AMD Ryzen 9 7950X Cooling Requirements & Thermal Throttling

[R0H1T]

You dump watt in your room. (Watt are a measure of energy). You don't dump Temperature (a measurement of how excited atoms are) in your room.

I don't dump anything in my room, it's those dumb rats & yes atoms, molecules, lattices them too!

 

[Bavor]

Maybe I missed it in the article, but how long in duration were the Blender, Cinebench multi core, and MP3 encode tests? Were they 5-10+ minutes or under a minute or two?

One thing I noticed in my own experience was in longer rendering and encoding workloads, the NH-D15 couldn't compete with a 360mm AIO and in some cases lead to earlier Ryzen CPUs thermal throttling in a high airflow case with a 200 watt(or higher) power limit.

 

[RainingTacco]

@Wizzard

All tested on Wraith Spire @ 100%
Compile: jumping between 85 and 94°C
Photoshop: around 75°C, jumping between 60 .. 88°C (this is a scripted workload, with no gaps between actions)

If the trend of CPUs targeting a certain temperature with their boost clocks continues, will TPU change their testing method to accomodate this? I mean temperature will be meaningless, so we will get graphs with boost clocks i presume? Seems a lot of work to redo tests for the batch of the cooler on the market. And then we get a problem of innate instability of certain bins -one can boost 100-200mhz more than the other, whereas with temperature testing this problem didn't exist -as most IHS and cpus were uniform and made to quite tight specs.

 

[Solid State Brain]

[...] And then we get a problem of innate instability of certain bins -one can boost 100-200mhz more than the other, whereas with temperature testing this problem didn't exist -as most IHS and cpus were uniform and made to quite tight specs.

Results under power-limited conditions are already affected by differences in silicon quality between different units of the same CPU, putting aside motherboard settings like corrections or offsets to applied CPU voltages and to the reported Package Power values.

 

[r9]

If the CPU were monolithic and big, your point would have more merit.

These CPUs use tiny chiplets so the heat is concentrated on very small areas, which is why it's so hard to have lower temperatures.

I wonder if it's possible to check ambient temp inside a case before turning the computer on and after performing one of the more "heavy duty" benchmarks: to see the delta between the temps.

EDIT

Preferably, without a GPU so that one could rule that as a source for the increase in temperature: now that these CPUs have an IGP, that's quite doable.

That CPU is connected via 1718 copper pins that are very good at passing heat to the motherboard.

 

[marios15]

Thanks a lot for the photos @W1zzard

So my concern is....if the reported temperature is higher than the surface temps of the die, it could be another case of offset temps.

1.If a delidded version shows 70 surface vs 95 sensor temps on the naked die too, then this was a very dumb decision by AMD.

2.If the naked die and sensor report same temps then the IHS is trash, and doesn't transfer heat fast enough from the dies.

In case 1, it wouldn't affect longevity of the chip at all, not negatively at least.

 

[Count von Schwalbe]

The temperature of the IHS is MUCH lower. I did a quick test with a thermal camera during first round of Ryzen reviews and the VRMs at 70°C were the hottest part in the socket area

How does this compare to an AM4 w/ 5950X? I do not remember seeing that on those reviews.

 

[TheoneandonlyMrK]

Thanks a lot for the photos @W1zzard

So my concern is....if the reported temperature is higher than the surface temps of the die, it could be another case of offset temps.

1.If a delidded version shows 70 surface vs 95 sensor temps on the naked die too, then this was a very dumb decision by AMD.

2.If the naked die and sensor report same temps then the IHS is trash, and doesn't transfer heat fast enough from the dies.

In case 1, it wouldn't affect longevity of the chip at all, not negatively at least.

Your misunderstanding some stuff, the core temperature is exactly that, in The core or near it, surface temperatures are irrelevant and are not reported though some motherboard have a thermocouple under the socket, with LGA I don't think they're is room.

And the surface of the chips and or heatsink are allllllllllllllllways lower than the core that's what thermal radiation does, spreads the heat out so by the time it hits the heatsink it's across a bigger area ala, exactly what the heatsink is for.

