90c+ CPUs

[sepheronx]

Hopefully not a repost:

 

[Zach_01]

My only though to this is do you think case design will change again for CPU's like it did with larger GPU's & the removal of the disc drives for better airflow?

What's next for case desgin? Baffles to keep hot GPU air from Flowing up the CPU's because the increase in wattage?
Or do w need to go back to Blower Cards?

I'm not a fan of these new heatsinks on cards that blow hot air all around the case. I prefer when the air flows out the back of the case.
I was recently looking at two AMD graphics cards from the same board partner that were only 20 watts away from each other. I noticed the main difference wasn't size of the heat sink it was weight the higher wattage one weighed more by 1/4 of the less wattage. The fin orinitation was different too one Horizonital the other veritcal, but they both had the same fans.

Desktop CPUs with this kind of wattage (200+W up to 250+W) exist for many many years now. How is Ryzen7000 any different?
On the other hand GPUs gone wild lately (last couple of years at 350+W) and they will continue further... see the 450W of the all new RTX4090 with the 3-4slots coolers

 

[A Computer Guy]

Neither of the things that you've just mentioned is fine. If someone wants you to believe it, then they simply want to sell you something bad.

Hot air is not fine - keep it mind that your case exhaust will put that air in your room.

It's an engineering fail:

If you tweak the PPT and apply curves you can reduce temps by 30 degrees with minimal performance loss.

 

[tabascosauz]

But isn't that exactly why we should distrust AMD's and Intel's statements that "its' fine" to run at these high temps? After all, they would rather just sell you a new CPU after 3 or whatever years... Personally, I have no interest in squeezing out the last drop of performance out of my hardware (at least through hardware measures) and so I have a very conservative setup that I hope can last 10 years or more if necessary.

Pretty much. I have a bit more faith in Intel since their 10ESF/7 process seems still pretty durable at 5GHz despite being a N7 class product. After my experience with 3700X and 4650G degrading I've pretty much accepted that if AMD says it's safe, it's safe......for 3 years

But that's the price of staying competitive. No sense in leaving extra headroom on the table. Raphael is certainly driving that point home, much, much more so than Zen3. Ryzen in games doesn't scale with clock, but in Cinebench it certainly does, and that's precisely the results AMD is trumpeting.

 

[Zach_01]

After my experience with 3700X and 4650G degrading I've pretty much accepted that if AMD says it's safe, it's safe......for 3 years

But that's the price of staying competitive. No sense in leaving extra headroom on the table. Raphael is certainly driving that point home, much, much more so than Zen3.

Was those 2 run at stock settings?

 

[tabascosauz]

Was those 2 run at stock settings?

The 4650G wasn't, and lasted about a week without that much load. The 3700X always was and lasted less than a year before it started fall apart. Of course, Ryzen is susceptible to different things than Core.

But like I said, bulletproof CPUs are a thing of the past.

 

[mplayerMuPDF]

Pretty much. I have a bit more faith in Intel since their 10ESF/7 process seems still pretty durable at 5GHz despite being a N7 class product. After my experience with 3700X and 4650G degrading I've pretty much accepted that if AMD says it's safe, it's safe......for 3 years

But that's the price of staying competitive. No sense in leaving extra headroom on the table. Raphael is certainly driving that point home, much, much more so than Zen3.

And that is exactly why I paid $150 for a 1600 AF (2600 had become ridiculously expensive) in 2020 instead of getting a 3600 for $10 more. I trusted Zen+ with its more conservative boost algorithm and 12 nm node more than Zen 2 with its hyper aggressive boost and (then new) 7 nm node. Because I don't know when I will have the money to upgrade my CPU (and motherboard). And if I do eventually have money to spend on PC parts, then there are multiple things that are a much higher priority (such replacing my 8 year old UltraStar and 8 year old monitor and getting a different case). People need to understand that these corporations are only looking out for themselves and that they also need to look out for themselves, no one else is going to do that for you, certainly not with the current state of the world.

 

[Night]

None of those are aluminium - silvery colored IHSes are nickel plated copper.

Are you sure? Look at the Athlon's IHS, it's fairly scratched and yet it's still "silvery", no sign of copper there.

 

[ARF]

If you tweak the PPT and apply curves you can reduce temps by 30 degrees with minimal performance loss.

