Info wet sanded 5950x and nh-d15

[Soulkeeper]

I was getting frustrated with 90C temps on my new build.
After thinking about the heat density of these chips (one of the highest ever), I decided i'd give the IHS and heatsink base a sanding.
Nothing too crazy:
150grit -> 220grit -> 400grit(wet) -> 600grit(wet)
thermal grizzly conductonaut (both before and after).


After this i'm seeing about 3C lower temps, incase anyone is wondering.
These screenshots were taken after compiling firefox and mingw while watching videos.
These are not exhaustive/scientific tests, but I can see ~3C.

Before:



After:

 

[igor_kavinski]

Nice. How about using this: https://www.amazon.com/Innovation-Cooling-Graphite-Thermal-Pad/dp/B07CKVW18G?th=1

Tech Junky on these forums swears by it.

If that doesn't reduce temps, maybe consider watercooling or this: https://www.newegg.com/deepcool-gamer-storm-assassin-iii/p/13C-002K-000H4

 

[Soulkeeper]

I wouldn't use a thermal pad over liquid metal. Unless the pecking order has changed drastically.
Most all the reviews i've seen show all the dual tower aircoolers in the same ballback as my NH-D15.
It seems that I need to just be happy with what I have for awhile (except optimizing the settings).
I'd love to get water cooling, I just can't afford it right now.

 

[thigobr]

Is the base making good contact with the IHS now? I have a Thermalright True Spirit Power 140 that is a really good tower cooler with 6x8mm heat pipes but the base is convex. It makes perfect contact with the IHS center but poor contact with the borders. Unfortunately Ryzen 5000 has the main chiplets placed on the border so this kills the cooling potential.

You could also try to mount the cooler with an offset, moving the base center closer to the main compute chiplets. I didn't try because I ended up moving to custom a WC but it's a next step

 

[Det0x]

For a comparison, with my overkill watercooling loop i manage ~70degrees load-temp while using 260w PPT in Cinebench 23, on my 5950x with LM and a TechN block. (in normal ~22-24 degrees celsius ambient)

https://imgur.com/a/rlI8N18

 

[Soulkeeper]

Is the base making good contact with the IHS now? I have a Thermalright True Spirit Power 140 that is a really good tower cooler with 6x8mm heat pipes but the base is convex. It makes perfect contact with the IHS center but poor contact with the borders. Unfortunately Ryzen 5000 has the main chiplets placed on the border so this kills the cooling potential.

You could also try to mount the cooler with an offset, moving the base center closer to the main compute chiplets. I didn't try because I ended up moving to custom a WC but it's a next step

Yes, I noticed that too when I was sanding. The heatsink base was convex and the IHS was concave. Now they are both about as smooth as I have the time to make them.
I think an offset would definitely help, but unfortunately the NH-D15 (non S) already basically touches the top of my video card. I've had bad luck in the past trying to bend heatpipes, they like to collapse, so i'm reluctant to even attempt it.

 

[Soulkeeper]

I think the lower 1 or 2 heatpipes are tasked with having to remove more of the heat than the rest. The front lower CCD is also strained more by the OS scheduler and gets worked the most compounding the issue.

 

[Soulkeeper]

For a comparison, with my overkill watercooling loop i manage ~70degrees load-temp while using 260w PPT in Cinebench 23, on my 5950x with LM and a TechN block. (in normal ~22-24 degrees celsius ambient)

https://imgur.com/a/rlI8N18

overkill is always good, that's one nice setup there

 

[eek2121]

I was getting frustrated with 90C temps on my new build.
After thinking about the heat density of these chips (one of the highest ever), I decided i'd give the IHS and heatsink base a sanding.
Nothing too crazy:
150grit -> 220grit -> 400grit(wet) -> 600grit(wet)
thermal grizzly conductonaut (both before and after).


After this i'm seeing about 3C lower temps, incase anyone is wondering.
These screenshots were taken after compiling firefox and mingw while watching videos.
These are not exhaustive/scientific tests, but I can see ~3C.

