AMD Ryzen 9 7950X Cooling Requirements & Thermal Throttling

[_Flare]

I liked it even before reading ... but i'm disappointed there are no efficiency numbers, which is what i most wanted to see.

just scroll down to the efficiency charts
https://www.techpowerup.com/review/amd-ryzen-9-7950x/24.html

 

[HTC]

just scroll down to the efficiency charts
https://www.techpowerup.com/review/amd-ryzen-9-7950x/24.html

Not the ones from THAT review: the ones with each cooler and fan speeds from THIS review.

 

[_Flare]

The Eco Mode at 65W will also lower the EDC Current-Limit at the MB VRMs to 90 Ampere.
That test from the "Hardware Unboxed" video with a 7600X wich "only" boosts upto 5450 MHz seems to reach 73A EDC.
Maybe the higher boosting and/or dual-CCD CPUs could hit the Eco-Mode EDC-Limit even in single or lightly threaded workloads.

would be great if that could be tested @W1zzard

 

[HTC]

The Eco Mode at 65W will also lower the EDC Current-Limit at the MB VRMs to 90 Ampere.
That test from the "Hardware Unboxed" video with a 7600X wich "only" boosts upto 5450 MHz seems to reach 73A EDC.
Maybe the higher boosting and/or dual-CCD CPUs could hit the Eco-Mode EDC-Limit even in single or lightly threaded workloads.

would be great if that could be tested @W1zzard

Apparently, he's busy with some review(s) that will be posted today or tomorrow: after that, he'll have the time to do the efficiency charts.

 

[_Flare]

Likely i´m wrong in my concerns regarding the Eco-Modes EDC limit. W1zzards test showed only 47W ST and i doubt that would be hindered by the Eco-Mode in any kind.

 

[HTC]

I said it back in page 3 of this topic:

Supposedly, because it doesn't boost as much, it should also consume less. However, since it takes longer to complete the same benches / tasks (not games, likely), it may actually consume more.

Hence the difficulty and why i requested it.

In games, it's expected to perform worse depending on how many MHz it loses VS the better cooler, but in tasks / benches where it must complete them to provide a result, while it will consume less in the same time period, since it takes longer to complete, it may end up consuming more overall.

Which is why the efficiency charts are required.

 

[RandallFlagg]

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?

*shrug*

I wasn't really talking about setting my own max boost, I was talking about setting power limits over time.

But you also said in the sentence before that there is a max boost on AMD, as well as power, and thermal limits. All 3 exist on Intel as well, all 3 are adjustable, along with Tau a power over time limit. If you want, you can sit there bouncing off a thermal limit.

I guess we will see how long these chips and motherboards last at 95C. AMD says they are designed for it, so there is that.

 

[AusWolf]

*shrug*

I wasn't really talking about setting my own max boost, I was talking about setting power limits over time.

But you also said in the sentence before that there is a max boost on AMD, as well as power, and thermal limits. All 3 exist on Intel as well, all 3 are adjustable, along with Tau a power over time limit. If you want, you can sit there bouncing off a thermal limit.

I guess we will see how long these chips and motherboards last at 95C. AMD says they are designed for it, so there is that.

If they say 95 °C is fine, then I guess the chip could handle more, but this is its maximum guaranteed safe temperature. Time will tell if I'm wrong.

Like someone said it before: the current issue is a mental one. We've seen CPUs pegged at their maximum turbo clocks, we've seen them bounce off their power limits, and we're used to the thought that only overclocked chips ever go near their thermal limits. Zen 4 is basically overclocked right out of the box, with far higher clock and power limits than thermal headroom. It's not wrong, it's not right, it's just something new.

 

[Mussels]

Lower nanometer products run at higher temperatures since cooling is harder without the surface area.

95C won't be where this stops.

Whilst I have no real issue with the dynamic clocking at 95c, I think AMD missed an opportunity to allow users to choose between an optimal profile, and the 95c profile.

It's part of PBO settings and has been since AM4 launced, cTDP i think it's called

You can literally type a max temperature and change that default - AMD's news was "it's safe, don't worry" and should have included "you can lower the value if you want"

Likely i´m wrong in my concerns regarding the Eco-Modes EDC limit. W1zzards test showed only 47W ST and i doubt that would be hindered by the Eco-Mode in any kind.

Eco should only limit MT, not ST

I'm sure there's exceptions somewhere, but i've always been keen on recommending eco mode to people who just do gaming, and didnt have the time or knowledge for screwing with curves

I said it back in page 3 of this topic:


In games, it's expected to perform worse depending on how many MHz it loses VS the better cooler, but in tasks / benches where it must complete them to provide a result, while it will consume less in the same time period, since it takes longer to complete, it may end up consuming more overall.

