AMD Ryzen 9 7950X Cooling Requirements & Thermal Throttling

[Garrus]

Heat output is not related to temperature. No his comment was not sarcasm. Ignorance here is astounding, even after it has been carefully explained and an entire article written about it. Your 90 degree CPU in your iPhone is not generating much heat, 3W usually.

Sure the exact same CPU will produce more heat overall with a higher performance and voltage target, but again, nothing to do with temperature. Think about it this way, if you run your fan at 1000rpm and hit 95 degrees, and you run it at 2000rpm and hit 65 degrees, does it affect the heat generated or the performance? No. Same wattage, same performance without throttling. Same for Intel. Nothing new here.

The only thing that matters is "does THIS CPU throttle performance significantly when it reaches high temperatures" and the answer is conclusively, no.

It seems a lot of people actually think lower temperatures are higher performance or less heat. All those people with $300 coolers and they get the same performance and the same amount of heat comes out of their cases, but they don't know that. This has always been the case. Nothing has changed. The only point was to avoid throttling.

 

[Outback Bronze]

So 95°C is the new 65°C.

Does this mean that not too far in the distant future that 125°C will be the new 95°C?

I remember when AMD's R9 290X would run @ 95°C with no problems and everybody said it running too hot. This was almost 10 years ago.

 

[DeathtoGnomes]

@W1zzard Thanks for the this article.

Having a Artic LF2 myself, I can say the hardest part will be installing it, IF, it fits in your case.

 

[Wirko]

Once you enable overclocking mode, the 95°C temperature target gets disabled and the CPU may run at up to 105°C, and only above that it will turn off automatically to protect itself

Ugh, that doesn't sound good. So the ability to throttle has been taken away completely?

 

[Crackong]

Nice article

 

[MrDweezil]

It will be hard to set up a temperature-regulated fan if the temperature tends to be near the maximum even with a 1T load (frequency dropping by 7% when Wraith is at 20% is an indication of that).

This is a good point. If the cpu is regulating its clocks to stay at 95, how should one program a sensible fan curve to ramp up cooling based on need?

 

[Wirko]

Heat output is not related to temperature. No his comment was not sarcasm. Ignorance here is astounding, even after it has been carefully explained and an entire article written about it. Your 90 degree CPU in your iPhone is not generating much heat, 3W usually.

Sure the exact same CPU will produce more heat overall with a higher performance and voltage target, but again, nothing to do with temperature. Think about it this way, if you run your fan at 1000rpm and hit 95 degrees, and you run it at 2000rpm and hit 65 degrees, does it affect the heat generated or the performance? No. Same wattage, same performance without throttling. Same for Intel. Nothing new here.

The only thing that matters is "does THIS CPU throttle performance significantly when it reaches high temperatures" and the answer is conclusively, no.

It seems a lot of people actually think lower temperatures are higher performance or less heat. LMAO. All those people with $300 coolers and they get the same performance and the same amount of heat comes out of their cases, but they don't know that. This has always been the case. Nothing has changed. The only point was to avoid throttling.

Given constant convection cooling, the way it was done in this test, the temperature difference (silicon minus ambient) is in linear proportion to the heat output.

What's absent in Zen 4 is hard throttling, which older CPUs use to protect themselves from thermal destruction. When the CPU reached that state, the performance didn't matter any more. Instead, we're getting soft, intelligent throttling now, and it can as well be called power regulation, which is something that the CPU (and its user) can bear in the long term.

 

[HTC]

@W1zzard : any chance you can add your usual power usage and efficiency numbers' charts to this review?

 

[Naito]

I've got a 240mm AIO on my 5800X3D at it usually sits at 80c while gaming. The 7950X appears to be within the same ballpark, so doesn't concern me too much

 

[cvaldes]

Awesome testing, thank you. As you said, the main issue is psychological, people for years have associated 95C with badness, and many people incorrectly equate temperature with heat flow (95C temps will heat up my house/room/case, etc).

I think a lot of those people have limited experiences.

My Mac mini 2018 (Intel Core i7) will peak at 100 °C during full CPU loads like a Handbrake encode. It's designed to run that way. On such a workload it will peak at 4.2? GHz at the start, then settle down to a lower frequency to maintain a temperature between 95 °C and 100 °C.

I have a Windows notebook PC that manages the CPU similarly. Peak frequency will be attained quickly and then tail off as frequencies are reduced to stay below a given limit. But I owned Mac notebooks for 18 years and they performed similarly whether it was a PowerPC or Intel CPU inside.

As for the heat output topic I can only shrug. Some people here still can't wrap their heads around the topic. Maybe they should head into the kitchen for 10-20 years and pay attention. There they might be able learn some practical concepts about heat output if they are paying attention and observing.