The heatsink is fine and anyone thinking otherwise is deluded or confused IMHO.

 

[rbgc]

As a personal opinion it would be difficult for me to accept 95 C all day.

It is not needed anymore. You can modify CPU limits using software.

You can use CPU default only when you need max perf from CPU. And when not needed and you want to change it, it is possible. You can set lower PPT limit (set "ECO" mode to any number you want, for example 45W), set max CPU freqency (for example 5 GHz), set max CPU temp (for example 60 C ). Or keep CPU limits at default and modify PBO curve with negative offset. All in software without reboot and in seconds. See or try PBO2 Tuner.

So only one decision is needed to make when you plan to buy modern AMD CPU. How many cores I need max (6, 8, 12, 16). CPU limits are configurable using software as needed.

 

[Max(IT)]

The biggest problem is probably psychological. For years we have been trained that "95°C is bad". This is no longer true. 95°C is the new 65°C.


This is utterly non sense. If you write something like that, you are completely ignoring physics law (Black's Equation... anyone ?).
AMD isn't using any "magic" in Ryzen 7000. They are not using any new innovative material. They are not even using a new design.
Everyone here is listening to AMD marketing BS, so 95° is the new 65° ?
Wrong.
AMD just care for the first 24 months of your CPU's life (warranty period in EU). If the silicon significantly degrades after that, is not their business anymore.
You can easily tune a Ryzen 7000 to stay at much lower temperature within 90% of the original performance, but Mrs. Lisa Su decided that a longer benchmark bar on the presentation would be better.

And the usual office defense by AMD cheerleaders here is more than expected.

Are Ryzen 7000 good CPUs ? Yes they really are. But a different IHS and a different default power management would have been the right choice. Even at the cost of a new cooler requirement (no AM4 cooler compatibility).

 

[MKRonin]

What was the fan speed set to on the AIO? 100%?

 

[docnorth]

Do you people render all day? otherwise this stuff does not hit even close to 95c for any significant period

It is obviously (I hope) a way of speaking, but even with a high-end air cooler at 60% speed the CPU stays above 80 C during gaming, based on TPU's test.

Btw I'll copy your (generally correct) comment and post it again when the 13900K reviews come out, if needed.

 

[r9]

What @TheoneandonlyMrK said.

The temperature of the IHS is MUCH lower. I did a quick test with a thermal camera during first round of Ryzen reviews and the VRMs at 70°C were the hottest part in the socket area

That sounds like the IHS is not doing the best job of heat transfer.
I wonder what what kind of improvement delidding would do.

 

[TheoneandonlyMrK]

That sounds like the IHS is not doing the best job of heat transfer.
I wonder what what kind of improvement delidding would do.

The circle of shite.

Read the whole thread, because this was already debated.

Wonder not, but do try Google FFS.

 

[GreiverBlade]

Ok nice, well for me it would be a 7700X if I decided to go AM5+DDR5
Or 5700X if I want to go cheap (could stretch for a 5800X3D )

 

[Max(IT)]

That sounds like the IHS is not doing the best job of heat transfer.
I wonder what what kind of improvement delidding would do.

we already know: something like 20°, that proves how bad designed it was.
But you know, saying they support AM4 coolers is good for marketing.

 

[RainingTacco]

The biggest problem is probably psychological. For years we have been trained that "95°C is bad". This is no longer true. 95°C is the new 65°C.


This is utterly non sense. If you write something like that, you are completely ignoring physics law (Black's Equation... anyone ?).
AMD isn't using any "magic" in Ryzen 7000. They are not using any new innovative material. They are not even using a new design.
Everyone here is listening to AMD marketing BS, so 95° is the new 65° ?
Wrong.
AMD just care for the first 24 months of your CPU's life (warranty period in EU). If the silicon significantly degrades after that, is not their business anymore.
You can easily tune a Ryzen 7000 to stay at much lower temperature within 90% of the original performance, but Mrs. Lisa Su decided that a longer benchmark bar on the presentation would be better.

