Question Need advice on 7950X build w/ NH-D15 high temperatures

[keithth]

Hello there,

Based on the excellent advice here, I've built a 7950x system with a Noctua NH-D15 CPU cooler(using both fans) in a Corsair 4000D Airflow case. Gigabyte x670 AORUS ELITE AX mobo with F20a firmware.

Full part list here

https://pcpartpicker.com/list/6k3H7R

I'm currently seeing temps under load using the free software Heavy Load, of around 92C within a few seconds on running the tests. The system in general seems relatively stable. The only "non-default" thing I've done is use some of the DDR5 features to run the RAM at 6400.

While under load, Core Speed viewed via CPU-Z, appears to be approximately 5140 mhz. HWinfo64 says 5036Mhz average effective clock. So I'd imagine it's throttling.

I used the Noctua supplied thermal paste, in the right size (3-4mm) round dot, in the center of the 7950x. The cooler fits snug, appears to be seated correctly. I originally thought I applied too much paste. Removed it, cleaned both surfaces, re applied being as careful as I could, but still seeing issues.

Idle steady state temps are about 45C.

While I can hear the fans increase in speed under load, I've also tried using the gigabyte control center to set the fans to full speed. Makes no difference.

I'm looking for advice on what I can do to lower those temps.

Thanks,
Keith

 

[COLGeek]

Use Ryzen Master for the most accurate temps. What is reported with it?

 

[punkncat]

What is your ambient temperature?

Are you using any supplemental case fans and what is their configuration?

I don't really feel like your synthetic benchmark temp is particularly worrisome. You could try to drop the side panel and run the benchmarks to see if there is any drop.

 

[boju]

Could try get thermal paste guard and spread the paste yourself rather than let heatsink pressure spread itself, to sure 100% coverage. How was paste spread when you checked it?

Noctua presents NA-TPG1 thermal paste guard for AMD AM5

Designed in Austria, Noctua's premium cooling components are renowned for their superb quietness, exceptional performance and thoroughgoing quality.

noctua.at

Can spread paste yourself with glad wrap on finger. That works well from my experience.

 

[keithth]

@COLGeek Ryzen Master reports the same..... Ball park 93C under load.

@punkncat 20.5C ambient....One 120mm output fan in the rear. One 120mm intake in the mid, front. It's not lined up with the front NH-D15 fan.

Yeah I'm not sure necessarily that there's a problem either. I do want to make sure I can get best, within reason, performance, and that's everything is ideal before I consider this done. And before I start relying on it regularly....

I should have mentioned, this is with the side off the case.

@boju While I'm clearly no expert, the paste spread pattern on the processor looked good. It was even, consumed most of the top of the processor, with no excess on the edges, or bleeding over. Thin layer. The bottom of the heat sink, which had most of the paste when I pulled it off, also had good coverage, but I noticed a ring of thicker paste around the edges.

Appreciate the help!

 

[boju]

Might be too much paste still or not enough pressure from heatsink. I screw tension screws all the way and then back it off half turn.

 

[Phaaze88]

I'm currently seeing temps under load using the free software Heavy Load, of around 92C within a few seconds on running the tests.

It's actually designed to do that, specifically when you load all the cores.

View: https://www.youtube.com/watch?v=OGuFq3jm9hM


It's still safe.

 

[keithth]

Thanks @Phaaze88 : this is the best video/presentation that I've seen on the topic.

The part that surprises me the most was that if you step up in AIOs sizes that all you end up with is slightly more performance before it throttles again. You're lucky if you get a couple hundred mhz. Cooling sufficiently so that the 7950x never throttles despite a heavy load seems very hard indeed.

I'm going to replace my front 120mm with (2) 140mm's, use the offset adapter, and use their latest H2 paste. Get it as good as I can, and be done. If I can move the needle a bit, then so be it.

Thanks

 

[ilukey77]

It's actually designed to do that, specifically when you load all the cores.

View: https://www.youtube.com/watch?v=OGuFq3jm9hM


It's still safe.

yeah was going to say 93c is fine if you want to seriously cool the 7950x start at the 420mm aio and then go to custom loop !!
my 7600x would hit 75c all core R23 with the 420mm arctic freezer ii .. even with some diminishing returns thats some crazy cooling for a 7600x !!

 

[keithth]

Thanks for chiming in @ilukey77 .

I installed one 140mm fan directly in front of the cooler stack, used the offset kit from Noctua, and their H2 paste. Maybe it helped a couple degrees, but not enough to "click" into the next faster clock mode. Same performance, slightly lower temperature.

This is definitely a fools errand. Noctua support wanted to see some HWINFO logs when running Prime95, so we'll see what they have to say. It looks like the IHS can't get the heat from the die(s) fast enough, so unless you're delidding, which I sure as heck am not, there's only so much an ihs-mounted cooler is gonna do for you.

I'm throttling to 5150mhz for an all-core load, which is still 89% of maximum. Not that many of my applications are going to driving such high loads anyways....

 

[Phaaze88]

I'm throttling to 5150mhz for an all-core load, which is still 89% of maximum.

As more cores become active, the max core clock scales DOWN, and the fewer that are active, the maximum goes UP. So the 5.15ghz is about the max for all the cores you've activated.
The advertised clock speeds from both Intel and AMD have been for bursty, low load scenarios, that see like 1~2 cores active at most.

 

[keithth]

Thanks everyone for the posts. I've done some more testing and still not happy with the current results. I might be chasing ghosts here, but I won't be happy until I see the results, or lack thereof, personally.

I'm running a Prime95 small FFTs torture test for 30 minutes. You really do need to run the test for that long because performance changes over time..... it gets worse. Maybe due to "heat soak" ?

My current setup is NH-D15 with NT-H2 spread manually, which was a pain in the butt. Don't care for the stickiness of that stuff. I'm using the noctua offset mount. And replaced the front 120mm fan with (2) noctua 140mm's.

Results have average effective clocks starting at about 4800mhz, dropping to 4700 after 7 minutes, to 4500 after 14 minutes, settling around 4400. There's some smaller troughs near 3800mhz for a few brief periods. Overall average 4334mhz.

Cinebench 2024's of 124 (single), 2200 (multi).

I think this sucks. Maybe it's a pipe dream, but I'd be much happier with sustained performance over 5ghz.

I've ordered an Arctic Liquid Freezer II 360. Gonna try that with their paste, but I also ordered Kyronaut just to get another data point if needed.

Thought I'd share journey in case others are out there doing the same.

 

[keithth]

So playing around with the thermal grizzly kyronaut --- it is actually making a noticeable difference from the NT-H2 noctua paste.

The kyronaut is generally keeping temperatures on average about 1.5 degrees C cooler, the system is able to maintain a higher clock frequency during my 30-minute Prime95 small FFTs run.

I can maintain 4842mhz or above 89% of the runtime with grizzly, and was only seeing that 58% of the time with the NT-H2. The next significant bucket is 4154mhz. So that means that 30% of the time, I'm running about 700mhz better. Averaging the full run, I'm seeing about 175mhz improvement. So between 5-14% improvement.....

This doesn't actually translate into higher cinebench 2024 scores, fwiw. But surely I'm heading in the right direction.

I'm super interested to see what difference the 360mm AIO makes.