3700X + NH D15 high temps

[Steelbom]

I'm seeing 50-60 idle and 60-70 degrees celsius with very light usage, like <10% CPU usage. Both CPU fans spinning at 800-1200 RPM. Two AP181's at 800 RPM.

I have the high performance setting enabled in Power Plans, so I expect it would be higher than if it were set to balanced.

Playing Fortnite for a couple of hours pushes the temps up to 85.

Are these temps reasonable? Ambient would be about 26 degrees.

Perhaps some other users with NH D15's or similar could chime in?

Appreciate any input.

 

[deustroop]

OK At idle
1) what power (watts) does the cpu call ?
2) What are the clocks ?

 

[ehume]

The NH-D15 is about the top of the line as heatsinks go. But your idle temps suggest a poor cooling solution, even if the Ryzen 3700X is a heat hog (I know nothing about your CPU). I would suggest taking the D15 off, examine the mount. Make sure it is all correct and that you use good TIM. Then put it all back together and measure temps again. I would also suggest measuring the temps inside your case. No cooling solution can go below its antecedent.

 

[fleshconsumed]

I have Scythe Ninja 4 with a single Scythe GenteTyphoon AP14 fan. Under full 100% load on all 16 threads I get 62-65 degrees C with the fan spinning around 1300RPM. Don't know about idle temps since my CPU rarely stays idle.

Since Zen2 is a chiplet design, it's possible the old one drop of thermal compound in the middle of heatspreader no longer works. When I mounted my heatsink I spread a thin layer of thermal paste over the entire chip. I suggest you try the same, try to make it as thin as possible though.

 

[Markfw]

I have Scythe Ninja 4 with a single Scythe GenteTyphoon AP14 fan. Under full 100% load on all 16 threads I get 62-65 degrees C with the fan spinning around 1300RPM. Don't know about idle temps since my CPU rarely stays idle.

Since Zen2 is a chiplet design, it's possible the old one drop of thermal compound in the middle of heatspreader no longer works. When I mounted my heatsink I spread a thin layer of thermal paste over the entire chip. I suggest you try the same, try to make it as thin as possible though.

I do that every time. I spread it with my finger as thin as possible. Use it on air and AIO and my custom water setup.

 

[Kenmitch]

Since Zen2 is a chiplet design, it's possible the old one drop of thermal compound in the middle of heatspreader no longer works. When I mounted my heatsink I spread a thin layer of thermal paste over the entire chip. I suggest you try the same, try to make it as thin as possible though.

Chiplets offset so one needs to keep this in mind when selecting a cooler. I went with coating the entire base of my cooler when I installed it. Probably used a little too much, but figured a little squeezing out around the edges may help.

One should look at the base of the cooler before purchasing it. Just because it's AM4 compatible doesn't mean it'll work well with the 3xxx series.

 

[DrMrLordX]

I had an NH-d15 on my 1800x when it was putting out over 210W in y-cruncher/Prime95/etc. It kept temps lower than that.

@Steelbom

What kind of power is that chip pulling from the socket? Do you have an external power monitor (Kill-a-Watt) to monitor system power? We need to know more about how much juice your system is pulling when producing such temperatures.

 

[Steelbom]

Thanks for the replies everyone!

OK At idle
1) what power (watts) does the cpu call ?
2) What are the clocks ?

As of this moment after a fresh boot on a cold morning here are the deets:

1.475v max of 1.487. Pulling 25-32W. 4299MHz. Temp as I'm writing is 45c. But has been up to 61 and as low as 36. According to my PSU read out, 71W in, 60W out total power.

The NH-D15 is about the top of the line as heatsinks go. But your idle temps suggest a poor cooling solution, even if the Ryzen 3700X is a heat hog (I know nothing about your CPU). I would suggest taking the D15 off, examine the mount. Make sure it is all correct and that you use good TIM. Then put it all back together and measure temps again. I would also suggest measuring the temps inside your case. No cooling solution can go below its antecedent.

I used the Noctua TIM and was pretty thorough with mounting it but... I have read a couple of times that a little pea drop in the center of the IHS isn't enough with Ryzen.

