Crazy-high Ryzen 3900X temperatures (95c immediately)

Centauri

2[H]4U
Joined
Mar 1, 2003
Messages
2,467
Ryzen 3900X with a Corsair H100i rad and MX-4 in between. ASRock X370 Taichi.

I just noticed that temps were idling at 70ish and immediately hitting the 95c thermal limit on load. Which got me looking at voltages and the BIOS had the CPU at 1.45ish for god knows what reason. So with a bit of research I dial it back all the way down to 1.275v. And while that lowered the idles into the lower 60s, it still hits 95c immediately on load.

And I see no problems with the contact between the heatsink and the CPU.

Ideas?
 
No airflow in case whatsoever? Very poor thermal goop application? Defective heatsink? defective heatsink fan(s)? Could also be defective thermal monitors, have seen that a few times.
 
reviews i see put the flow rate of that at .25gpm With a delta temp for an i7-5930k in the mid 50's over ambient. Put a bit less of that heat but in 60% of the space and I'm thinking it's just not good enough to cool down such a concentrated amount of heat to reasonable levels
 
Obviously there's something wrong. First, is it even working? Feel the hoses while it's running, there should be a small amount of vibration from the liquid moving through. Do you have it directly connected to a power connector (molex/sata)? Does the pump head feel loose against the CPU? Some AIO mounting hardware has very conservative thread stops to prevent overtightening which can be a problem on thinner motherboards.
 
You probably have poor contact or too much or to little tim.

Remount it again and give it a little jiggle on the socket.

Make sure protective plastic isn't stuck to copper heat plate.
 
Poor contact most likely -- but why use one when you could just get a Noctua air cooler that works better and is near silent?
 
It's important to get complete coverage of the TIM / paste across the entire heat spreader. I can't stress this enough. I've used the dot method for years and it just isn't good enough for modern CPU's like the 3900X.
 
Check the hoses. If one is hot and the other is cool you likely have a dead pump or theres gunk in the loop clogging up the block or one of the hoses.
 
What is iCUE telling you? Pump speed, fans, coolant temp, etc.?
 
No airflow in case whatsoever? Very poor thermal goop application? Defective heatsink? defective heatsink fan(s)? Could also be defective thermal monitors, have seen that a few times.

This is an existing build that has been running since 2017. Only recent addition is the change to a 3900X. I also can't imagine an airflow situation that would cause a CPU to rocket 30 degrees instantaneously.

Obviously there's something wrong. First, is it even working? Feel the hoses while it's running, there should be a small amount of vibration from the liquid moving through. Do you have it directly connected to a power connector (molex/sata)? Does the pump head feel loose against the CPU? Some AIO mounting hardware has very conservative thread stops to prevent overtightening which can be a problem on thinner motherboards.

I re-applied the heatsink last night to ensure even tightness and snugness. I can feel the vibration of the pump when touching its heatsink housing and the hoses have a slight vibration to them as well.

You probably have poor contact or too much or to little tim.

Remount it again and give it a little jiggle on the socket.

Make sure protective plastic isn't stuck to copper heat plate.

It's a 2 year old build, no plastic to remove. It's definitely mounted securely at this point.

I can't imagine this thing hitting 95c instantly even with zero thermal compound.

Poor contact most likely -- but why use one when you could just get a Noctua air cooler that works better and is near silent?

Which air cooler is better than a 2x 120 radiator setup?

It's important to get complete coverage of the TIM / paste across the entire heat spreader. I can't stress this enough. I've used the dot method for years and it just isn't good enough for modern CPU's like the 3900X.

I'll check that again.

Check the hoses. If one is hot and the other is cool you likely have a dead pump or theres gunk in the loop clogging up the block or one of the hoses.

One is definitely warmer than the other.

What is iCUE telling you? Pump speed, fans, coolant temp, etc.?

3k~ RPM pump, 1600~ RPM fans and a temp of 41c
 
I had an H100i V2 that exhibited the same behavior. It just quit cooling even though everything was reading ok. Corsair RMA'd it for me. The pump even though it was running wasn't pumping fluid properly. Unit was a couple of years old IIRC.
 
Yeah, I'll reach out next week. Frustrating but I'd rather it be the AIO than the chip I suppose. I actually went to bed last night wondering if I somehow had the luck of getting a chip that didn't have its IHS welded. :p

Thanks for the help guys!
 
I just don't trust the AIO coolers. Noctua NH-D14 is pretty solid

I've used several Corsair and Thermaltake units over the years with great results. I've never had one fail on me. Some of them are more than five years old now.
 
My 3900x in low core count/high clock speed turbo scenarios will set a high voltage like 1.4-1.45v, this is normal. Under an all core load like cinebench I see 1.2875 to 1.3125v using precision boost overdrive. Under low load/no load (<1ghz any single core) I sit at around 0.95v.

I highly recommend using ryzen master and just setting it up to either do auto overclocking or precision boost overdrive. I found that auto overclocking pushed slightly higher voltages on my motherboard than precision boost without much of a clock increase so I'm sticking with precision boost.
 
