Hot Temps for 5800X3D

[JrRacinFan]

I don't think it matters but you can try swapping CPU and GPU in your loop.

Correct. The loop order wont matter, it will all settle into equilibrium. What could matter is push/pull configuration. @Chicken Patty I cant recall, whats your radiator placement?

 

[dgianstefani]

What thermal paste do you use?

These are my temps under 240 Hz gaming load.

Prime 95 is not a realistic load, so I wouldn't really worry.

 

[ThrashZone]

hello guys

I recently upgraded my 5600X to a 5800X3D. My understanding is that this chip due to the v-cache design runs a bit hotter than a 5800X per say. Now I won't compare it to my 5600X as that was a Ryzen 5, this is a Ryzen 7, so more cores, etc. CPU is the only thing that changed. It's cooled by an Bykski CPU block, to a Bykski distro plate with two EK S360 rads. the GPU is in the loop, it's a 3090Ti but my understanding is this should be enough cooling for both.

So idle is about 45ºc then if I put it under 100% (Prime 95 Small FFT's) it goes straight to 90ºc and stays there. CPU doesn't throttle, but should it not be running cooler than this under water?

The way this distro plate works is RESERVOIR > PUMP > GPU > RESERVOIR > CPU > RESERVOIR > TOP RAD > FRONT RAD > RESERVOIR > PUMP and so forth. that's the routing. Only thing I can think of is that after coming ouit from the GPU it doesn't go through a radiator and it's already hot? my GPU is running great. Idles around 36ºc, full load stress test 58ºc.

Originally I did have an EK block on, so when I installed the Bykski block I re-did the thermal paste obviously, switched to Thermal Grizzly Kryonaut Thermal Paste which is what I have on the GPU. I've tried loosening the CPU block, tightening. Are these temps normal?

Thanks in advance for any help guys.

Hi,
Hottest item should be hit first with coolest fluid possible either coming from a rad or reservoir and cpu usually is the hottest item.
Example reservoir-pump/ cpu/ gpu/ rad/ rad/.. back to reservoir.

Fluid in general does equalize but you'd have to have multiple water tem sensors to be accurate say before and after cpu/ gpu
I'd personally prioritize the hottest item/ cpu from pump or rads which ever is more convenient pipping wise

So with that in mind I use two d5 pumps
Res-pump combo/ rad/ rad/ 2nd d5 pump/ cpu/ gpu back to res-pump combo. repeat.
Of course I use soft tubing as well and don't mind the octopus look either because it's about oc'ing not looks

 

[Minxie]

Stop worrying and enjoy that beast. Temperatures do not matter on Zen 3 and above. AMD is obviously building these things to redline to get the most performance possible.

 

[QuietBob]

idle is about 45ºc

With 25c room ambient you should be idling below 40, and most likely in the 30-35c range. Perhaps your CPU does not enter deep sleep due to power plan settings or background processes?

 

[freeagent]

ARGB software can be a real biznatch. I don’t have armoury crate installed or anything like that. I use open RGB because it has a low footprint and i can just turn it off. Hwinfo is the one program I use that is always polling.

 

[ThrashZone]

the sensor on the block says the water is 34c, that's pretty warm I'd say. Just finished a run though. I will play around with it, see what settings work best, but atleast for now i'm not scared it'll melt off heheh

Hi,
34c water temp at 90c core temp well if good or bad depends on amount of rads and fan rpm's
You'd hope for only a few celsius above room temp max but room 25c is pretty typical I'd say unless you're lowering room temp for maximum cooling sessions.

I'd doing nothing really water temp is nearly 32c
I have fans at silent range on 2 gtx-280 rads.
I'm sure if I ran anything that maxed turbo clocks I'd hit somewhere near 90c to and of course water temp would go higher

Distro plates are purely for looks
In a typical reservoir config fluid exist out the bottom and hot fluid would dump in and mix in with existing from the top so using distro plate fluid isn't mixed together nearly as well as a typical tube reservoir.

 

[dgianstefani]

ARGB software can be a real biznatch. I don’t have armoury crate installed or anything like that. I use open RGB because it has a low footprint and i can just turn it off. Hwinfo is the one program I use that is always polling.