I have Scythe Ninja 4 with a single Scythe GenteTyphoon AP14 fan. Under full 100% load on all 16 threads I get 62-65 degrees C with the fan spinning around 1300RPM. Don't know about idle temps since my CPU rarely stays idle.

Since Zen2 is a chiplet design, it's possible the old one drop of thermal compound in the middle of heatspreader no longer works. When I mounted my heatsink I spread a thin layer of thermal paste over the entire chip. I suggest you try the same, try to make it as thin as possible though.

That sounds preferable. I have two Noctua 140mm fans spinning at 1300+ RPM at around 80-85 degrees.

Yeah, I may need to re-apply. I've heard that an X shape can also be good.

I had an NH-d15 on my 1800x when it was putting out over 210W in y-cruncher/Prime95/etc. It kept temps lower than that.

@Steelbom

What kind of power is that chip pulling from the socket? Do you have an external power monitor (Kill-a-Watt) to monitor system power? We need to know more about how much juice your system is pulling when producing such temperatures.

I downloaded cinebench and I'll also run Fortnite quickly to test.

10 mins of FN resulted in temps of 64 - 73. Voltage hit 1.5v on a couple of cores, max package power was 61W. Total power draw 270W in 250W out.

In Cinebench R20

All core clock speed was 3949MHz, 1.118 VID across all cores. CPU hit 75c. Power draw from package was 90W.

 

[deustroop]

Not an amd user myself, but it looks to me like it could be that the intel MCE equivalent is engaged by the mainboard. Its called Precision Boost Overdrive I believe. What suggests that condition is the idle power/heat profile. If you are recording a cpu draw of 25-32 w and thermals of 60c at idle they are both very high for normal use but would be MCE territory on a 9900K which I have.

 

[ehume]

It looks like something is not drawing off heat the way it should. I say again, remount your heatsink

 

[IEC]

Core area of each chiplet is approximately 30mm^2. Couple the tiny die surface area with aggressive boost behavior ("race to idle") and 1ms clock and voltage adjustments and you have GPU-like heat density and fluctuations.

You will have higher temperatures versus what you are used to on 14nm CPUs.

 

[Indus]

1.475v seems high

 

[Steelbom]

Not an amd user myself, but it looks to me like it could be that the intel MCE equivalent is engaged by the mainboard. Its called Precision Boost Overdrive I believe. What suggests that condition is the idle power/heat profile. If you are recording a cpu draw of 25-32 w and thermals of 60c at idle they are both very high for normal use but would be MCE territory on a 9900K which I have.

Ahh I see I see, thanks.

It looks like something is not drawing off heat the way it should. I say again, remount your heatsink

Yeah, I think I will. Bit scared, considering when I took the stock cooler off it ripped the CPU out of the socket and I bent some pins, but don't think that'll happen with my Noctua paste.

Core area of each chiplet is approximately 30mm^2. Couple the tiny die surface area with aggressive boost behavior ("race to idle") and 1ms clock and voltage adjustments and you have GPU-like heat density and fluctuations.

You will have higher temperatures versus what you are used to on 14nm CPUs.

I don't have the Ryzen power plan selected (in fact, it's not even an option) so will I be benefiting from the 1ms clock/voltage adjustments? Or do I need that option?

Yeah you have a point the temperature does jump around rapidly depending on what's happening. I'm used to a 14nm 4c 4t so this is a big change.

1.475v seems high

Just ran cinebench with Ryzen Master open. Voltage dropped from 1.45 to 1.22 under full load. All cores at 3933MHz and temps capped out at 73. Dropped back down to the 50-60 range. It bounces around a lot.

 

[IEC]

1.475v seems high

Up to 1.5V for single/lightly-threaded loads is within spec and won't degrade the chip beyond expected limits. All-core is a different matter. Voltage, amperage, and resistance all matter when worrying about degradation. I wouldn't recommend more than 1.3V for an all-core OC. Probably less, especially with aggressive LLC (read: voltage spikes). Most users will have better results just running stock and letting the CPU and motherboard figure out what speeds to boost to depending on the load.

Ahh I see I see, thanks.

Yeah, I think I will. Bit scared, considering when I took the stock cooler off it ripped the CPU out of the socket and I bent some pins, but don't think that'll happen with my Noctua paste.

