Hot Temps for 5800X3D

[tabascosauz]

@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

65W limits are 88/60/90. I keep EDC at 90 and just adjust PPT upwards as necessary, though you can decide for yourself based on % utilization of PPT/TDC/EDC reported by HWInfo.

How hot is your room ambient? If it's not hotter than 30C, everything actually looks a little on the warm side, by 5-10C. I would have suggested remounting, but noticing IO die temps and cache temps ambient just seems quite high. If above 30C, temps look a lot more reasonable.

Otherwise, things look fine, except that Vcore is slightly on the low side. It's good news that -30 works but that usually lands you around the 1.15-1.2V mark. You can play around with Vcore (one of the options in the dropdown is a + or - offset), see if your clocks improve from running say +0.025V.

Also, does -30 actually give a regression in clocks compared to -20 or -25. You're clock stretching a fair bit (4.125 effective vs 4.15 clock). Shouldn't be more than a 10MHz difference

 

[Upgrayedd]

Isn't there a Ryzen offset mounting option?

 

[3x0]

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.

Sorry for off topic question, but does 1206 have MSI Kombo strike or is it only in 1207?

 

[tabascosauz]

Sorry for off topic question, but does 1206 have MSI Kombo strike or is it only in 1207?

Unify-X or other boards? Unify-X only has Kombo Strike in A.70. A.50 has a PBO menu that MSI forgot to hide though (no CO). Can't say for other boards, BIOSes are a mess. 1206 was X3D launch BIOS, MSI didn't release Kombo Strike before 1207.

 

[Mussels]

@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
View attachment 268853

On my 5800x, certain BIOSes wouldnt work with the curve undervolt - that same BIOS worked with the CU on the x3D, maybe you ran into similar

 

[LordsnLadies]

How hot is your room ambient? If it's not hotter than 30C, everything actually looks a little on the warm side, by 5-10C. I would have suggested remounting, but noticing IO die temps and cache temps ambient just seems quite high. If above 30C, temps look a lot more reasonable.

Room ambient was 25-30 when I ran those previous tests. I've tried repasting and remounting but it didn't make much difference. I think my chip is just bad silicon lottery at this point, because my friend has the same mobo, cpu and memory except he is running a sythe fuma 2 and he gets 200-300mhz higher at the same power/voltage limits while using agesa 1207. We did some more tests yesterday and our temps are actually a lot closer than "10-15c" figure I was saying in my previous post, but in general his runs colder than mine.

For the tests below I would say the ambient temp is probably about 21-25c.

88/60/90 @ -20mv

88/60/90 @ -30mv

Stock limits @ -30mv

Otherwise, things look fine, except that Vcore is slightly on the low side. It's good news that -30 works but that usually lands you around the 1.15-1.2V mark. You can play around with Vcore (one of the options in the dropdown is a + or - offset), see if your clocks improve from running say +0.025V.

I'll give that a try.

 

[Mussels]

You dont have your system specs filled out so i've already lost what you're running, but your cooler, mounting pressure/method and how you applied thermal paste are kinda critical here

First image:
85C for a 4.1GHz effective clock isn't great. It's not end of the world, but it's a 10% performance loss and still too hot

second run:
4150MHz at 70C? heaps improved

third run:
4300 at 85C, best of the lot clearly performance wise, except for the temp

A few wattages are cut off there, but looks to be around 115W PPT - you can either limit that slightly lower (110/105) to drop the temps, or you can work on the cooling and keep it there.

 

[LordsnLadies]

You dont have your system specs filled out so i've already lost what you're running, but your cooler, mounting pressure/method and how you applied thermal paste are kinda critical here

Sorry, didn't realize I could fill out my system specs in my profile, I have now updated my profile with the specs.

Just to recap, I was running a Phanteks PH-TD12DX 120mm air cooler initially and then switched to the corsair h150i rgb elite. All results posted here have been with the corsair though.

With the phanteks I used the X method to paste it and repaste it the second time. Saw the same clock and similar temps compared to the above results.

With the corsair AIO I initially used preapplied paste and when I repasted I used the X method. There weren't any differences in thermals.

As for mounting pressure, I tightened both coolers as much as the screws would allow.


I asked my friend to limit his clock speed to mine and use stock settings without undervolts or limits. He has identical specs except is running a sythe fuma 2 and a midtower.

Friends result:

My result:

 

[tabascosauz]

@LordsnLadies are you sure the 88/60/90 -30 result actually applied -30 in PBO2 Tuner? 1.26V is kinda high for -30. Usually at -30 all core you'll peak around 1.2V max. 1.25V is stock Vcore basically.

How was temps on the Phanteks 120mm tower?

You are still pulling more power than your friend at stock, so even considering sample variations it's not apples to apples. Over 100W, you start to really feel every 10W increase, so 110W vs. 125W is not negligible.

TC12DX is not quite on the same level as Fuma 2 (more U12S). You don't need much paste for X3D, it's a 1CCD part, larger rice grain works fine. How does your cooler contact look, good spread on IHS?

 

[HD64G]

Curve optimizer test and applying a bit less optimistic values than the ones it will show (i.e. -20 instead of -23) in Ryzen Master would be more than enough if power limit was at 105-110W to get >95% of max performance and -10-15C of temp decrease.