+1 for open RGB, I use it in place of iCUE for my RAM, Synapse I don't really have issue with.

 

[Deleted member 185158]

hello guys

I recently upgraded my 5600X to a 5800X3D. My understanding is that this chip due to the v-cache design runs a bit hotter than a 5800X per say. Now I won't compare it to my 5600X as that was a Ryzen 5, this is a Ryzen 7, so more cores, etc. CPU is the only thing that changed. It's cooled by an Bykski CPU block, to a Bykski distro plate with two EK S360 rads. the GPU is in the loop, it's a 3090Ti but my understanding is this should be enough cooling for both.

So idle is about 45ºc then if I put it under 100% (Prime 95 Small FFT's) it goes straight to 90ºc and stays there. CPU doesn't throttle, but should it not be running cooler than this under water?

The way this distro plate works is RESERVOIR > PUMP > GPU > RESERVOIR > CPU > RESERVOIR > TOP RAD > FRONT RAD > RESERVOIR > PUMP and so forth. that's the routing. Only thing I can think of is that after coming ouit from the GPU it doesn't go through a radiator and it's already hot? my GPU is running great. Idles around 36ºc, full load stress test 58ºc.

Originally I did have an EK block on, so when I installed the Bykski block I re-did the thermal paste obviously, switched to Thermal Grizzly Kryonaut Thermal Paste which is what I have on the GPU. I've tried loosening the CPU block, tightening. Are these temps normal?

Thanks in advance for any help guys.

When I did custom.

Pump - cpu - gpu - rad - rad - rad - rad - res.

On most cases, it doesn't matter, but typically run the water to the cpu first if it's a high wattage chip.

 

[dgianstefani]

When I did custom.

Pump - cpu - gpu - rad - rad - rad - rad - res.

On most cases, it doesn't matter, but typically run the water to the cpu first if it's a high wattage chip.

Even if GPU runs higher wattage, CPU is a much smaller die, so temperature differential between die and liquid should be max for maximum heat transfer coefficient.

If I could be bothered to delid my 5800X3D and watercool it with a contact frame I would see much better temperatures, but they're already fine and the chip can't be OC'd without resorting to BLCK so it's not worth it.

 

[Deleted member 185158]

Even if GPU runs higher wattage, CPU is a much smalller die, so temperature differential between die and liquid should be max for maximum heat transfer coefficient.

If I could be bothered to delid my 5800X3D and watercool it with a contact frame I would see much better temperatures, but they're already fine and the chip can't be OC'd without resorting to BLCK so it's not worth it.

I dunno, just was they way I'd set them up. GPU is higher wattage, but generally a full cover block and a lot more surface area, so cpu first.

That's a shame the X3D has its nuts chopped like that. You're right, not worth a delid though.

I love delidded chips .

 

[LordsnLadies]

This is what I get with a corsair 150i elite 360mm AIO at -30mv on all cores and stock power limits.

Tried everything to get all core 4.4ghz on cinebench, but the max I can do is 4.2ghz. Setting some power limits allows me to get 10c lower temps at 4-4.1ghz.

A friend of mine with basically the same specs, except a sythe fuma 2 air cooler, gets 15c~ lower temps at 300mhz higher clock than me. So my chip is just the worst lottery ever.

 

[Mussels]

hello guys

I recently upgraded my 5600X to a 5800X3D. My understanding is that this chip due to the v-cache design runs a bit hotter than a 5800X per say. Now I won't compare it to my 5600X as that was a Ryzen 5, this is a Ryzen 7, so more cores, etc. CPU is the only thing that changed. It's cooled by an Bykski CPU block, to a Bykski distro plate with two EK S360 rads. the GPU is in the loop, it's a 3090Ti but my understanding is this should be enough cooling for both.

So idle is about 45ºc then if I put it under 100% (Prime 95 Small FFT's) it goes straight to 90ºc and stays there. CPU doesn't throttle, but should it not be running cooler than this under water?

