AMD Ryzen 9 7950X Cooling Requirements & Thermal Throttling

[AusWolf]

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.

The AIO didn't make a difference in temperature, but it did make a difference in clock speed. The CPU auto-overclocks itself to reach 95 °C regardless of your cooler. Whatever cooler you have, it will be used to its full capacity to give you the best performance possible. It's not like CPUs of the old days that ran at fixed clock speeds that you had to tune yourself. Manual overclocking is a thing of the past.

 

[Mussels]

People have been trained by over simplified cooler reviews that "60C is good" "90C is bad"

You throw in intel CPU's that reported really low temperatures for a long time because of the slow reporting times showing averages, and the fact that distance to TJMAX was never intended to be accurate the further it got from the throttle point that gave everyone very wrong ideas of how coolers work


The amount of times i've seen social media stuff "But this review shows 40C why am i at 60C" with entirely different hardware is just... overwhelming.
Oh and the "These are my temps at idle are they good" threads every single day on facebook gaming groups....




And yeah, who buys a 7950x to get low temps? no one. You buy it to either run expensive workloads on like rendering that *need* fast completion times, or you do it because you're a bored rich kid mistaking envy of the masses as the love you never got as a child

It looks like you can tune PBO or use voltage curve offsets and drastically cut the temperatures down too, very similar to the 5800x3D - the default PBO settings have a good 40W or so that does nothing for performance you can shave off, even before undervolting

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.

It's clocking higher on the bigger cooler.
It's clocking higher and using more wattage, to run at the same temperature.

These are designed to boost forever. Just boost boost boost, as far as the cooler can take it.
Obviously, if you limit things like intel with PL2 on a time limit, they'd settle down after a while - AMD has chosen to let the cooler be the limit.

 

[panjang]

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

me too, i would love to see power usage in relations to which cooler used. Does lower frequency means lower power?

 

[Punkenjoy]

Not only modern CPU have way more temperature (and power usage) sensors, they are smaller.

In the past, they had way less, so you had to have a higher margin temperature wise as you could reports 80° C but some parts of the CPU could be above 100°C

Now, CPU are much more smaller, it have way more temperature sensors. The Sensors density is much higher (since things are smaller they are closer to the heat sources). It's quite possible that this only is what allow us to run these things at a higher reported °C all days.

This is how modern boost algorithm work. This is what multiples sensors allow you to do. Get as close to the edge without going too far. The fact that you can get an air cooler and get almost all the performance of the CPU is somehow a proof that heating or throttling isn't an issue.

I would still run an AIO or custom loop but that is personal preference. I think people that prefer air cooling can run these CPU without problem.

 

[Scrizz]

I'd love to see something like this with a Noctua NH-D14 or NH-D15
Thanks for the quick review.

 

[zlobby]

Great article, @W1zzard !

Do we have a chance to see core and package temps vs time and core(s) load?

I get 95C is the TARGET temp, so the CPU will AIM for it under max possible clocks, but I'd like to see how aggresively it does so.

 

[Minus Infinity]

The AIO didn't make a difference in temperature, but it did make a difference in clock speed. The CPU auto-overclocks itself to reach 95 °C regardless of your cooler. Whatever cooler you have, it will be used to its full capacity to give you the best performance possible. It's not like CPUs of the old days that ran at fixed clock speeds that you had to tune yourself. Manual overclocking is a thing of the past.

True, but still no reason I can see for the very thick IHS. Also you won't get 5.75GHz over 50C so that's either a joke as it's impossible and why advertise it or did they hineslty believe you could somehow cool the cpu to achieve that. I'd rather my cpu didn't have throttle down due to high temps.

 

[HTC]

me too, i would love to see power usage in relations to which cooler used. Does lower frequency means lower power?

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.

 

[Count von Schwalbe]

True, but still no reason I can see for the very thick IHS. Also you won't get 5.75GHz over 50C so that's either a joke as it's impossible and why advertise it or did they hineslty believe you could somehow cool the cpu to achieve that. I'd rather my cpu didn't have throttle down due to high temps.

Why not? The only reference point was the M1, as no other desktop CPU used the TSMC N5 node before. AMD also decided that no thermal throttling whatsoever would happen before 95C. Between those, I see no reason not to believe the advertised boost clocks (given half-decent cooling, naturally).

 

[Steevo]

True, but still no reason I can see for the very thick IHS. Also you won't get 5.75GHz over 50C so that's either a joke as it's impossible and why advertise it or did they hineslty believe you could somehow cool the cpu to achieve that. I'd rather my cpu didn't have throttle down due to high temps.

Intel sockets are bendy with thinner IHSs.

https://www.techpowerup.com/forums/threads/thermal-grizzly-announces-contact-frame-for-intel-alder-lake-promises-to-reduce-temperatures-by-up-to-10º.295334/#post-4764631

 

[thestryker6]

Thanks for the straightforward testing.

While I don't like the temps I use liquid cooling so I'm not worried about the ambient increase. The thing which concerns me most is that when Der8auer did direct die the measured power consumption dropped. If the software measurement is accurate to reality that means the IHS thickness is not only causing higher temps than needed, but also increasing power consumption.