It's important to get complete coverage of the TIM / paste across the entire heat spreader. I can't stress this enough. I've used the dot method for years and it just isn't good enough for modern CPU's like the 3900X.

Care to expand on this a little? I have trouble seeing how the end result would be different if I just let the pressure of the heat sink spread the tim out
 
Care to expand on this a little? I have trouble seeing how the end result would be different if I just let the pressure of the heat sink spread the tim out

It doesn't really matter, as long as there is enough TIM you'll be fine.

 
I've used several Corsair and Thermaltake units over the years with great results. I've never had one fail on me. Some of them are more than five years old now.

Same. Still have a 7 year old H80i goes strong. Have a few year old Arctic Liquid Freezer 240 that works perfectly as well. Actually, if that Liquid Freezer had an AM4 mount I would be using that on my CPU instead of going custom loop. That thing is a champ at cooling.
 
This is why Der8auer suggested spreading paste in to a thin layer across the entire IHS. Kryonaut comes with a little spatula just for that. It's how I've been doing it for probably 10 years now. You actually use less TIM since you only need a very thin layer. I usually make it damn near transparent.
 
OP, glad the stock cooler ended up working out! what kind of of CPU-only power consumption are you seeing with those temps?

This is why Der8auer suggested spreading paste in to a thin layer across the entire IHS. Kryonaut comes with a little spatula just for that. It's how I've been doing it for probably 10 years now. You actually use less TIM since you only need a very thin layer. I usually make it damn near transparent.
seconded to this... also my method since like Athlon & Pentium 4 days. I do the "big X and spread w/ a plastic credit/gift card until its totally even and thin" method and it hasn't done me wrong.
 
Yeah I guess it's becoming a "to each their own" thing because I see people still using dot/line/X methods successfully and in the tests I've seen there's less than 2 degree variance between the "best" and "worst" type of applications. I'll stick with mounting my heatsink properly and saving however many seconds it takes to spread tim
 
Sounds like either not enough pressure on the block / cpu contact , or a dead pump. I guess it is also possible that air got into the system somehow?

Not sure why so many people are so mad at AIOs, did they steal your lunch or something?
 
custom loops wont be able to cool the zen2's any better than correctly working aio's. You'll get maybe 2-4C difference between them in peak and average workload temps. Hardly justifying the additional cost.

Unlike previous cpu's the problem isn't in the block transporting heat to the radiator ...which is where aio's have been deficient historically due to low flow rates.

The reason why both are sucking at cooling the cpu's is because copper can't move the heat from the tiny cores to the water fast enough to keep the delta T down. Both aio's and diy's utilize the same basic water block makeup, they use a copper plate to transport heat to the surface the water contacts and that part is the same in both. That's why there's little difference between them.


We'll have to wait and see if the diy water cooling market comes out with tiny heatpipe assisted water blocks to combat the slower performance out of solid copper.
 
custom loops wont be able to cool the zen2's any better than correctly working aio's. You'll get maybe 2-4C difference between them in peak and average workload temps. Hardly justifying the additional cost.

Unlike previous cpu's the problem isn't in the block transporting heat to the radiator ...which is where aio's have been deficient historically due to low flow rates.

The reason why both are sucking at cooling the cpu's is because copper can't move the heat from the tiny cores to the water fast enough to keep the delta T down. Both aio's and diy's utilize the same basic water block makeup, they use a copper plate to transport heat to the surface the water contacts and that part is the same in both. That's why there's little difference between them.


We'll have to wait and see if the diy water cooling market comes out with tiny heatpipe assisted water blocks to combat the slower performance out of solid copper.

This. And it's only going to get worse with 7nm+/5nm/3mn etc.

My 3600 @ 76W is getting the same temps as my 1600 @ 150W, both on 280mm AIO. My Vega at 150~180W gets the same or lower temps despite being on a 240mm AIO. Vega is ~twice the size of a Zen1 die, Zen1 die is ~twice the size of a Zen 2 chiplet... things are gonna get toasty. Zen 3 or 4 on 5nm is gonna be a nightmare to cool if this keeps up. 60-80W on a 50mm^2 chiplet? jesus
 
Care to expand on this a little? I have trouble seeing how the end result would be different if I just let the pressure of the heat sink spread the tim out

Because there is no sure way of knowing if you applied enough to cover the whole span. You have to disassemble and inspect and by that time have to reapply.

It's easier to just use your finger tip and spread it around thinly.
 
Because there is no sure way of knowing if you applied enough to cover the whole span. You have to disassemble and inspect and by that time have to reapply.

It's easier to just use your finger tip and spread it around thinly.
I've always been a fan of a doubled edged razor blade for spreading TIM - easy to get nice THIN and even coating and less mess than using a finger (even with a glove.)
 
This. And it's only going to get worse with 7nm+/5nm/3mn etc.