The way this distro plate works is RESERVOIR > PUMP > GPU > RESERVOIR > CPU > RESERVOIR > TOP RAD > FRONT RAD > RESERVOIR > PUMP and so forth. that's the routing. Only thing I can think of is that after coming ouit from the GPU it doesn't go through a radiator and it's already hot? my GPU is running great. Idles around 36ºc, full load stress test 58ºc.

Originally I did have an EK block on, so when I installed the Bykski block I re-did the thermal paste obviously, switched to Thermal Grizzly Kryonaut Thermal Paste which is what I have on the GPU. I've tried loosening the CPU block, tightening. Are these temps normal?

Thanks in advance for any help guys.

No, that's too hot

What wattages are you seeing in HWinfo during that test? If you've maxed out PBO and got it blasting 140W, then yeah you're gunna cook it like that

Mine maxes out in the 90-110W range, and sits around 70C (rarely i see like one single reading hit 80C, when nothing else matches it - i assume theres some minor readings errors at times)

Like this 81C reading doesnt fit any other result, despite all being the same hardware (Current - min - max)
I guess over 46 hours the sensor just had one brief millisecond of heat?

This is what I get with a corsair 150i elite 360mm AIO at -30mv on all cores and stock power limits.

View attachment 268828

Tried everything to get all core 4.4ghz on cinebench, but the max I can do is 4.2ghz. Setting some power limits allows me to get 10c lower temps at 4-4.1ghz.

A friend of mine with basically the same specs, except a sythe fuma 2 air cooler, gets 15c~ lower temps at 300mhz higher clock than me. So my chip is just the worst lottery ever.

you're pushing 125W through it, you need to lower that. You look like you've got boosted PBO wattages with no curve optimiser.


Heres mine with 25C ambients: R23 makes it hotter than literally anything else, but it's not as hot not using as much wattage as what you guys are getting

You likely have some "performance" or overclocking options turned on that are kicking voltages higher than expected like LLC - i'm running really high SoC voltages and RAM at 3866 here, which adds a fair bit of heat (16W on my SoC vs 7.5W on one post above mine)

 

[LordsnLadies]

you're pushing 125W through it, you need to lower that. You look like you've got boosted PBO wattages with no curve optimiser.

Here, some more aggressive power limits. Dropped to 4ghz effective.

 

[tabascosauz]

No, that's too hot

What wattages are you seeing in HWinfo during that test? If you've maxed out PBO and got it blasting 140W, then yeah you're gunna cook it like that

Mine maxes out in the 90-110W range, and sits around 70C (rarely i see like one single reading hit 80C, when nothing else matches it - i assume theres some minor readings errors at times)

Like this 81C reading doesnt fit any other result, despite all being the same hardware (Current - min - max)
I guess over 46 hours the sensor just had one brief millisecond of heat?

you're pushing 125W through it, you need to lower that. You look like you've got boosted PBO wattages with no curve optimiser.

Most boards will push 110-120W through 5800X3D at stock. If you're not at least over the 100W mark bone stock with normal cooling, then your board is markedly more conservative than is typical.

CCD temp sensor is often an outlier usually either noticeably higher than anything else or captures spikes that the others don't, which is why I just keep it hidden. Fan control is tied to Tctl/Tdie anyway, and per-core temps exist.

However.......the L3 temp sensor I suspect is still located in the same place as other Zen 3 (ie. the base 32MB layer), as it behaves the exact same; still, I'm not entirely convinced that stacking some more cache and operating at even lower Vcore is enough to fundamentally change thermal characteristics of Zen 3 uncore, so CCD1 sensor most likely is just doing the same bizarro shit it does on other CPUs - thus no cause for concern.

Here, some more aggressive power limits. Dropped to 4ghz effective.

Thermals not surprising really, pretty much everyone hits 90 out of the box, then once you drop to 90-100W then temps come down drastically.

If you want to bring your clocks back up you will need to mess with curve optimizer undervolting, via PBO2 Tuner. Temps look fine but stock V-F curve is bringing your clocks down with it.

Also, low R23 clocks (~4.2GHz even with -30 all cores) is characteristic of most AGESA 1207 BIOSes. Unsure what board you have but 5800X3D only has a choice of 2 AGESAs, 1206 and 1207. You can try a 1206 BIOS for your board for better clocks.

