90c+ CPUs

[BSim500]

For AMD 7000, reviewers keep saying that AMD has notify them about it and that 95C is the maximum safe 24/7 everyday temperature and not the absolute max operating temperature.

Personally I wouldn't say they're designed / going to fail, but aggressive thermal cycling / electro-migration is still bad for electronics regardless of "100c CPU's are normal" marketing used to wallpaper over the fact Intel vs AMD are engaged in some dumb pre-overclock Pentium-4 style p*ssing content pushing chips way past their sweet spot on the efficiency curve. I can't say I like this 'direction' here at all vs simply leaving overclocking to overclockers. What's the point of K/X branded unlocked chips if there's no headroom for those who want to overclock, whilst those who don't end up with sh*tty thermals they never asked for anyway. I'd definitely second adding a page to the reviews testing performance / thermals at 65w, and test with a 65w rated stock cooler / Noctua L9i. That will "separate the wheat from the chaff" as far as "we made our chip faster (IPC)" vs "it's only faster because we turned it into a furnace..."

 

[Vayra86]

Personally I wouldn't say they're designed / going to fail, but aggressive thermal cycling / electro-migration is still bad for electronics regardless of "100c CPU's are normal" marketing used to wallpaper over the fact Intel vs AMD are engaged in some dumb pre-overclock Pentium-4 style p*ssing content pushing chips way past their sweet spot on the efficiency curve. I can't say I like this 'direction' here at all vs simply leaving overclocking to overclockers. What's the point of K/X branded unlocked chips if there's no headroom for those who want to overclock, whilst those who don't end up with sh*tty thermals they never asked for anyway. I'd definitely second adding a page to the reviews testing performance / thermals at 65w, and test with a 65w rated stock cooler / Noctua L9i. That will "separate the wheat from the chaff" as far as "we made our chip faster (IPC)" vs "it's only faster because we turned it into a furnace..."

There isn't a point in K-chips since Coffee Lake.

Kaby Lake was the last good overclocker and even then ran hot, and ironically also the last quad core top end consumer CPU. Intel didn't progress, they just took our OC headroom. AMD took 4 iterations of Zen to get to the point Intel got to over 9 generations of Core. GPU is similarly pointless since Pascal. The only real CPU progress we got since Skylake is new DDR and elevated core counts, let's face it. The rest is... margin of error nonsense at best. It shows because you can still run most stuff on ancient CPUs. The elevated core counts now go along with better multithreading than we had in the quadcore era, that's where the real movement in perf is at. Big Little is also all about more multithreading while not sacrificing ST. As are these new temp targets. The simple fact is, only IPC advances are not enough to warrant new CPU generations, but money must roll, and we support this, every time.

The new OC is the undervolt. Its a perspective thing mostly, but to me a good OC has always been about an efficiency/perf sweet spot, not just yanking the voltage up for maximum clock.

 

[mb194dc]

Only a tiny amount of use cases really needs 250 to 300w cpu at 90c+. Mainly for servers and some workstation.

E.g from tpu review, at 4k all core 5.1ghz 7950x consumes 55w and 1% difference with max power config.

Almost all users won't see any productivity or FPS difference if they're using a processor from the last few generations and game at a modern resolution like 4K. 1080p was the high end 10 years ago... shouldn't even be in reviews now?

Only a tiny portion of users do so much encoding, rendering or low res gaming it'd make sense to pay the premium for the latest chips.

I can't agree with the review conclusions for those reasons. Almost everyone better off on AM4 for years more.

 

[AlwaysHope]

....

Inflation is present in the states, but if you managed to retain employment through the pandemic, it's not cutting in as much as many would have you believe.

Of course, if you are homeless/jobless, a cpu may not be your primary worry.

Getting a little off topic here, however security of employment is threatened by rising inflation in the private sector. This dynamic never changes throughout economic history.

 

[Valantar]

They're all on the take from Big Tech.

Shit. You're right - they're literally paying them! Like, salaries, even! Check mate, man.



As has been said by several people above here, I'm not worried about these thermals. Laptops and a lot of other hardware operates at these temperatures entirely safely. Yes, laptops fail more often than desktops, but when that is caused by thermals it is mostly VRMs failing, not CPUs. And you obviously still need VRM cooling on your desktop. There also won't be excessive heat dissipated into the board from these chips despite their high thermals - the high thermals are caused by an inability to efficiently dissipate heat into the cooler, after all. The additional heat through the socket will be negligible over what is caused by the increased CPU power.