I have another idea. How about you don't buy the said CPU and don't tweak the PPT?

 

[Deleted member 185158]

That's entirely possible - when you're working wiht a cooling medium that's that cold, you probably want a buffer in between it and the cores in order to have some control over thermals. Also, the boiling of the LN2 if done directly on the die would be a serious problem for cooling, as you'd have wildly fluctuating temperatures.

Of course, an LN2 pot isn't an IHS, nor is it a thermal interface as much as it is a cooler. It does the same job that your heatpipes and fins on your heatsink do - dissipate heat into another medium - just through boiling LN2 instead of heating air. And since the medium is sub-ambient and the main problem with it is too much cooling, you don't have the problem of heat soaking IHSes and coolers that you get wiht ambient coolers - turning the slow heat transfer of thicker pieces of metal into a useful buffer slowing down fluctuations rather than a hindrance for effective cooling. But this is due to the specific nature of LN2 cooling, and not anything even remotely transferable to other types of above-ambient cooling.


Yes, if we're talking LN2, which fundamentally changes how cooling operates in the first place - where the main problem is no longer dissipating heat, but rather not cooling too much. That makes it fundamentally incomparable.

That depends - delidded cooling can absolutely be good, but getting perfect contact with a tiny bare die is essentially impossible, especially given how fragile they are, so an IHS helps immensely with clamping a beefy cooler to the chip. Your LN2 pot, assuming it's pure copper, would still technically transfer thermal energy between the die and LN2 more efficiently without an IHS - but clamping a giant, >1kg copper pot to a 10x20mm die and expecting good contact? When good contact means microns of height difference across the die? That's not going to happen.

No, the thickness is to make up for the lower Z-height of the LGA socket while maintaining AM4 cooler compatibility, as without it the vast majority of AM4 coolers would no longer have been compatible with AM5. Nor is it nit-picking - the chips would run cooler with a thinner IHS. But as I've been saying all along: as long as performance is there (which it is), then it doesn't matter. The chip doesn't care if it's 95°C or 75°C, and boosts happily until that limits. People would be more comfortable with lower temperatures, but whatever. The IHS isn't ideal, but it does what it needs to do.

See how you broke all that up into different quotes and stuff?? It really makes 6 conversations instead of 1.

Yes LN2 Pot is EXACTLY like an air cooler or waterblock.
BUT (the part about thermal storage comes in here....)
It's massive size and mass is where you STORE a temperature, which all temps above 0 kelvin are considered a heat.

But you couldn't explain why that Big ass plate works better with TEC cooling than the IHS plate that's soldered to a chip.

It's impossible to TEC cool a 2700X. I only know one person that's accomplished it though.

And it wasn't believing a smaller cold plate is better. Not at all. The opposite in fact. A small cold plate doesn't store enough thermal energy in either direction.

Be it you consider it being "cooled" off, or in my perspective heating it up with a little dash in front of the number, say like -25c idle or so.

Sure this blows my mind thinking about it. I've removed a lot of Top plates man. Probably just about every single chip I've ever owned or touched.... some 125 chips maybe.... I dunno, would have to look.

But proof comes in pudding, and I have some of that.

2700X 4ghz vs TEC -

Just before the run ended with load temp - and the score/idle temps... well near idle. w/e time it took to take the screen shot. This set up hit -30 with a water delta of about 50f **Tap to drain **Geothermal liquid cooling on the TEC)

The last picture was to "raise up" so I could contact the chip with my Dice Pot. And it's fun to show off an Xbox chip in a s939 board. And it did get frozen, -26c at load 3800+X2 Toledo.

Look at that nice big plate on lidless 2700x. Doing a fine job storing thermals. Decent temp gradient.

The leakage reduction was really decent too. On average this chip is 4000mhz 1.40v -

Here depicted at only 1.19v but under 20c load.
At idle on defaults, the CPU sits all core at it's maximum frequency 4350mhz. But even at load, because the algorithm isn't told it's frozen, runs 3.9ghz all core. So this below is done with a statically given overclock and v-core reduction. Obviously for testing, I'd rather do it myself. AI doesn't know what I'm trying to do here haha.