My 3600 @ 76W is getting the same temps as my 1600 @ 150W, both on 280mm AIO. My Vega at 150~180W gets the same or lower temps despite being on a 240mm AIO. Vega is ~twice the size of a Zen1 die, Zen1 die is ~twice the size of a Zen 2 chiplet... things are gonna get toasty. Zen 3 or 4 on 5nm is gonna be a nightmare to cool if this keeps up. 60-80W on a 50mm^2 chiplet? jesus


Future chips are just gonna have to spread the chiplets farther apart. This will need bigger physical package size. But means the heat will be less concentrated.
 
I've always been a fan of a doubled edged razor blade for spreading TIM - easy to get nice THIN and even coating and less mess than using a finger (even with a glove.)

Of course but the finger tip.is mundane but it works! Razor is superior of course.
 
the argument for using a dot and pushing it flat to spread out vs spreading it evenly and thinly off the bat with something like a credit card is _only_ to waste as little heat sink paste as possible . Is it really that precious to you that you'd risk not having uniform full coverage? that fraction of a ml worth that you scrape off? ... Skip a morning coffee - there's your extra thermal paste ... put a full coat on and be done with it.
 
the argument for using a dot and pushing it flat to spread out vs spreading it evenly and thinly off the bat with something like a credit card is _only_ to waste as little heat sink paste as possible . Is it really that precious to you that you'd risk not having uniform full coverage? that fraction of a ml worth that you scrape off? ... Skip a morning coffee - there's your extra thermal paste ... put a full coat on and be done with it.

Some people slather the stuff on like buttered bread and instantly have massive heat problems.
 
you can't make everything idiot proof. But if you have two methods and one is objectively easier to get right, then that's the right method to use. Spreading things out thin and uniform with a straight edge is by far a smarter choice than relying on perfectly even pressure while screwing the heatsink down to do the job for you. You only do the latter when you absolutely can't afford to waste any of the paste or it's some liquid with such a high surface tension that it wont stay evenly spread out without the pressure of the heatsink down on it. Even gallium doesn't get applied that way so I'm not sure what kind of heat sink fluid that would be.
 
Future chips are just gonna have to spread the chiplets farther apart. This will need bigger physical package size. But means the heat will be less concentrated.
AMDs on the right track with big MCM packages. Even the dual-die Opty 6344 I had a few years back ran pretty cool with middling cooling.

I imagine designing coldplates with offset & asymmetrical die arrangements in mind might help too? changing locations of inlets/microfins/etc? and moving heatpipes around on air coolers
 
AMDs on the right track with big MCM packages. Even the dual-die Opty 6344 I had a few years back ran pretty cool with middling cooling.

I imagine designing coldplates with offset & asymmetrical die arrangements in mind might help too? changing locations of inlets/microfins/etc? and moving heatpipes around on air coolers

I imagine soon we'll have TEC devices built into the chips themselves before too long, along with micro channel water cooling flowing through the dies themselves.
 
the argument for using a dot and pushing it flat to spread out vs spreading it evenly and thinly off the bat with something like a credit card is _only_ to waste as little heat sink paste as possible . Is it really that precious to you that you'd risk not having uniform full coverage? that fraction of a ml worth that you scrape off? ... Skip a morning coffee - there's your extra thermal paste ... put a full coat on and be done with it.

Saving paste isn't the reason. The reason is metal to metal is a lot better than metal to paste to metal contact. Paste is there to fill in the gaps/imprefections but metal to metal is always more desirable when possible. The problem is we can't see between the IHS and heatsink and the size of today's head spreaders makes the dot method inadequate.
 
AMDs on the right track with big MCM packages. Even the dual-die Opty 6344 I had a few years back ran pretty cool with middling cooling.

I imagine designing coldplates with offset & asymmetrical die arrangements in mind might help too? changing locations of inlets/microfins/etc? and moving heatpipes around on air coolers

Most waterblocks have the inlet channel directly in the center, where most cpu cores were located before zen2. The actual heat radiating fins were to the sides of that channel. So, they're probably mostly already over the cores now. It doesn't make a difference.

The limiting factor is solid metal between the cores and the water moves heat too slowly to keep the delta T down to where we're used to seeing it with bigger chips. Moving the heat fins around a mm or two here and there wont matter because the problem occurs before the heat even reaches them.

Something better than solid metal needs to transport heat from the cpu outward. We have such a thing already in the form of heatpipes. Properly made ones can move heat almost 90 times faster than solid copper. I imagine we'll see heatpipe infused water blocks coming to market in the not too distant future. Tiny thin heatpipes setup to draw heat away from the tiny core area and out to the sides of the block and micro-fins everywhere inside to soak it all up.

Best case for manufacturer (on cpu die) solutions would be to move to a non-silicone centric wafer design, there are gallium alloys/compounds that significantly reduce resistance of circuits that are going into 5g modems this coming year that could make it's way to cpu's. But that doesn't really solve the issue of heat density, just reduces the heat. Either way, good for us.
 
Back
Top