AMD is saying these chips can run 24/7 at 95°C, and I trust them when saying that. The Zen3 boost and voltage control and chip protection system has already proven itself to be very capable, and this is improved over that. Us enthusiasts need to change our thinking though. Our already rather irrational preference for more or less arbitrarily low temperatures will no longer work - we need to start trusting the chips to govern themselves. They can do so far better than we can, after all. No human can keep track of core voltages thousands of times a second, or temperatures, or clock speeds. That's just reality. PCs are for using, not monitoring.

That increased CPU power worries me more, especially as a long-time SFF builder. Luckily it seems that these chips have tons of potential for power limit tuning while sacrificing minimal performance, especially in ST but also MT. I'd be very interested to see what a 7600X could do in my Densium 4+ with a Noctua L9, but I won't be upgrading that system any time soon.

A more significant issue IMO is that this change necessitates a paradigm shift in how we think of fan control - a linear relation between fan speeds and core temperature is insufficient now that CPUs are doing this degree of fine-grained boost and power control at high temperatures. Controlling fans by CPU power, but somehow also weighted by temperature, would likely be better. This is what laptops do after all, and it allows them to run relatively quietly (for their size).

 

[BSim500]

1080p was the high end 10 years ago... shouldn't even be in reviews now?

80% of the market use 1080p or less (vs 10.9% 1440p, 2.5% 4k & 2.4% Ultrawide). It's as mainstream relevant as you can get and simply an example of "good enough = here to stay" (just like we went from 3-8TB HDD's back to 0.5-2TB SSD's). If the average person doesn't feel they need more, then that is what mainstream becomes.

Only a tiny portion of users do so much encoding, rendering or low res gaming it'd make sense to pay the premium for the latest chips.

Agreed, but even then a lot of consumers over-estimate it. Eg, a lot of 'content creators' have figured out there's zero point buying a 16-core CPU for CPU-based archival quality x264 encoding to upload to Youtube when the first thing Youtube does it recompress it at their end using hardware-encoding anyway, ie, might as well use NVEnc, etc, in the first place at which point even 2 vs 16 cores becomes moot for Youtubers. Same goes for streamers using external HDMI capture / broadcast devices, video captured on phones, all GPU fixed-function encoder based. Not to mention editing software has gotten smarter, ie, add 5mins worth of overlays / fade-outs / on-screen displays to 2hrs of footage, and it will intelligently recompress only that 5mins rather than brute-force the whole 2hrs. It's ironic that the era that produced mega-cored CPU's is the same one that needs them less for consumer video vs batch ripping DVD to Divx overnight all those years ago.

 

[AusWolf]

I was surprised while reading the review that even Arctic Freezer II wasn't able to get it under 90 °C when loaded, though it wasn't mentioned which size of the radiator was used. I wonder how the "next gen" AIOs will have to be designed to get those temperatures lower, even though that's a new safe temperature.

That's because of how the CPU boosts. If you have a better cooler, it'll push itself higher, and you'll still reach 95 °C. Only when you reach the chip's architectural limitations with thermal headroom will you see temps below 90, but you'll probably need liquid nitrogen, or at least a massive custom loop with multiple radiators for that.

If you think about it, Zen 4 is designed to give you everything that your cooling allows without any limitation on performance whatsoever.

OK, guys... I made a little sketch about the way I understand Zen 4's boosting behaviour so that we're all on the same page. I don't want to confuse anyone, so if someone knows better, please correct me.

 

[Valantar]

That's because of how the CPU boosts. If you have a better cooler, it'll push itself higher, and you'll still reach 95 °C. Only when you reach the chip's architectural limitations with thermal headroom will you see temps below 90, but you'll probably need liquid nitrogen, or at least a massive custom loop with multiple radiators for that.

If you think about it, Zen 4 is designed to give you everything that your cooling allows without any limitation on performance whatsoever.

OK, guys... I made a little sketch about the way I understand Zen 4's boosting behaviour so that we're all on the same page. I don't want to confuse anyone, so if someone knows better, please correct me.

View attachment 263423

I think it's a bit more complex than that. Rather, we could say pre-Zen4 had "temperature limit" and "temperature limit before reducing boost clocks", and the boosting algorithm targeted the latter in long-term scenarios but the former in short-term ones. With Zen4 these are made one and the same, meaning that you don't get the same drop-off in boost clocks as the chip steps back from "oh shit I'm too hot" territory into "nice and toasty", as it no longer reaches anything it considers too hot, even while boosting.

 

[AusWolf]

I think it's a bit more complex than that. Rather, we could say pre-Zen4 had "temperature limit" and "temperature limit before reducing boost clocks", and the boosting algorithm targeted the latter in long-term scenarios but the former in short-term ones. With Zen4 these are made one and the same, meaning that you don't get the same drop-off in boost clocks as the chip steps back from "oh shit I'm too hot" territory into "nice and toasty", as it no longer reaches anything it considers too hot, even while boosting.