SO NO, I cannot agree that a smaller plate would be better even with an ambient water cooling because I've used the smaller plate in the pic from my previous post for that. It works better than the IHS plate and I can prove that with a 5.7ghz FX-9590 (hottest damn chip of all times, which I did de-lid also...) a larger plate stores thermal energy. Just enough to make a difference like you see below.

Hats off to the Nay sayers. It's all good, I ain't mad atcha.

EDIT: I'll include a short thread. It was moved quickly over from a forum that has closed, so it's not really as complete as it once was.
There is a smaller plate in this thread that I didn't get a picture of because it's in another box somewhere. Any how it's a little larger than the IHS plate.
The results, are not quite as good as the larger plate. Really had a difficult time with it actually, it would heat up pretty fast.

2700X De-lid + TEC cooling experiments. | Warp9-systems

Well, it was a pretty smooth de-lid. The glue is cheaper and easier to cut. I noticed the adhesive on the cpu covering the PCB is much cheaper also. You can pick and peel it off with your finger

warp9-systems.proboards.com

Air coolers are meant to be operated at a higher temperature when they are most efficient.

Water blocks with larger cold plates or full copper work better. Not cheap, but additional surface area and thermal storage. Usually better at keeping temp spikes low as well. Because it takes longer to warm the block and coolant and then the radiators. When warm, the radiators are most efficient. So a water loop as a LOT MORE thermal storage than an air cooler and pretty much WHY they work better and AIOs sell like hot cakes. Most people just think water cooling is better, but don't understand why.

The you upgrade to a LN2 pot and it's got almost as much mass as the entire water loop in a giant chunk on top of the CPU. Thermally storing energy from LN2. Holes for additional surface area and depending on those two variables mass and surface area, is how your temp gradient will act.

So as you see, as you step up your cooling solution, you step up your available thermal storage. Be it a plate, or some coolant in a reservoir.

I think AMD should have gone with a much larger chip size than 40mm so they could not only increase the surface area of the chip, but it would have additional thermal storage as well.

HOWEVER!!!

I will agree at some point TOO MUCH or too large of a plate has it's down sides. I'm speaking about in comparison to the current AMD IHS plates released, they should have gone bigger. Probably should have, but hey, engineers design it and technicians repair it or make it better with some tweaks.

Enjoy

 

[Valantar]

See how you broke all that up into different quotes and stuff?? It really makes 6 conversations instead of 1.

No, it makes it possible to respond to individual statements so that things are at least moderately readable, unlike this wall of spaced-out sentences you've posted here.

[massive wall of random statements with dubious relations to the topic]

Okay, I had a long answer here, but reading to the end of your post has made it abundantly clear to me that we're not discussing the same thing. You've been working off of radically different assumptions for what is being discussed here, and arguing against my statements about a given platform with statements that assume every single variable is up for discussion. It isn't. Yes, there is a relation between die size, IHS size, and IHS thickness. For a given platform, IHS size is fixed, as it's determined by the socket and package dimensions (with maybe a mm or two of leeway for tweaking the design in X and Y directions. So: the area of the IHS for any given platform is fixed, and not up for discussion - at least unless it's brought up at the beginning of the damn discussion. Which you didn't. You responded to me saying the IHS was too thick by arguing that no, a thicker IHS would be better. Which ... no. Just no.

The question I've been trying to discuss with you this whole time has been this: given the size of AM5 heatsinks, assuming a cooler designed for the platform, would it be better with a thinner IHS? This is what I have argued for. Neither more nor less. You've come here dragging in all kinds of absurd examples, which are entirely irrelevant to this discussion. If you have been trying to argue this whole time that the fundamental characteristics of the platform are wrong, then why didn't you say so? It is so utterly and completely frustrating to try have a discussion with someone who doesn't even try to clarify the parameters of what they're discussing. This is also why you've been constantly contradicting yourself in all kinds of ways: because you've been responding to my arguments that presuppose that all other factors are pre-given with statements that assume that everything is in flux, making your logic fundamentally inapplicable to the question at hand.

To sum up: for any given surface area, a thinner IHS is better, assuming it has sufficient thickness to be mechanically stable and isn't so thin as to entirely hinder spreading heat outwards (which generally overlaps with mechanical stability - an IHS so thin as to stop heat spreading outwards sufficiently would also bend under pressure). Which I have come to believe that you agree with.