Which is why the efficiency charts are required.

Agreed - except the lower the clocks, the more efficient they are.
If you use a PPT limit or a thermal limit to lock the CPU to 4GHz instead of 6GHz, it's going to use the voltage for that clock speed and be more efficient, even if it's slower


AM5 feels like intel with all power limits disabled.
Hopefully as more info comes out and more testing is done, people find some common settings to use for PBO that control the wattage and heat, while keeping performance the same - we did it with AM4, and over time AGESA updates improved the stock settings too



AM5 also may come out like Intels 12th gen did, with updates to windows and software drastically altering performance over time - we've already seen that with AVX512 support, with hurried patches giving large changes.

This might be the kind of review w1zzard should re-do in the future, say in 6 months

For all we know, a windows update or chipset driver could alter how the cores are used and boost gaming performance on the dual CCX chips tomorrow

 

[DemonicRyzen666]

I how like everyone has glossed over the fact that Intel has had Cpu's that hit 95C for what like 15 years?

 

[Bwaze]

Lower nanometer products run at higher temperatures since cooling is harder without the surface area.

95C won't be where this stops.

But it can't go on much higher, due to material properties. Very high temperature is also one of the reasons Zen 4 isn't as efficient as previous Zen iterations, due to higher resistances at higher temperatures.

Also, one question - if for instance Zen 4 chips do start degrading, when do the alarm bells sound?

Specs on AMD site have:

Max. Boost Clock Up to 5.7GHz
Base Clock 4.5 GHz

How low must the "up to", non guatanteed boost clock fall for chip to be considered degraded? Below base?

 

[Mussels]

But it can't go on much higher, due to material properties. Very high temperature is also one of the reasons Zen 4 isn't as efficient as previous Zen iterations, due to higher resistances at higher temperatures.

Also, one question - if for instance Zen 4 chips do start degrading, when do the alarm bells sound?

Specs on AMD site have:

Max. Boost Clock Up to 5.7GHz
Base Clock 4.5 GHz

How low must the "up to", non guatanteed boost clock fall for chip to be considered degraded? Below base?

Degradation wont alter the boost speeds, it'd go unstable.

As someone who's 3700x did degrade over time from overclocking (although not by much) - even after years of all core overclocks it's like a 0.2v difference in voltage required at the high end, which still runs it's stock and PBO clocks problem free

 

[HTC]

Agreed - except the lower the clocks, the more efficient they are.
If you use a PPT limit or a thermal limit to lock the CPU to 4GHz instead of 6GHz, it's going to use the voltage for that clock speed and be more efficient, even if it's slower

Imagine this scenario:

- with the best cooler @ it's highest fan speed, the CPU takes 6 minutes to complete task A @ 5.7 GHz
- with the worst cooler @ it's lowest fan speed, the CPU takes 7 minutes to complete task A @ 5.0 GHz

If in the extra time it takes for the CPU with the worst fan speed to complete task A, the CPU ends up consuming more power than the difference in power they both had @ 6 minutes, then it ends up consuming more total power for that task, which means less efficiency. If not, then it's more efficient.

 

[AusWolf]

Imagine this scenario:

- with the best cooler @ it's highest fan speed, the CPU takes 6 minutes to complete task A @ 5.7 GHz
- with the worst cooler @ it's lowest fan speed, the CPU takes 7 minutes to complete task A @ 5.0 GHz

If in the extra time it takes for the CPU with the worst fan speed to complete task A, the CPU ends up consuming more power than the difference in power they both had @ 6 minutes, then it ends up consuming more total power for that task, which means less efficiency. If not, then it's more efficient.

A good theory, but didn't the article show that there is a minimal difference in performance between the "normal" and "low power" modes?

 

[HTC]

A good theory, but didn't the article show that there is a minimal difference in performance between the "normal" and "low power" modes?

According to this review's frequency scaling:

With the Wraith @ 20% fan speed, it lost more than 2 GHz VS the Arctic AIO, using 32 threads, so not so minimal.

Ofc this won't be the case for all tasks and there will be some with minimal speed loss where IT WILL be more efficient, but also others where ...? And this is exactly what i want to know.

 

[Bwaze]

Degradation wont alter the boost speeds, it'd go unstable.

As someone who's 3700x did degrade over time from overclocking (although not by much) - even after years of all core overclocks it's like a 0.2v difference in voltage required at the high end, which still runs it's stock and PBO clocks problem free

You do not know that. Zen 2 and 3 boost was much less aggressive than Zen 4 is. Even Zen 3 was much more aggresive than Zen 2, and it was basically impossible to lift the boost clock by any meaningful degree, and a lot of ways to lower it.