I don't have the Ryzen power plan selected (in fact, it's not even an option) so will I be benefiting from the 1ms clock/voltage adjustments? Or do I need that option?

Yeah you have a point the temperature does jump around rapidly depending on what's happening. I'm used to a 14nm 4c 4t so this is a big change.

Just ran cinebench with Ryzen Master open. Voltage dropped from 1.45 to 1.22 under full load. All cores at 3933MHz and temps capped out at 73. Dropped back down to the 50-60 range. It bounces around a lot.

That's normal behavior. All-core loads use much lower voltage by default. Since you are seeing 1.22V on all-core loads it's likely you don't have PBO enabled (a good thing, PBO currently just increases heat for little to no gain).

If you want to take advantage of the 1ms clock/voltage adjustments you need to install the latest AMD chipset drivers for your motherboard. Those will install the Ryzen power plans.
https://www.amd.com/en/support/chipsets/amd-socket-am4/b450

 

[Steelbom]

Up to 1.5V for single/lightly-threaded loads is within spec and won't degrade the chip beyond expected limits. All-core is a different matter. Voltage, amperage, and resistance all matter when worrying about degradation. I wouldn't recommend more than 1.3V for an all-core OC. Probably less, especially with aggressive LLC (read: voltage spikes). Most users will have better results just running stock and letting the CPU and motherboard figure out what speeds to boost to depending on the load.



That's normal behavior. All-core loads use much lower voltage by default. Since you are seeing 1.22V on all-core loads it's likely you don't have PBO enabled (a good thing, PBO currently just increases heat for little to no gain).

If you want to take advantage of the 1ms clock/voltage adjustments you need to install the latest AMD chipset drivers for your motherboard. Those will install the Ryzen power plans.
https://www.amd.com/en/support/chipsets/amd-socket-am4/b450

Thanks for the info!

PBO is on "auto" so I'm not sure if it's on or not. Apparently turning it on voids the warranty so I'm not touching it.

Just installed the Ryzen High Power plan, might test out the balanced plan later as well.

Also I've disabled AMD Cool n Quiet in the BIOS. Temps were still high before this and I'm not sure I've noticed any difference. What does that actually do and is it worth keeping on?

 

[Kenmitch]

Just ran cinebench with Ryzen Master open. Voltage dropped from 1.45 to 1.22 under full load. All cores at 3933MHz and temps capped out at 73. Dropped back down to the 50-60 range. It bounces around a lot.

What monitoring software are you using to see voltages and temps?

The reason I ask is I mainly use HWiNFO64 and it shows a CPU core voltage along with a vcore voltage. If a person views the CPU core voltage they'll see mainly the voltage hovering around 1.4v's at idle, but it'll drop when cores are loaded. If a person views the vcore they'll see the actual voltages shown by cpu-z as it looks to be more real time. Both of these voltage readings show pretty much the same when stressing all the cores with something like CB R20. If a person just idles in Ryzen Master and views the voltage it matches the CPU core voltage, however if one watches the cores do their thing dropping clocks really low, going to sleep, etc one would believe what is shown as vcore is more realistic/believable.

 

[PUN]

A lot of the poeple here with the responses dont have Ryzen 3000 chips.
These chips are all idling at high 40's and 50's, maxing at 80's under full load. It has nothing to do with TIM nor how your Noctua is seated. Irratic behaviour of Ryzen MB with CPU with wide range of voltage is causing your temps to spike. Hopefully AMD will patch this soon. Meanwhile, my advice is limit your max boost with power draw. This will limit your performance by 2-3% but your temps will drop by 10-20C

 

[Steelbom]

What monitoring software are you using to see voltages and temps?

The reason I ask is I mainly use HWiNFO64 and it shows a CPU core voltage along with a vcore voltage. If a person views the CPU core voltage they'll see mainly the voltage hovering around 1.4v's at idle, but it'll drop when cores are loaded. If a person views the vcore they'll see the actual voltages shown by cpu-z as it looks to be more real time. Both of these voltage readings show pretty much the same when stressing all the cores with something like CB R20. If a person just idles in Ryzen Master and views the voltage it matches the CPU core voltage, however if one watches the cores do their thing dropping clocks really low, going to sleep, etc one would believe what is shown as vcore is more realistic/believable.