Whether the AM5 socket should have had a larger IHS is an entirely different discussion.

Conclusion:



AMD Ryzen 7000 CPU Direct-Die Cooling Can Offer Up To 20C Lower Temps, IHS Hot Spots & Temps Analyzed (wccftech.com)

AMD just produced the most terrible shitshow in its history.

Why on earth are you linking garbage rumor mill site WCCFtech instead of Der8auer's video for this? You prefer to use a secondary source notorious for terrible analysis rather than the original source? Cool.

Neither of the things that you've just mentioned is fine. If someone wants you to believe it, then they simply want to sell you something bad.

Hot air is not fine - keep it mind that your case exhaust will put that air in your room.

Jesus, have you read anything that has been posted in all the various 7000-series threads? CPU core temperature does not equal heat exhausted into your room. Power draw equals heat exhausted into your room. The absolute temperature of your CPU is not directly tied to the amount of heat energy put out by the system - if it was, then you wouldn't be able to lower temperatures with better cooling on any CPU ever.

The absolute temperature of the CPU is a function of power consumption and how quickly the cooling setup (TIM, IHS, TIM, cooler, ambient air temp) is able to move heat away from the core.

The heat output into ambient air is a function of the power consumption of the core. Period. If your CPU is consuming 230W consistently, then 230W of heat energy is being dumped into your room, regardless if the CPU cores are 30°C or 95°C.

But isn't that exactly why we should distrust AMD's and Intel's statements that "its' fine" to run at these high temps? After all, they would rather just sell you a new CPU after 3 or whatever years... Personally, I have no interest in squeezing out the last drop of performance out of my hardware (at least through hardware measures) and so I have a very conservative setup that I hope can last 10 years or more if necessary.

That's what Eco mode is for. IMO, that should have been the default setting for these CPUs, but sadly the competitive situation makes things different right now.

Are you sure? Look at the Athlon's IHS, it's fairly scratched and yet it's still "silvery", no sign of copper there.

It takes quite a lot of effort to scratch through a nickel plating. It's not tin foil.

 

[tabascosauz]

That's what Eco mode is for. IMO, that should have been the default setting for these CPUs, but sadly the competitive situation makes things different right now.

At least on 1CCD Eco Mode doesn't solve anything on its own, it's still way too much Vcore. Same with only CO, too much power. Gotta combine them to make a difference to temps, and that's where the problem lies - Intel generally has pretty hefty undervolt headroom across the board but AMD has pretty wild variance in achievable CO from sample to sample. Haven't really ever seen great CO headroom from 2CCD parts in general

 

[AusWolf]

Pretty much. I have a bit more faith in Intel since their 10ESF/7 process seems still pretty durable at 5GHz despite being a N7 class product. After my experience with 3700X and 4650G degrading I've pretty much accepted that if AMD says it's safe, it's safe......for 3 years

But that's the price of staying competitive. No sense in leaving extra headroom on the table. Raphael is certainly driving that point home, much, much more so than Zen3. Ryzen in games doesn't scale with clock, but in Cinebench it certainly does, and that's precisely the results AMD is trumpeting.

The 4650G wasn't, and lasted about a week without that much load. The 3700X always was and lasted less than a year before it started fall apart. Of course, Ryzen is susceptible to different things than Core.

But like I said, bulletproof CPUs are a thing of the past.

The 3700X degraded at stock? What were the symptoms if you don't mind me asking? I'm getting less tempted to invest in an AMD system again when I hear about things like this.

Edit: How did you cool it? How hot did it run? Sorry for the million questions.

 

[tabascosauz]

The 3700X degraded at stock? What were the symptoms if you don't mind me asking? I'm getting less tempted to invest in an AMD system again when I hear about things like this.

Edit: How did you cool it? How hot did it run? Sorry for the million questions.

Just air. Some combination of U9S, Dark Rock Pro 4 and C14S. The 3700X doesn't really boost so doesn't really get hot, 70s ish max maybe. Most of the time the VRM on that shitty board ran hotter (B450I Aorus Pro Wifi). Both Fabric and cores degraded, about a year in started seeing both 18s (Cache Hierarchy) and 19s (Bus/Interconnect).