I'm just asking since many people have defended AMD with boost clock debacle. When Der8auer's poll has shown most users didn't achieve the rated boost clock, defenders were quick to point out that "up to" wording means there is no guarantee that the processor needs to reach the boost clock - only the (very low) base clock is officially guaranteed spec.

 

[AusWolf]

According to this review's frequency scaling:

With the Wraith @ 20% fan speed, it lost more than 2 GHz VS the Arctic AIO, using 32 threads, so not so minimal.

Ofc this won't be the case for all tasks and there will be some with minimal speed loss where IT WILL be more efficient, but also others where ...? And this is exactly what i want to know.

Yeah, but with the Noctua at 20%, it had 93% of its performance with 60 W lower power draw. That's 7% performance lost with 35% less power. I imagine the Noctua was dead silent at 20%.

I also recommend this video. It was linked somewhere, but I can't find where.

TLDR (TLDW): Enabling Eco mode to bring down the TDP from 105 W to 65 W (PPT from 142 W to 88 W) on the 7600X resulted in basically no performance loss, but a huge reduction in power and thermals.

 

[Dirt Chip]

I how like everyone has glossed over the fact that Intel has had Cpu's that hit 95C for what like 15 years?

Not that way.

But it can't go on much higher, due to material properties. Very high temperature is also one of the reasons Zen 4 isn't as efficient as previous Zen iterations, due to higher resistances at higher temperatures.

Also, one question - if for instance Zen 4 chips do start degrading, when do the alarm bells sound?

Specs on AMD site have:

Max. Boost Clock Up to 5.7GHz
Base Clock 4.5 GHz

How low must the "up to", non guatanteed boost clock fall for chip to be considered degraded? Below base?

Below 4.5, everything else is non issues according to warranty.

 

[dj-electric]

Lower nanometer products run at higher temperatures since cooling is harder without the surface area.

95C won't be where this stops.

The problem here is that we could have had it much better. AMD has made a decision with one good karma solution while also having a bad karma side effect to it.

The Ryzen 7000 CPUs could have been cooler, a lot cooler. When I say cooler I don't mean consume less power (well, that's also the case if you manually power tune them, but besides the point).

It does not take a degree in mechanical engineering to see that AMD's decision to keep a similar Z height compatibility with AM4 socket and cooler cost them a lot of heat bottleneck in Ryzen 7000 CPUs.

The temperature delta between the CPU die and the IHS surface is monstrous, and exists thanks to the particularly high thickness of the IHS and its general design.

So if you are AMD you have to choose between supporting many (I would say even most) of AM4 compatible coolers, or make your new CPUs being able to run coolers, possibly slightly faster, or even shift your entire frequency to temperature goal 10 degrees lower to stay away from the reputation of those CPUs running particularly hot when stressed. I wholeheartedly believe that we can reach such large delta if AM5 CPU IHS was to be design with thermal flow priority first, and completely disregard AM4 cooler compatibility.

AMD's decision about this is at least a part on why we ended up with this interesting situation.

It is easy to just say "AMD should have decided to change their compatibility and drop AM4 support entirely", but the backlash from that might have been more severe from the one on the whole "designed to run at 95c" one. I for one, wouldn't want all my AM4 cooler equipment to be incompatible with AM5, that would suck, especially for the lower end parts.

 

[pressing on]

I agree that the decision to design the AM5 socket around compatibility with AM4 coolers was a good one from the point of users but less good when it came to the implications such as the thickness of the IHS. A different result could have been achieved if the aim had been modified to compatibility with AM4 coolers with an adapter. But I guess that AMD might have looked at what happened to Intel with the LGA 1700 adapters that were necessary for existing coolers to work with 12th Gen (Alder Lake) CPUs and decided that was not for them. In the event complete compatibility with AM4 coolers was not achieved with AM5 because the ILM is screwed into the backplate, effectively making it non-removable and this is an issue with the (admittedly small) number of coolers that require the backplate to be removed/replaced.

 

[catulitechup]

Socket AM5 must throw away any compatibility with am4 size for prevent this situation, with more bigger socket them can prevent this temperature situation

However said somethink like 95 degrees are perfectly safe and normal that dont sound good, maybe seems a cheap trick to try sell a product with temperature problems

Personally seems something like that: you need buy new expensive cpu, new expensive mainboard, new expensive memories aka higher frecuencies memories but you dont need buy new cpu cooler

If buy new expensive components, buy new cooler dont be a problem because users buy am5 must be have enough money and new cooler dont change so much in money

Again personally i think ryzen 5 7600x dont must be exist (same than size dimensions of socket am5) because give a strange message in prices side

Because with all things showed this platform are high end and dont be avalaible for averaje joe users, for them amd still have am4

 

[Dirt Chip]

The problem here is that we could have had it much better. AMD has made a decision with one good karma solution while also having a bad karma side effect to it.