 

[Mussels]

Clearly that's helped - 90C down to 70C
What you need to do is leave it like that, and test your BIOS settings.

I've got an unlocked BIOS so i've got every setting available even when most don't work, so I don't know what you'll have visible but LLC definitely should be


VID (requested voltage) changes the clock speed, so a straight voltage offest can reduce performance since it alters VID

Anything that raises or lowers actual voltage instead of VID (like the LLC settings and various "performance enhancers" wont - so it could ask for 1.25v and receive 1.20 and run colder, or 1.30v and run hotter

Set LLC to 1, see if its faster MHz at that same 90W PPT

Most boards will push 110-120W through 5800X3D at stock

Not 125W with 3200 RAM, that's on the really high end


Definitely do a CMOS clear and load optimised defaults and set things up again, I had various settings from my 5800x visible and 'saved' until I did this. altering some of the defaults slightly

 

[tabascosauz]

Not 125W with 3200 RAM, that's on the really high end

Package power has nothing to do with memory and Fabric (at least not how you're suggesting). Because stock 5800X3Ds don't max out anywhere close to the 142W limit, stock wattage is almost wholly dependent on board and AGESA. Both my boards are north of 120W stock.

 

[LordsnLadies]

Thermals not surprising really, pretty much everyone hits 90 out of the box, then once you drop to 90-100W then temps come down drastically.

If you want to bring your clocks back up you will need to mess with curve optimizer undervolting, via PBO2 Tuner. Temps look fine but stock V-F curve is bringing your clocks down with it.

Also, low R23 clocks (~4.2GHz even with -30 all cores) is characteristic of most AGESA 1207 BIOSes. Unsure what board you have but 5800X3D only has a choice of 2 AGESAs, 1206 and 1207. You can try a 1206 BIOS for your board for better clocks.

Running an Asus ROG Strix B550-A with ComboV2_1207.

Out of the box my chip hits 3.8ghz on cinebench at 70c, so it power limits before it can even hit higher clocks. I had to undervolt by -30mv to get over 4ghz.

I first thought it was cooling that wasn't adequate but its the chip itself. I was running with a 120mm phanteks and the temps weren't much worse than I have now on an AIO.

 

[tabascosauz]

Running an Asus ROG Strix B550-A with ComboV2_1207.

Out of the box my chip hits 3.8ghz at 70c, so it power limits before it can even hit higher clocks. I had to undervolt by -30mv to get over 4ghz.

I first thought it was cooling that wasn't adequate but its the chip itself. I was running with a 120mm phanteks and the temps weren't much worse than I have now on an AIO.

1207 is the main culprit for the low clocks in R23. It's a very visible difference while only changing AGESA and nothing else. Step back to any available 1206 and see if things improve.

On 4006 (a 1206 BIOS) I'm usually at about 4.35-4.4GHz in R23 at 100W. On 4201 (a 1207 BIOS) I'm down in the 4.1GHz range at same 100W. Same behaviour for A.50/A.71 BIOSes on my Unify-X.

Normally you have to use PBO2 Tuner which resets upon reboot, but you should have access to beta BIOSes here, which unhides PBO/CBS/AMD OC settings and you go about it like a normal CPU:

[Sammelthread] - ASUS B550 Strix/TUF/Prime Series Sammel- und Laberthread | Hardwareluxx
B550-A (mediafire.com)

Unfortunately, I've only found a 2803 BIOS for the Strix-A (which is a 1207 BIOS). I'd use the vanilla 2604 BIOS and PBO2 Tuner for now, until you can find a modded 2604, or someone to do it. Reous released a bunch of 1206/1207 BIOSes for the ROG C8 boards; he might point you towards a 2604 mod or make one if you ask. It looks like he made the 2803 mod, so best to ask, he probably still has it somewhere.