...unless you reach 95 °C, which you definitely will. I guess it's more like a target temperature for Zen 4 rather than what we used to consider a "limit".

Pre-Zen 4 said: "If you reach X temperature, it's too hot and I'll decrease boost."
Zen 4 says: "If you don't reach 95 °C, you're not getting the most out of me."

 

[Valantar]

...unless you reach 95 °C, which you definitely will. I guess it's more like a target temperature for Zen 4 rather than what we used to consider a "limit".

Pre-Zen 4 said: "If you reach X temperature, it's too hot and I'll decrease boost."
Zen 4 says: "If you don't reach 95 °C, you're not getting the most out of me."

Pretty much. For Zen3 (and before) 95 was "nope, not going there, slowing down now" territory, and boost clocks started dropping around 75. For Zen4, it instead treats 95 as fine, and tries to get as high clocks as possible while staying at or slightly below that threshold.

In a way this is a simplification of thermal management: instead of having a dynamic relationship between the three variables of temperature, power draw and clocks, it tries to peg one of them to a stable point and adjust the other two to keep it that way (though not completely - it's not like it stops fans at idle and allows the CPU to sit at 95 then too).

 

[Vario]

My experience so far is stock, on air, the i9 12900KS will boost and hold 5.2 GHz at 1.4V, sit at 100C and consume 350 watt package during Cinebench R23. Undervolted to 1.18V and restricted to 5GHz, i9 12900KS consumes 240 watt package and sit at 83C during Cinebench R23. In gaming loads, it merely consumes 65 watts and sits at 55C game load. The extra 100-400 MHz is nearly worthless unless you like wasting electricity. I am sure the Zen 4 stuff is similar, one could probably dial the voltage and core clock back slightly and reduce power consumption by 50 to 100 watts, and probably get to 70C instead of 95C.

 

[Toothless]

I run Ryzen 5950x at perfect cool 80C tops, undervolted underclocked to 4 ghz. It's silent, fast, I don't need extra 20%, I am not in a hurry.
Also fans doesn't spin - TDP doesn't exceed 130W TDP even in harderst tasks

Mate my 5950x hits 80c tops with PBO/curve tweaked, 4.4 all core with a 4.95 single, as the chip is intended. Your cooler is just not enough for that chip if it can't cool 130w.

 

[Valantar]

Mate my 5950x hits 80c tops with PBO/curve tweaked, 4.4 all core with a 4.95 single, as the chip is intended. Your cooler is just not enough for that chip if it can't cool 130w.

Hey now, don't crap on people's insistence on holding on to outdated and poorly suited ways of clock and power tuning. Not everyone can adjust to new developments

 

[AusWolf]

My experience so far is stock, on air, the i9 12900KS will boost and hold 5.2 GHz at 1.4V, sit at 100C and consume 350 watt package during Cinebench R23. Undervolted to 1.18V and restricted to 5GHz, i9 12900KS consumes 240 watt package and sit at 83C during Cinebench R23. In gaming loads, it merely consumes 65 watts and sits at 55C game load. The extra 100-400 MHz is nearly worthless unless you like wasting electricity. I am sure the Zen 4 stuff is similar, one could probably dial the voltage and core clock back slightly and reduce power consumption by 50 to 100 watts, and probably get to 70C instead of 95C.

I watched JayzTwoCents experimenting with a 7950X overclock/undervolt in this video. If I remember right (I was pretty tired), he didn't have much success. It might be better to leave clocks and voltages alone and enable Eco mode, or decrease PPT and call it a day.

 

[Vario]

I watched JayzTwoCents experimenting with a 7950X overclock/undervolt in this video. If I remember right (I was pretty tired), he didn't have much success. It might be better to leave clocks and voltages alone and enable Eco mode, or decrease PPT and call it a day.

Might also be in need of a future bios update or motherboard revision before overclock/undervolting is worthwhile.

 

[Valantar]

I watched JayzTwoCents experimenting with a 7950X overclock/undervolt in this video. If I remember right (I was pretty tired), he didn't have much success. It might be better to leave clocks and voltages alone and enable Eco mode, or decrease PPT and call it a day.

Tbh, at these power levels I see any attempt at overclocking as utterly pointless. Your CPU is already consuming massive amounts of power, and performing very well. Is it really worth increasing that power draw even further for a 1% performance increase? No. Just no. The only even marginally interesting form of tuning for these chips is power limiting and undervolting. OCing is just boring at this point, at least outside of LN2 and other exotic cooling.