 

[Wirko]

This is a good point. If the cpu is regulating its clocks to stay at 95, how should one program a sensible fan curve to ramp up cooling based on need?

There are other consequences I can think of. Everything that used to have an effect on temperature of silicon will have an effect on performance instead: room temperature, case cooling, application and quality of thermal paste, variations in packaging quality (TIM thickness? IHS flatness?), and so on.

Also, are the (many) temperature sensors on the chip accurate? In many CPUs and other chips that contain sensors, they aren't.

 

[TheoneandonlyMrK]

This is a good point. If the cpu is regulating its clocks to stay at 95, how should one program a sensible fan curve to ramp up cooling based on need?

However you like.

As stated a spire at 20% is fine.

Silent too, though performance suffers but your better off with something decent.

Great article W1zzard.

@Wirko temperature sensors are not accurate at LOW values but fine at high.

 

[AusWolf]

Very nice test, thanks a lot for doing it.

Hopefully, this will dispel some doubts of air coolers being enough for Zen 4.

Old CPUs ran at a constant clock speed and your temperature varied by your cooler choice. You could overclock them manually if you had headroom.

Zen 4 runs at a constant temperature and your clock speed varies by your cooler choice without the need for any manual OC.

Edit: one more way to check against allegations that Zen 4's ISH is too thick would be to check for coolant liquid / exhaust air temperature. If it's cold, it means the IHS isn't transfering heat efficiently enough. If it's warm, it means the IHS is doing its job just fine.

 

[Garrus]

Given constant convection cooling, the way it was done in this test, the temperature difference (silicon minus ambient) is in linear proportion to the heat output.

What's absent in Zen 4 is hard throttling, which older CPUs use to protect themselves from thermal destruction. When the CPU reached that state, the performance didn't matter any more. Instead, we're getting soft, intelligent throttling now, and it can as well be called power regulation, which is something that the CPU (and its user) can bear in the long term.

Wirko: "the temperature difference (silicon minus ambient) is in linear proportion to the heat output"

You know this to be false, it doesn't even make any sense if you think about it clearly. Does the heat output of a heater in your room change if you blow a fan on it? NO.

Basic high school physics here: One of the major effects of heat transfer is temperature change: heating increases the temperature while cooling decreases it. We assume that there is no phase change and that no work is done on or by the system.

The CPU is doing work in the system as electricity is converted to heat. The CPU is not a hot piece of metal that just radiates in the environment to its surroundings. It is consuming electricity.

A CPU consuming 10W is adding 10W of heat to the system, even if it is running at -100 degrees Celsius.

 

[Solid State Brain]

Even with weaker cooling, the processor won't bug out with thermal-trips or emergency system shutdowns, provided you have some form of cooling. It will still hold on to the 95°C temperature at the highest possible clock speed. This in and of itself isn't new. When any chip hits its temperature limit, it either shuts down, or dramatically drops voltages and clock speeds to attempt to lower temperature; only that's not what is happening with Zen 4. The 95°C temperature doesn't necessarily cause the processor to freak out in any way. Instead 95°C is what the processor is trying to run at, by picking the highest possible boost frequency that doesn't cause the temperature to go over the 95-degree mark. Remember, this isn't a classic TJMax, but a load-temperature limit at which the processor begins to lower its highest boost frequency based on the cooling performance. Once you enable overclocking mode, the 95°C temperature target gets disabled and the CPU may run at up to 105°C, and only above that it will turn off automatically to protect itself—this is the real TJMax.

Have you tried running modern Intel CPUs at the thermal limit under stressful multithread workloads?

Just for the sake of safety, on Z-series motherboards you could decrease the temperature limit to a value similar to that of these AMD CPUs and do such test, for example by disabling the cooler fan. As the CPU keeps working at the themal limit you will see that there's no thermal trip, nor emergency shutdown, nor the CPU will freak out. Effective frequency and power will instead gradually decrease, and performance will follow accordingly.

I just tried with my 12700K and Cinebench R23. I started a long run at 4700 MHz at 170W and ended up at about 3800-4000 MHz with the passively-cooled CPU at about 105W (turns out the Noctua NH-D15S can passively cool quite a lot of power). When I quickly repeated a briefer CB23 test with the CPU at about that power due to the thermal throttling, the score seemed consistent with previous tests with the CPU power-limited that way but proper cooling.

It could be argued that under such unlikely conditions the CPU will not work efficiently, but the excerpt I quoted from today's article seems outdated at the least and worth revisiting to make sure that it still holds in 2022.

 

[Wirko]

Maybe they should head into the kitchen for 10-20 years and pay attention. There's they will learn some practical concepts about heat output.