The Ryzen 7000 CPUs could have been cooler, a lot cooler. When I say cooler I don't mean consume less power (well, that's also the case if you manually power tune them, but besides the point).

It does not take a degree in mechanical engineering to see that AMD's decision to keep a similar Z height compatibility with AM4 socket and cooler cost them a lot of heat bottleneck in Ryzen 7000 CPUs.

The temperature delta between the CPU die and the IHS surface is monstrous, and exists thanks to the particularly high thickness of the IHS and its general design.

So if you are AMD you have to choose between supporting many (I would say even most) of AM4 compatible coolers, or make your new CPUs being able to run coolers, possibly slightly faster, or even shift your entire frequency to temperature goal 10 degrees lower to stay away from the reputation of those CPUs running particularly hot when stressed. I wholeheartedly believe that we can reach such large delta if AM5 CPU IHS was to be design with thermal flow priority first, and completely disregard AM4 cooler compatibility.

AMD's decision about this is at least a part on why we ended up with this interesting situation.

It is easy to just say "AMD should have decided to change their compatibility and drop AM4 support entirely", but the backlash from that might have been more severe from the one on the whole "designed to run at 95c" one. I for one, wouldn't want all my AM4 cooler equipment to be incompatible with AM5, that would suck, especially for the lower end parts.

Or in short- a "design choice" that is basically a design flaw with manageable workaround (aka 95 degree is fine and set your cap your fan speed to 60% to escape the noise).
Time will tell but this decision might put AMD in bigger problem with zen5 and feutre zen on AM5. The density will just get bigger and bigger as lithography lowers.

 

[TheoneandonlyMrK]

Or in short- a "design choice" that is basically a design flaw with manageable workaround (aka 95 degree is fine and set your cap your fan speed to 60% to escape the noise).
Time will tell but this decision might put AMD in bigger problem with zen5 and feutre zen on AM5. The density will just get bigger and bigger as lithography lowers.

The thermal envelope Am5 operates in is directly caused by the node not the heatsink and all you saying otherwise need to show your qualies and test data because your chatting total balls.

And clearly do not understand the design choices made were to optimise the performance in every scenario, automatically.

A lower Z hight on some KS chip's was not the heatsink, the actual chip was ground thinner, which will have meant that the heatsink on those would be thicker than standard to align height with the cooler's.

It isn't the IHS it's the power and frequency curve control circuitry causing the DESIREd high temperatures.

Delidded chips with direct die cooling are always upwards of 20° c cooler because there's two less layers for heat to pass through.

Arm chair engineering= asss brained assuming IMHO.

And I can't wait to hear some of the yeah butting, and whataboutism in the 13900K review thread.

 

[thestryker6]

Delidded chips with direct die cooling are always upwards of 20° c cooler because there's two less layers for heat to pass through.

This is simply not true, and something der8auer addressed in his delidding/direct die video: he saw 18-21c temperature reduction the typical is more in the 10-12c range (he was using prototype liquid metal which he said should be 1-2c with maximum of 3c improvement). The IHS is clearly hurting the efficiency of these chips more than it would have if it had been better designed.

Personally I don't think this is a dealbreaker, especially if you're okay with Intel's 240W+ chips, but it was a poor design choice to hurt their own product for cooler compatibility. Both Intel and AMD have made it easy for me to skip this generation and wait to see what HEDT/MTL has to offer next year.

 

[TheoneandonlyMrK]

This is simply not true, and something der8auer addressed in his delidding/direct die video: he saw 18-21c temperature reduction the typical is more in the 10-12c range (he was using prototype liquid metal which he said should be 1-2c with maximum of 3c improvement). The IHS is clearly hurting the efficiency of these chips more than it would have if it had been better designed.

Personally I don't think this is a dealbreaker, especially if you're okay with Intel's 240W+ chips, but it was a poor design choice to hurt their own product for cooler compatibility. Both Intel and AMD have made it easy for me to skip this generation and wait to see what HEDT/MTL has to offer next year.

I disagree with you still, surprise.

And I have to ask are you going to work on Hedt or just game?!.

With either supplier your future looks bleak I'd say, newer chip's are not going to be cooler.

And when Intel kicks in with chiplets and stacks your ideology is going to get wrecked IMHO.