Been through the wringer with AMD but I wouldn't be so pessimistic. Just Zen 2 sus - N7FF yields and SP problem, CPPC, WHEA, etc. all new concepts. AMD started dialing it back on current when B550 released (AGESA 1006??), then Zen 3 got much more conservative, and now 5800X3D feels like a locked Intel CPU lol (Robert said 1.35V tops, more like 1.28V and 1.15V after CO)

that said, N5 is a new process and you know how I feel about AMD when they get their hands on brand new toys

 

[AusWolf]

Just air. Some combination of U9S, Dark Rock Pro 4 and C14S. The 3700X doesn't really boost so doesn't really get hot, 70s ish max maybe. Most of the time the VRM on that shitty board ran hotter (B450I Aorus Pro Wifi). Both Fabric and cores degraded, about a year in started seeing both 18s (Cache Hierarchy) and 19s (Bus/Interconnect).

Been through the wringer with AMD but I wouldn't be so pessimistic. Just Zen 2 sus - N7FF yields and SP problem, CPPC, WHEA, etc. all new concepts. AMD started dialing it back on current when B550 released (AGESA 1006??), then Zen 3 got much more conservative, and now 5800X3D feels like a locked Intel CPU lol (Robert said 1.35V tops, more like 1.28V and 1.15V after CO)

that said, N5 is a new process and you know how I feel about AMD when they get their hands on brand new toys

Maybe that was AMD still trying to figure out how a chiplet-based CPU works. You scared me by saying that it degraded at stock nonetheless. That should never happen with any CPU.

Maybe I'll just buy an A770 GPU for sh*ts and giggles now, and save up for a platform upgrade for my birthday in March. Early adopter problems will have surfaced by then, and prices might come down a bit as well.

 

[Vario]

Just air. Some combination of U9S, Dark Rock Pro 4 and C14S. The 3700X doesn't really boost so doesn't really get hot, 70s ish max maybe. Most of the time the VRM on that shitty board ran hotter (B450I Aorus Pro Wifi). Both Fabric and cores degraded, about a year in started seeing both 18s (Cache Hierarchy) and 19s (Bus/Interconnect).

Been through the wringer with AMD but I wouldn't be so pessimistic. Just Zen 2 sus - N7FF yields and SP problem, CPPC, WHEA, etc. all new concepts. AMD started dialing it back on current when B550 released (AGESA 1006??), then Zen 3 got much more conservative, and now 5800X3D feels like a locked Intel CPU lol (Robert said 1.35V tops, more like 1.28V and 1.15V after CO)

that said, N5 is a new process and you know how I feel about AMD when they get their hands on brand new toys

Could it have been your motherboard's power delivery degrading?

 

[tabascosauz]

Maybe that was AMD still trying to figure out how a chiplet-based CPU works. You scared me by saying that it degraded at stock nonetheless. That should never happen with any CPU.

Maybe I'll just buy an A770 GPU for sh*ts and giggles now, and save up for a platform upgrade for my birthday in March. Early adopter problems will have surfaced by then, and prices might come down a bit as well.

Just follow the rule - if you want less problems, wait 6 months after AMD launch

I'm not even kidding. Whenever I follow the rule I'm happy, whenever I don't I get burned

Could it have been your motherboard's power delivery degrading?

Had that CPU in 4 separate boards. Last one also hosted my 5900X for months

though......now that you mention it, the B450 Aorus had something like 76% power reporting deviation..............but the Fabric side of things would still be inexcusable

@Vario on that note, I'd like to share what AMD said about HWInfo's power reporting deviation:

"We want to be clear with our customers: AMD Ryzen processors contain a diverse array of internal safeguards that operate independently of external data sources. These safeguards enforce the safety and reliability of the processor during stock operation. Based on our initial assessment, we do not believe that altering external telemetry in the manner described by those public reports would have a material impact on the longevity or safety of a user's processor."

So, even if it was Gigabyte that shafted me (not the first time), AMD still wouldn't be off the hook I can also guarantee that there is zero way a C14S or Dark Rock Pro 4 can keep a supposedly 115W 1CCD below 75C

 

[AusWolf]

Just follow the rule - if you want less problems, wait 6 months after AMD launch

I'm not even kidding. Whenever I follow the rule I'm happy, whenever I don't I get burned

Sounds like a plan. A770 it is (for now)!

