't think that is the issue.

 

[EarthDog]

Yes. They are just that hot/it's how today's processors work. The temperature variances are the P-cores (hottest) and E-cores (cooler). What you're seeing is normal. You can try a negative voltage offset and see if that helps.

Also, processors, when running P95 or Cine, were always an 'instant' to temperature as I recall it... Just a second to so and bam. there's a little creep, but, they ramp up really fast.

I'd try front to back airflow as you're just hoovering a lot of that hot air off of a 450W card when gaming.

 

[grendel0501]

Yes. They are just that hot/it's how today's processors work. The temperature variances are the P-cores (hottest) and E-cores (cooler). What you're seeing is normal. You can try a negative voltage offset and see if that helps.

Also, processors, when running P95 or Cine, were always an 'instant' to temperature as I recall it... Just a second to so and bam. there's a little creep, but, they ramp up really fast.

I'd try front to back airflow as you're just hoovering a lot of that hot air off of a 450W card when gaming.

This is normal?!?

I wanted to try to lock the P cores at their Turbo frequency for what I thought would be a 'simple' overclock, but I don't think thats a good idea with it getting this hot.

With the same air cooler, NH-D15 I am able to keep my OC'ed 9700k below 75 degrees on stress tests... But this thing is hitting 100.. Yikes.

 

[EarthDog]

9700k is a 95W chip. Yours is 150W base and up to 253W turbo...

The chips are good to that hot. If you want to lock them, lock them and see what voltage you need to be stable. Remember, a stress test is just that... absolute worst case... gaming is going to be 20C less easy.

 

[grendel0501]

9700k is a 95W chip. Yours is 150W base and up to 253W turbo...

The chips are good to that hot. If you want to lock them, lock them and see what voltage you need to be stable. Remember, a stress test is just that... absolute worst case... gaming is going to be 20C less easy.

Lock as in turn off turbo?
I don't want to do that. I want to atleast be able to run stock speeds, if not more.

I was just hoping for some low temps at stock clocks, so I could push the chip alittle further.
At 100c doesn't the chip turn down its own speed/voltage?

 

[EarthDog]

Lock as in, what you said.......to turbo clocks.

It throttles at 100C, yes. try a negative voltage offset. Or you can set it manually. I run ~1.35V for a 5.6/4.5 GHz (from 5.5 and 4.3) overclock. You lose some single/dual thread speeds (turbo boosts to 5.8), but there I don't throttle either.

 

[grendel0501]

This is the stuff of nightmares. lol

 

[EarthDog]

It's the stuff of....these are just about maxed out of the box. It's ok, these temps and clocks, etc.

 

[grendel0501]

My worry is that each generation is going to get hotter, and chips will end up being too hot for anything other than a large custom water setup.

Last I checked the NH-D15 was the best air option available, and its showing its limitations here.

 

[EarthDog]

Well, it's not so much the cooler as it is the CPU itself. Getting the heat out of the more dense (transistor count) die, is the issue we've been running into over the last few generations. Now that AMD and Intel are 'maxing' their chips out from the factor, it's more apparent then when then ran these closer to their sweetspot.

 

[dejo]

I have my 13700k undervolted to -.14 and it runs cinebench23 for a single run with a high core of 78c in a 75f room. This is in a mini itx case with an EK 280AIO for cooling. with some clocks thrown in, I can manage 5.7ghz on an all core load with cinebench 23 before I really get into the higher 90c range.
I would venture that if you bring your voltage down the temps will drop nicely too

 

[stompah]

Are your temps with the side panel on? Was this combo ever run with the 2080TI?

You might just be at the max capacity of what your case can handle. And as ED pointed out pulling hot air off the backplate of the 4090 is not helping.

 

[grendel0501]

Are your temps with the side panel on? Was this combo ever run with the 2080TI?

You might just be at the max capacity of what your case can handle. And as ED pointed out pulling hot air off the backplate of the 4090 is not helping.

It doesn't make much of a difference to take my panels off, my case is like one of those hurricane simulators in the mall.

Post magically merged: Nov 8, 2022

I may try the vertical config on the heatsink to see if it helps at all.

I ordered some Kryonaut thermal compound to replace the included NT-H1.

Is de-lidding still a thing?

I remember doing that to a Q6600 and getting -10c on all cores.

 

[EarthDog]

I may try the vertical config on the heatsink to see if it helps at all.

Horizontal. It's currently vertical... or blowing the air up (right?).

Thermal paste may help a degree or two if youre lucky.

 

[grendel0501]

Yeah, its currently blowing up.

 

[Johan45]

I would switch that cooler direction and undervolt the CPU as suggested. In our front page review, overclocking the 13900K, I got an extra 200 MHz (5.6) on the P-cores with a negative offset leaving everything else on auto. The stock voltage on these CPUs is quite a bit higher than it needs to be.

 

[bwanaaa]

Der8auer had a youtube video a couple of weeks a go where he tamed a 13900k simply by limiting the power draw to 90 watts (a setting in the bios) with minimal reduction in performance.

 

[Zerileous]

Well, it's not so much the cooler as it is the CPU itself. Getting the heat out of the more dense (transistor count) die, is the issue we've been running into over the last few generations. Now that AMD and Intel are 'maxing' their chips out from the factor, it's more apparent then when then ran these closer to their sweetspot.

I think it's also how temperature is measured. Not only is the thermal sensor getting architecturally closer to the parts of the CPU doing the work, the distance is that much smaller because of the increased density. 20 years ago we put a thermocouple *next to* the bare die, or even had the board measure temperature under the socket. Eventually the sensors got moved onto the cores themselves. What I'm trying to say, because I'm not super technical, is that it's all relative. If the engineers say 100c is okay, then who's to say last generation's 90c isn't equivalent to this generations 100c because the point of measurement is closer to the hottest part of the core. 40c used to be dead CPU territory, but we were not measuring from the same point so the same temperature scale can't be used.

I'm just saying you can't compare measurements if the technique used to measure is not consistent. And the technique used to measure changes every generation, both in terms of hardware and software. So we are left to rely on the engineering specifications, even if something "feels wrong" based on the temperatures we are familiar with.

 

[EarthDog]

To me, it doesn't matter where the sensor is or 90C or 100C. If that's what the BIOS uses to say 'no mas' than it is. It's intrinsically harder to cool these chips because of the increase in power use, transistor density, and die thickness. The fact that they can throttle higher or the temperature sensor reads from X, doesn't change the physics of getting the heat out.