ASUS ROG X570 Crosshair VIII Overclocking & Discussion Thread | Page 631 | Overclock.net

Not all are, and it's definitely a combo of the CPU wattage + the SoC wattage
Some people mix up PPT and combined, which definitely screws with things

90W PPT still has your SoC on top, which could be 95 to 115W depending on the RAM setup and screw with the temps - in this case seeing the low SoC with the 125W total says somethings being pushed too hard

PPT and Package Power are both inclusive of SOC and have always been this way......since the beginning of AM4

His SOC draw is well below 10W and par for the course for 1.0V VSOC. Nearly 120W coming out of just VDDCR_CPU, is quite a bit of power.

Blender is not R23.

 

[Mussels]

Package power has nothing to do with memory and Fabric (at least not how you're suggesting). Because stock 5800X3Ds don't max out anywhere close to the 142W limit, stock wattage is almost wholly dependent on board and AGESA. Both my boards are north of 120W stock.

Not all are, and it's definitely a combo of the CPU wattage + the SoC wattage
Some people mix up PPT and combined, which definitely screws with things

90W PPT still has your SoC on top, which could be 95 to 115W depending on the RAM setup and screw with the temps - in this case seeing the low SoC with the 125W total says somethings being pushed too hard

You gotta do the opposite to a normal board, set the LLC settings to allow the voltage to droop - theres enough headroom to lower the voltages and the overclocking heavy boards tend to push LLC harder by default, making them a worse choice on defaults for these sometimes


Look at w1zzards review of the x3D
CPU only at 89W, he's not seeing 125W or close to it

 

[Calenhad]

As long as you have decent flow rate, the order of the components in your custom loop has very limited impact on temps. Water does not heat up or cool down between them. Temps are uniform throughout the loop. I have traditionally run two temp sensors in my loops, "hot" and "cold" sides, and never once seen a measurable delta between them. Now I use a combined temp/ flow rate sensor instead.

People seriously underestimate how fast the water circulates with decent flow rate. In my current setup I have just under 1 litre of coolant. Based on the measured flow rate, my loop moves about 5 litres per second. There is no time for any hot spots to develop when the entire coolant quantity has been through the loop several times per second.

So, unless the flow rate in your setup is rubbish. You should not worry about the order of things.

 

[LordsnLadies]

@tabascosauz Downgrading to official 2604 bios increased the stock result to 4.1ghz and an extra 100mhz when undervolting/power limiting.

90w ppt, 60A tdc, 80A edc with -30mv on all cores

 

[INSTG8R]

I mean mine runs much “warmer” than my 5600X did under a 360 custom water loop. But I did change pumps to something stronger(x2 RPM) and I may need to look at my rad fans as they don’t seem to pulling as strrong(there were many errors made that led up to this build) That said while definitely running warmer than I’m used to I still find it quite manageable.
My also new MSI board has a setting for the 3D called Kombo Strike that kinda acts like PBO amd Curve Optimzer togethe)obviously no OC just keeps the MT boost maxed better) Thanks to @Super Firm Tofu for that one. It has kept my temps around 80-85 in say an R23 run vs it pushing 91 before

 

[Calenhad]

I mean mine runs much “warmer” than my 5600X did under a 360 custom water loop. But I did change pumps to something stronger(x2 RPM) and I may need to look at my rad fans as they don’t seem to pulling as strrong(there were many errors made that led up to this build) That said while definitely running warmer than I’m used to I still find it quite manageable.

I just want to point out that double the rpm says close to nothing about performance. You could get twice the rpm because the impeller is half the radius and in reality you have lost flow rate and/or pressure as a result.

That being said, the DDC pump in your system specs is about as good as it gets

So I will take my cola bottom lens glasses and leave

 

[INSTG8R]

I just want to point out that double the rpm says close to nothing about performance. You could get twice the rpm because the impeller is half the radius and in reality you have lost flow rate and/or pressure as a result.

That being said, the DDC pump in your system specs is about as good as it gets

So I will take my cola bottom lens glasses and leave

Let’s just say it was replacing some ancient and inferior but kept at least 5 builds always cool and was always reliable up until I “killed it”

So long old friend…

Welp after 15 years of faithful service after doing a loop flush I accidentally pulled the wires out of the Molex. I even tried it in 12 bare wires. No joy… May you rest well old faithful… I’m grabbing a Corsair res/D5 combo just to get something quick hopefully have it back up next weekend...

www.techpowerup.com