And the usual office defense by AMD cheerleaders here is more than expected.

Are Ryzen 7000 good CPUs ? Yes they really are. But a different IHS and a different default power management would have been the right choice. Even at the cost of a new cooler requirement (no AM4 cooler compatibility).

Do you have any silicon degradation graphs for MTTF or percentage of failure vs temp/time graph?

 

[r9]

we already know: something like 20°, that proves how bad designed it was.
But you know, saying they support AM4 coolers is good for marketing.

20 ouch that heat spreader more like a heat trap

 

[THU31]

See? These CPUs are not power hungry. All you have to do is use a cheap air cooler.

 

[_Flare]

Very nice review, BIG THANKS
These CPU can regulate surprisingly smart, good work AMD.

 

[TheoneandonlyMrK]

Do you have any silicon degradation graphs for MTTF or percentage of failure vs temp/time graph?

For 5Nm@ Tsmc.

I'm sure as shit they put more scientific hours into knowing the answer then Max(IT)(my asss) did

 

[LuxZg]

A little preview for TLDR (TLDW?):
It looks like Eco mode basically lowers your TDP class from 105 W to 65 W, your temperature by a lot, and your performance by 5-10%.
View attachment 264135

Now I'm sure that if I end up buying a Zen 4 CPU at some point (the B650 motherboard lineup looks pretty grim at the moment, but we'll see), I'll definitely run it in Eco mode.

I'm looking at 7600/7700 mostly, and I would "kill" for that hardware unboxed video adding a combined -10 offset + 65W ECO + Wraith Spire.

My reasoning is that offset would push less watts while keeping performance same, so it should have something in between of two options, but probably closer to offset numbers. And that would be true 65W CPU that should've been released, with boxed cooler.

I really don't care about Celsius, but Watt, those I want at up to 75W.

Looking forward to more articles and videos from everyone. I have plenty time till B650E drops in price a bit.

 

[RandallFlagg]

Sounds like AMD implemented Intel's TVB (Thermal Velocity Boost).

I'm not a fan of that, I prefer setting my max boost by core count, max short power draw, and set average power limits over time after testing with my cooling setup. When I do that, and I hit my thermal limit, that is a separate thing that throttles CPU and if it happens it means I screwed up.

Old school I know but I much prefer it.

With the way the AMD thermal based turbo works, I can see why all the AM5 motherboards have like 16+ VRMs. Keeps someone from frying their VRMs by putting a really good cooling solution into a cheap X670 and letting it pull 200W for an hour straight while running a video game.

I don't suppose Zen 4/AM5 has anything like PL1/PL2 and tau limits?

 

[80-watt Hamster]

You dump watt in your room. (Watt are a measure of energy). You don't dump Temperature (a measurement of how excited atoms are) in your room.

Energy is Joules; Watts are power, or energy over time (Joules/second). EDIT: Granted, Watts are generally a more useful measurement, but given the amount of misconception around this particular topic, it's probably best to be precise.

 

[AusWolf]

Sounds like AMD implemented Intel's TVB (Thermal Velocity Boost).

I'm not a fan of that, I prefer setting my max boost by core count, max short power draw, and set average power limits over time after testing with my cooling setup. When I do that, and I hit my thermal limit, that is a separate thing that throttles CPU and if it happens it means I screwed up.

Old school I know but I much prefer it.

With the way the AMD thermal based turbo works, I can see why all the AM5 motherboards have like 16+ VRMs. Keeps someone from frying their VRMs by putting a really good cooling solution into a cheap X670 and letting it pull 200W for an hour straight while running a video game.

I don't suppose Zen 4/AM5 has anything like PL1/PL2 and tau limits?

No. AMD boosts up until you reach any limit (power, thermal or max boost).

Setting your own max boost achieves nothing on modern AMD CPUs. You're basically restricting yourself to one performance limiter when you could let the chip run faster to hit another limit that it's designed to hit. You could set it up to boost to X MHz with 1 core, or Y MHz all-core, but why would you if it can boost higher in lighter workloads by default?