I'm using HWMonitor (but I didn't have that extra section open on the right like you do) and Ryzen Master (not at the same time). Thanks, good to know.

A lot of the poeple here with the responses dont have Ryzen 3000 chips.
These chips are all idling at high 40's and 50's, maxing at 80's under full load. It has nothing to do with TIM nor how your Noctua is seated. Irratic behaviour of Ryzen MB with CPU with wide range of voltage is causing your temps to spike. Hopefully AMD will patch this soon. Meanwhile, my advice is limit your max boost with power draw. This will limit your performance by 2-3% but your temps will drop by 10-20C

Does that mean lowering PPT in BIOS? What should it be set to approx?

 

[coercitiv]

@Kenmitch would do mind testing your idle temps & CPU Package Power with sleep states disabled? (if MSI X570 BIOS is anything like previous AMD versions you should be able to flip the switch in BIOS )

Tech Yes City did a 3600 review and talked about their sample also exhibiting high idle temps, but he also showed monitoring software with idle Package Power being tracked around 30+W. From what I see your CPU idles around 18W. To me this would explain the idle temperature delta and looks like different software configs which may prevent the CPU Package from entering sleep states properly.

 

[lopri]

What everyone says seems correct to me:

1. You have to account for the MCM design when you spread thermal interface material. Grain size drop in the center is inadequate.
2. There are BIOS versions that give extra voltages to the chips. See you can find more recent BIOS for your board.
3. Manually limiting the board's power supply to the CPU will lower the temperature by quite a lot. You may have to find a sweet spot because undervolting too much will have negative effect on performance.

 

[IEC]

What everyone says seems correct to me:

1. You have to account for the MCM design when you spread thermal interface material. Grain size drop in the center is inadequate.
2. There are BIOS versions that give extra voltages to the chips. See you can find more recent BIOS for your board.
3. Manually limiting the board's power supply to the CPU will lower the temperature by quite a lot. You may have to find a sweet spot because undervolting too much will have negative effect on performance.

You can also manually enable PBO and instead of raising PPT and EDC limits, lower them to maximize efficiency.

 

[DrMrLordX]

10 mins of FN resulted in temps of 64 - 73. Voltage hit 1.5v on a couple of cores, max package power was 61W. Total power draw 270W in 250W out.

In Cinebench R20

All core clock speed was 3949MHz, 1.118 VID across all cores. CPU hit 75c. Power draw from package was 90W.

That is not a lot of power output. Even taking power density into account, hotspots in a "GPU-like" situation on TSMC 7nm shouldn't be much higher than +25-30C than die edge temperature. A stock NH-D15 can handle heat loads of 180W with aplomb. The "cool" parts of the chiplets should be very close to ambient. I see no reason why your hotspots would be getting as high as 75C (assuming that's what's being reported). Also, package power should be chiplet + I/O die combined (CPU + SoC) so the power output from the chiplet alone should be even lower than what is reported.

I suspect a bad mount or something.

 

[Kenmitch]

@Kenmitch would do mind testing your idle temps & CPU Package Power with sleep states disabled? (if MSI X570 BIOS is anything like previous AMD versions you should be able to flip the switch in BIOS )

I'll take a look and see tonight after work. I don't remember seeing any setting, but then again I wasn't looking to turn them off.

I think it boils down to the uEFI version and how it's been implemented by the MB makers. MSI looks to be holding strong so far at least with the uEFI that was released at launch. AMD ComboPI1.0.0.3a is what my MB shows. I'm hoping they don't jump the gun and release a buggy uEFI for my MB. My gut told me to go MSI for this adventure and I'm hoping it was correct....So far so good.

Now if I can only shake off the demon in my head that's telling me to pick up a 3700X I'll be in good shape. I've been saving up my BestBuy rewards and currently have $45 accumulated in my account....My local store has 1 in stock which isn't helping the battle at all.

 

[ehume]

PBO=Precison Boost Overdrive. What does that mean?

 

[moinmoin]

PBO=Precison Boost Overdrive. What does that mean?

It's overclocking by removing/changing the limits that govern PB, as such it voids the warranty.