 

[Camm]

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.

 

[AusWolf]

True, but still no reason I can see for the very thick IHS. Also you won't get 5.75GHz over 50C so that's either a joke as it's impossible and why advertise it or did they hineslty believe you could somehow cool the cpu to achieve that. I'd rather my cpu didn't have throttle down due to high temps.

The very thick IHS is only to make sure AM4 cooler compatibility. If we didn't have it, people would cry for having to swap coolers. Kind of a lose-lose situation.

5.75 GHz below 50 °C just represents what the chip is actually capable of. It's not something to aim for, purely marketing. Not that I agree with such marketing (I don't), but if we complained about every single marketing BS that AMD/Intel/Nvidia came up with through the years, TPU's servers would be full in no time. I'm not saying that any marketing BS is good - just take the product at face value, without any marketing crap.

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.

That's what Eco mode is for, I guess.

 

[Camm]

That's what Eco mode is for, I guess.

I'd argue thats an efficiency preset rather than an optimal one. But yeah, if anything AMD's communication definitely sucked on this.

 

[AusWolf]

I'd argue thats an efficiency preset rather than an optimal one. But yeah, if anything AMD's communication definitely sucked on this.

I agree that it could use some more testing and media coverage. I'd love to know more about it for sure. Considering how much power and heat you can save on most modern CPUs and GPUs while only losing a few percent of your max performance, I think we can call any kind of Eco setting the optimal one nowadays.

 

[Forza.Milan]

Thanks for the straightforward testing.

While I don't like the temps I use liquid cooling so I'm not worried about the ambient increase. The thing which concerns me most is that when Der8auer did direct die the measured power consumption dropped. If the software measurement is accurate to reality that means the IHS thickness is not only causing higher temps than needed, but also increasing power consumption.

it seem, siliconlottery.com will be open and on business again...

 

[HTC]

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.

That's what Eco mode is for, I guess.

You can change the max temp in the BIOS (different setting than ECO mode), but i don't know the exact setting where it's changed, and it's name also varies from vendor to vendor: i know this from one of AnandTech Forum's RyZen 7000 related topics but i'm having trouble locating it.

EDIT

Haven't found the post but i found this video talking about it (timestamped):

That's the setting for this board: "Platform Thermal Throttle Ctrl".

 

[[XC] Oj101]

This is 7 years ago: https://www.techpowerup.com/cpu-specs/?released=2015&sort=name

That's for aging me seven years with a single post.

Man, that feels like yesterday.

 

[nguyen]

@W1zzard What is the ambient temp for your testing lab?

 

[Bwaze]

How do you set the fan curve with a processor behaviour like that?

Let's say you do it as usual, so the fans are silent at idle, and gradually ramp up as the temperature gets higher.

But the processor reaches close to 95 degrees with almost any load - doesn't matter if it's 40W single core load or 200W+ rendering. Heat generated is of course very different, but this isn't visible just by looking at CPU temperature.

Of course you can set the fan speed at 100% at 95 degrees. And then watch your fans ramp up and down even with processor at idle, just doing the standard Windows background tasks. That was very apparent in Zen 2 and 3, and it's probably just gotten worse.

 

[nguyen]

How do you set the fan curve with a processor behaviour like that?

Let's say you do it as usual, so the fans are silent at idle, and gradually ramp up as the temperature gets higher.

But the processor reaches close to 95 degrees with almost any load - doesn't matter if it's 40W single core load or 200W+ rendering. Heat generated is of course very different, but this isn't visible just by looking at CPU temperature.

Of course you can set the fan speed at 100% at 95 degrees. And then watch your fans ramp up and down even with processor at idle, just doing the standard Windows background tasks. That was very apparent in Zen 2 and 3, and it's probably just gotten worse.

Set your CPU fan speed according to VRM temp, CPU temp fluctuate too fast.

 

[LuxZg]

"Limiting power to conserve energy is another thing of course and perfectly reasonable. Early results indicate that a lot of efficiency can be gained from this approach—more on this soon"

Oh yes! Can't wait for more in this subject Thanks for all the info!

 

[Niceumemu]

Now add direct die cooling and see if there are any gains from that too, would be cool to see

 

[zenlaserman]

Looks like a good job done here, but I'm not convinced that electromigration it's a thing from the past, and
continuous high temperature over the years... doesn't seems right.

Electromigration is usually attributed to overvolting, not necessarily a temperature thing. Look, laptop hardware has been seeing stuff run this hot for decades, and some of the really old high-end GPUs would run 85-90C constantly. Still have an AIW X800XT that works just fine, always hits 85C in games. I stopped caring about temps back in the Prescott days.

Temperature monitoring software gets people paranoid, I think. I gotta laugh at all this hullabaloo. Historically, the newest jump in lithography has always brought interesting temperature characteristics.

 

[Jimmy_]

Nicely written and well explained Kudos to uh!
So, the 95C is the new normal for CPU