Good idea. When I get back, I will be:
- acquainted with the fact that liquid water in an ordinary pot doesn't want to go above 100°C, no matter how much heat you put in, which proves that temperature is not related to heat
- sixty or seventy, so maybe a bit slower at thinking
- more willing to buy prebuilt PCs than I am now
- stuck in the past, marvelling at new PCs, all connected to three-phase power
- reading about the world adopting some new temperature scale, "so chips look less hot"

 

[robal]

Heat output is not related to temperature. No his comment was not sarcasm. Ignorance here is astounding, even after it has been carefully explained and an entire article written about it. Your 90 degree CPU in your iPhone is not generating much heat, 3W usually.

At least someone gets it !
Temperature (state) and Heat (energy output) are completely different things.
CPU that operates OK in higher temperatures is just better, because it's actually way easier to cool.

 

[wheresmycar]

Great review and i couldn't agree more:

"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"

I'm still in the same boat... gonna have to ride those waves and get used to these new highs. Actually ignore me, i have no use for these high-core-count Zen 4 chips other than a 7600X/7700X which sees "excellent" thermal performance in gaming (esp. considering the new highs)

W1zzard, i'm curious - is there a dedicated page where you announce your up-n-coming reviews?

 

[Garrus]

Good idea. When I get back, I will be:
- acquainted with the fact that liquid water in an ordinary pot doesn't want to go above 100°C, no matter how much heat you put in, which proves that temperature is not related to heat
- sixty or seventy, so maybe a bit slower at thinking
- more willing to buy prebuilt PCs than I am now
- stuck in the past, marvelling at new PCs, all connected to three-phase power
- reading about the world adopting some new temperature scale, "so chips look less hot"

Read the fine print when you study the laws of physics. I'm not trying to be insulting, but educative. But people explain the truth over and over and other people don't listen.

We assume that there is no phase change and that no work is done on or by the system. Otherwise your law is not true.

At least someone gets it !
Temperature (state) and Heat (energy output) are completely different things.
CPU that operates OK in higher temperatures is just better, because it's actually way easier to cool.

exactly!

 

[cvaldes]

Good idea. When I get back, I will be:
- acquainted with the fact that liquid water in an ordinary pot doesn't want to go above 100°C, no matter how much heat you put in, which proves that temperature is not related to heat
- sixty or seventy, so maybe a bit slower at thinking
- more willing to buy prebuilt PCs than I am now
- stuck in the past, marvelling at new PCs, all connected to three-phase power
- reading about the world adopting some new temperature scale, "so chips look less hot"

Another thing people might learn in the kitchen is the notion of thermal capacity and energy.

Boil water (100 °C / 212 °F ) and stick your hand in it. You will probably remove it in a tenth of a second and immediately seek a cold water bath for your hand.

Preheat oven (100 °C / 212 °F ) and stick your hand in it. You can probably leave it in there for a while. Not a big deal.

Boiling water and steam are the same temperature (100 °C / 212 °F ). Which one is a nastier burn?

Heat oil to (175 °C / 350 °F ). Heat oven to same temperature. Throw in some chicken legs. Will they be done in the same amount of time?

A lot of people here simply don't understand basic concepts of heat because they are poor observers. Unfortunately at this point, the opportunity to teach them how to be good observers is probably over for most of them.

As we have seen, an entire article about heat was written and a lot of people still don't get it. It's not like a Q&A forum discussion is going to fix deficiencies in observational skills.

 

[Count von Schwalbe]

@W1zzard any plans to test with a 7600X? This testing is amazing but as you mentioned yourself, is a less likely scenario than someone slapping a low-end cooler on a low or mid rang CPU.

 

[Tyl3n0L]

Very good article. Thanks

 

[Wirko]

Effective frequency and power will instead gradually decrease, and performance will follow accordingly.

Is this the effect of one (or more) of Intel's "boosts"? They are described here, but Intel didn't care to update the page for Alder Lake.

 

[Solid State Brain]

Is this the effect of one (or more) of Intel's "boosts"? They are described here, but Intel didn't care to update the page for Alder Lake.

It should be a different mechanism which Intel datasheets call Adaptive Thermal Monitor. CPU voltage and frequencies will decrease to keep temperature below the configured limit, and in some cases clock modulation can be introduced too. It works seamlessly and progressively, and it's a behavior many Intel-based notebooks already take advantage of while boosting above PL1.

 

[Minus Infinity]

Fantastic article, but until I see how Raptor Lake 13900K behaves as regards clocks, heat, throttling etc, I'll reserve judgement on Zen 4. I'm gobsmacked a massive AIO water cooler didn't make any difference to temperature in blender. Talk is the IHS is just way too thick to allow efficient transfer of heat. I'm bemused as to why AMD chose that path. I hope when the v-cache models come they go with a thinner IHS and/or a Zen 4 revision with thinner IHS. I still won't accept 95C as being normal long term. I tend to keep cpu's a long time.