Before:

View attachment 58887

After:
View attachment 58888

Man, if my 5950x hit 90C I would not be happy. The highest I've seen with my AIO is 86C, and that was with Prime95 torturing the daylights out of my system after PBO was enabled.

 

[Soulkeeper]

After spending countless hours tuning(and stability testing) curve offsets, I managed to get my temps down another few C.

 

[igor_kavinski]

After spending countless hours tuning(and stability testing) curve offsets, I managed to get my temps down another few C.

What are the settings and how low are the max temps now?

 

[beginner99]

Isn't the point of ryzen to boost as far and long as there is thermal headroom? Unless you are using extreme watercooling like the MORA3 in this thread or some other exotic solution, won't ryzen just boost higher/longer and hence limited gains are possible in terms of temp?

I'm also not really clear why this is a huge issue. 90 is safe for a chips. Not like it will break.

This assumes no manual overclock is set. else well scale down on the OC.

 

[igor_kavinski]

I'm also not really clear why this is a huge issue. 90 is safe for a chips.

Just because it won't break doesn't mean there is no adverse effect. Prolonged high temperature will reduce the chip's life by a few years. It may also lead to premature failure depending upon the silicon lottery.

 

[eek2121]

Isn't the point of ryzen to boost as far and long as there is thermal headroom? Unless you are using extreme watercooling like the MORA3 in this thread or some other exotic solution, won't ryzen just boost higher/longer and hence limited gains are possible in terms of temp?

I'm also not really clear why this is a huge issue. 90 is safe for a chips. Not like it will break.

This assumes no manual overclock is set. else well scale down on the OC.

Ryzen has a power limit, so the answer is no.

 

[tomatosummit]

Ryzen has a power limit, so the answer is no.

I've forgotten the name of the effect but the electical efficiency of silicon is affected by temperature slightly.
The cooler you run a cpu the more efficient it is and can boost higher within it's power limit.
There were some tests done on zen2 launch that showed the small performance difference a decent cooler (vs stock) has on even stock cpus, only a handful of percent but it was outside of the margin of error.
Zen2 was highlighted because the hotspot concerns and actually having stock coolers shipped with cpus.

 

[beginner99]

Ryzen has a power limit, so the answer is no.

that is exactly my point. As long as it is in the power limit, it will use it. So you are not getting cooler temps but more performance with a better cooler. Up to a limit. But this "my ryzen is running too hot" has been a thing since Zen2 release and it has been explained why that is the case a gazillion times.

if you don^t want this, then you need to turn off the automatic stuff and to a manual OC or in this case Underclock/undervolt.

 

[dr1337]

Prime95 makes my 3600 peak at 88c with a d15, seems like a normal temp to me. Unless you're getting that hot while gaming or something I don't think there is any need to worry. Even in the most CPU intensive games I don't see temps higher than 70c, 80c+ is a peak and something I only see if I'm looking for it.

 

[Soulkeeper]

Yes, the lower the temps the more boost.
I've also noticed a slight improvement to neighboring cores when lowering the volts of others.

 

[Soulkeeper]

What are the settings and how low are the max temps now?

I'm using a -0.08125v Dynamic vcore offset, then optimized the curve from there.
Highest temps are in the 70's now, and 5GHz boost on all cores is pretty consistent.
+0 boost clock overrride. I have not bothered to try to find higher clocks yet.
My first CCD is also worse than my second CCD. The second one can take a lot of undervolt and only has 1 relatively bad core.

It's worth noting that each time I lower the dynamic vcore, temps go up and boost goes down untill I redo the curve.
This is probably about as low as I want to push dynamic vore.
Essentially, by lowering the dynamic, I give the cores less undervolt at lower frequenices. This gives more potential undervolt for the high clocks with the curve offset. For curve offsets it's -0.005*co TO -0.003*co. Dynamic is less agressive on the lowend, since it is static across the range.