Had that CPU in 4 separate boards. Last one also hosted my 5900X for months

though......now that you mention it, the B450 Aorus had something like 76% power reporting deviation..............but the Fabric side of things would still be inexcusable

@Vario on that note, I'd like to share what AMD said about HWInfo's power reporting deviation:

Is it possible that because of the power reporting deviation, the CPU actually got more power than it should have and that's why it deteriorated?

 

[tabascosauz]

Sounds like a plan. A770 it is (for now)!

Is it possible that because of the power reporting deviation, the CPU actually got more power than it should have and that's why it deteriorated?

It is possible but again, the CPU didn't get nearly as much power as 76% suggests, because it would be straight uncoolable. AMD is probably right that it's not enough to bypass safeties. More likely a few watts extra - the other 3 boards have normal power deviation and the difference was only ever a few dozen points and maybe 2-3C.

You want to be Intel's early adopter (before they pull the good yields to make -KS CPUs), you do not want to be AMD's early adopter (before good yields even exist).

 

[Vario]

Just follow the rule - if you want less problems, wait 6 months after AMD launch

Generally a solid strategy with any Tech product launch.

edit: Also true about the Intel yields, as well. Anecdotally, from my experience it often seems like early bins of K (even a few years ago before they even had KS as a SKU) were better overclockers, and then they start dumping garbage silicon near the end of the cycle.

 

[AusWolf]

It is possible but again, the CPU didn't get nearly as much power as 76% suggests, because it would be straight uncoolable. AMD is probably right that it's not enough to bypass safeties. More likely a few watts extra - the other 3 boards have normal power deviation and the difference was only ever a few dozen points and maybe 2-3C.

Still, I think I'll stay with Asus. I've gone through a few B550 and A520 boards from them (I like tinkering), and not only is their quality excellent, but their power reporting deviation seems to be spot on as well.

You want to be Intel's early adopter (before they pull the good yields to make -KS CPUs), you do not want to be AMD's early adopter (before good yields even exist).

I've never thought about it this way, but it makes sense.

Actually, it's not even possible to not be an early adopter of Intel because they pull generations out of their hat faster than lightning. By the time you get used to your new system, the next platform is already out. They played this with me with Kaby Lake - Coffee Lake came out only a couple months after I bought my 7700 (just when I thought they would stay on 4 cores forever). And then, they played the same trick with Alder Lake that came out about half a year after I bought the 11700. I can't even follow all their releases anymore.

AMD on the other hand, keeps their platforms and CPU SKUs around for longer, so they have more time to mature and initial hiccups to be ironed out.

 

[Zach_01]

Power Reporting Deviation value under 100% while the CPU is on 100% load only means worse thermals due to increased power. Its a push-over to advertised stock PB/PBO limits. Nothing more.
This cannot bypass the internal silicon safeties of the CPU.

It is possible but again, the CPU didn't get nearly as much power as 76% suggests, because it would be straight uncoolable. AMD is probably right that it's not enough to bypass safeties. More likely a few watts extra - the other 3 boards have normal power deviation and the difference was only ever a few dozen points and maybe 2-3C.

Was it PRD 76% along with 88W PPT? Because you do know PRD value alone doesn't mean anything.
I've seen some CPUs on screenshots here at TPU that had a awfully low PRD but also their PPT value was too low at the same time during all core loads/tests.

I mean if you had 88W PPT, 76% PRD this is:
88 / 0.76 = 115~116W

But if PPT was also lower that 88W (lets say 75) then true wattage is something else
75 / 0.76 = 98~99W

 

[AusWolf]

I mean if you had 88W PPT, 76% PRD this is:
88 / 0.76 = 115~116W

This is what I was thinking. Is it possible that this led to the degradation of the CPU? (Regardless of its temperature)

 

[Vario]

This is what I was thinking. Is it possible that this led to the degradation of the CPU? (Regardless of its temperature)

Probable. My understanding was some board manufacturers were stealth overvolting in this manner.

 

[Zach_01]

This is what I was thinking. Is it possible that this led to the degradation of the CPU? (Regardless of its temperature)

Unlikely

@tabascosauz talked about IF degradation which related to DRAM OC and IF v/f.
Has nothing to do with false CPU telemetry to force it exceed stock package power.