 

[AusWolf]

Tbh, at these power levels I see any attempt at overclocking as utterly pointless. Your CPU is already consuming massive amounts of power, and performing very well. Is it really worth increasing that power draw even further for a 1% performance increase? No. Just no. The only even marginally interesting form of tuning for these chips is power limiting and undervolting. OCing is just boring at this point, at least outside of LN2 and other exotic cooling.

He only tried to overclock with the logic "if it overclocks by X, then it also undervolts by X". This logic might be flawed for Zen 4 which works at its peak pretty much straight out of the box.

 

[Denver]

So now that Intel i9 is 100c and now even amd 7600X is 90c.
What's the overall sentiment about running r5 class cpus at 90+c or any of these new cpus so hot?
For the longest or was always run it cooler or you will degrade your chip.
All of a sudden we're accepting 90C+? No one is worried about the 7900x dying in 3 years right out warranty?

The overclock era is over, welcome to the undervolt era.

 

[Valantar]

He only tried to overclock with the logic "if it overclocks by X, then it also undervolts by X". This logic might be flawed for Zen 4 which works at its peak pretty much straight out of the box.

Yeah, they've probably tuned it so that it goes really close to some combination of architectural/node limits at its nominal power budget. There's definitely more to be had with exotic cooling, but you might not get much more without it, as that might simply require unsafe core voltages.

 

[Zach_01]

I was surprised while reading the review that even Arctic Freezer II wasn't able to get it under 90 °C when loaded, though it wasn't mentioned which size of the radiator was used. I wonder how the "next gen" AIOs will have to be designed to get those temperatures lower, even though that's a new safe temperature.

Because the more cooling you're adding the more the CPU boosts so its 95C all over. This continues until power limit reached. Its as "simple" as that

If any one wants decreased temp can either set lower temp limit or power limit.
Again simple as that...

I watched a test of the 7950X on stock settings with extreme cooling (LN2).
The CPU has reached a max of 5.7+GHz all core boost within stock power setting (200+W PPT) at Tdie: -60~70C

EDIT: typo

 

[pavle]

...The engineers know what they are doing.

Of course they do. No. They do what they are told by the (mis)management.
Just run the parts at next lower setting (105W or 65W or somewhere in between) and they'll be fine.

 

[Valantar]

Of course they do. No. They do what they are told by the (mis)management.
Just run the parts at next lower setting (105W or 65W or somewhere in between) and they'll be fine.

They'll still be fine at 170W. Just hotter.

 

[Bones]

My take on it:
Heat is always an enemy of electronics, even if they are "Supposed" to tolerate it.

I've always had the mindset of "Cooler is better" for electronics since we all know they always fail in terms of overheating/burning up, not from being too cool/freezing.... That would only apply if on Ln2 and even in that case they could still burn up - It's happened before on Ln2 cooling.

When they do fail to work by being too cool or frozen (CB/CBB), you simply let them warm back up and they start working again.
Note that doesn't mean being frozen itself is exactly "Good" for them either, esp the extreme cooling kind of cold - That can make a chip fail over time too if it's frozen one too many times.

If you let one get too hot and it stops working, cooling it probrably won't work - That's the real difference between those terms and conditions.

If I ever get one I will be making all efforts to keep it as cool (Not frozen) as possible but at the same time that's just me - you guys can run them as you see fit to.

 

[TheoneandonlyMrK]

I don't get it personally, so much drama.

ALL my CPUs have been run 24/7 365 at their top thermal limit.

And that's from a q6600, fx8350, 2600X 3800X and many more.

Wtaf do some of you do with your pc, nothing?!.

This shits only new because this generation chips are MADE to run hot, but but if you actually used your shit they already did.

 

[Zach_01]

Its not confirmed yet, and I don't know if it ever will/can but Tjmax of these CPUs is said to be 110~115C.
So 95C is not that bad considering this. Its just us, most of us, that do not accept this kind of temperature in our minds for electronics.

And do not confuse heat with temperature. Its very different things. And most rules of yesterday will be broken by the tomorrow's.
(most)Humans like and feel comfortable when nothing changes.

If I ever get one I will be making all efforts to keep it as cool (Not frozen) as possible but at the same time that's just me - you guys can run them as you see fit to.

No real effort needed really. I mean dont make it sound like it will take a load of money and/or personal work (to set a loop for example).

You can slap the best ever water cooler loop on the AM5 system. It will still hit the 95C. What you are gaining out of better cooling is more boost.

You want lower temp?
Limit the temperature or the power and loose the whatever boost.

1+1=2