 

[Mussels]

Yeah i'm not sure your mounting method is correct
X method isn't ideal for this sort of CPU either, you want either the blob or spread methods - personally i do both, spread some paste and then a small extra central blob

You need to be finger tightening those screws in opposite orders diagonally so top right, bottom left, top left, bottom right 1-2 turns at a time


Behold, my MS paint skills!
The CPU die is in the red box, that is where you need to be cooling. Not the center. Get the paste wrong or tighten in the wrong order and you can easily push the TIM everywhere except where it's needed

You also need to make sure you've got the pump at 100% speed while you're testing, and show us images of your fan and radiator setup - it's way more common than people think to put rads in bad spots, or have fans facing the wrong way... or just have a terrible case.
Hell make sure it's not clogged with dust and animal hair too, it can happen really fast in some environments.

The Filthy, Rotten, Nasty, Helpdesk-Nightmare picture clubhouse

Friends PC had issues. Friends PC ate the whole dog. The compacted hair springloaded the front panel so it shot off onto the floor

www.techpowerup.com

 

[LordsnLadies]

are you sure the 88/60/90 -30 result actually applied -30 in PBO2 Tuner? 1.26V is kinda high for -30. Usually at -30 all core you'll peak around 1.2V max. 1.25V is stock Vcore basically.

How was temps on the Phanteks 120mm tower?

Its certainly doing something because I notice changes if I decrease down to 20mv.

When i was running the phanteks I was looking at the ryzen master reported temp which shows the hottest temp and that was identical to the temp i get now on the corsair AIO. So I can't really say if the cores themselves run differently now.

How does your cooler contact look, good spread on IHS?

You also need to make sure you've got the pump at 100% speed while you're testing, and show us images of your fan and radiator setup - it's way more common than people think to put rads in bad spots, or have fans facing the wrong way... or just have a terrible case.

My setup

Also great job on those paint skills Mussels.

I've contacted my supplier/retailer to RMA the chip because I'm convinced its just a bad sample, but thanks for all the insight into this.

Edit:

90/60/90 @ -30mv with pump/fans balanced profile

90/60/90 @ -30mv with pump/fans extreme profile

 

[GerKNG]

The 5600X with PBO and custom power with curve can do in excess of 130w PPT, and over 150w PPT with a static clock.

in what? Prime95?
i have a 5600x and even at 4.8 Ghz all core i barely reach more than 105-110W in R23 or Blender.

 

[freeagent]

in what? Prime95?
i have a 5600x and even at 4.8 Ghz all core i barely reach more than 105-110W in R23 or Blender.

I can do it with Linpack, I posted a few screens but they are probably lost in the mix

 

[Mussels]

On the images:

Cable management isn't great, but nothing that should alter performance i can see
Fans all look oriented correctly, and that thermal paste looks 99% of the way to perfect - you do want a tiny bit more than that so it goes out over the edges and makes an airtight seal, but its a flat contact which is critical

IMO you should move your SSD's down to the lower bay and remove this entire HDD caddy - its blocking airflow for no reason

SSD's dont vibrate so you could simply use double sided tape (i prefer command strips like these) and mount the SSD's to the floor of the case and remove both cages

These are 'reusable' in that you can add/remove the drive many times, but not reusable in that you could peel it from the drive/case and move it around

This kind of reusable nylon cable tie is dirt cheap on amazon and can be cut to shrink them down (halves work great, thirds for really small wires in tight areas) to tidy up cables in easy to adjust ways

Any VRM overheating will be here:
With one temp sensor over two areas, it's possible for one side to overheat and the one with the sensor be cold. You should be able to hold a finger on that area (and the PCB right next to it) for an extended time without it getting too hot to touch at the temps we've seen reported
If the PCB is burning hot and the heatsink is cold, we know you need new thermal pads there

Just to be clear: You're on AGESA 1.2.0.7 (latest BIOS), you've done the "load optimised defaults" and you have no settings altered regarding CPU voltages (offsets) other than the PBO settings and curve optimiser?


Your "90/60/90 @ -30mv with pump/fans extreme profile" looks pretty much perfect, you could add 10-20 to each value and probably reach the full 4.45GHz without overheating, and then work out what fan/pump settings you need


It's also important to note that just because its an AIO or from a big brand, doesnt mean anything about its performance

As noted here with a ryzen 3600x, its chart topping on those extreme pump and fan settings but on quiet is beaten by a mere 120mm noctua air cooler
Corsair iCUE H150i Elite Capellix CPU Cooler Review | TweakTown

With overclocked settings (and the 3600x is still low wattage in this situation) quiet mode became almost worthless, higher wattage CPU's would have issues like you have been
Balanced does okay, but it's matching air coolers

 

[tabascosauz]

Its certainly doing something because I notice changes if I decrease down to 20mv.

When i was running the phanteks I was looking at the ryzen master reported temp which shows the hottest temp and that was identical to the temp i get now on the corsair AIO. So I can't really say if the cores themselves run differently now.

I've contacted my supplier/retailer to RMA the chip because I'm convinced its just a bad sample, but thanks for all the insight into this.

I was going to say, if you were handling the RMA yourself, just overload the AMD rep with your troubleshooting data, and say you've already tried everything on his canned response troubleshooting list. Usually persuades them to skip the bullshit and just get the RMA done (which itself is quick at least in NA). But if someone else is handling it, great.

Tbh if the paste is covering your entire IHS, you don't see any areas with noticeably poor contact, and you're using a good enough paste, paste application on AM4 is hard to fuck up to the point of affecting your temps beyond margin of error. At one point with my 5900X, I was very picky about the 3 dot method for best coverage on 2CCD - it's not worth fussing over if you have even the faintest idea of what you're doing, at least not for ambient conditions.

Your friend's X3D looks about right in all respects, at least - yours is both slightly warmer than expected and still clocking a lot lower than expected. With half-respectable cooling at -30, there's no reason to ever be sitting below 4.2GHz, tbh. Even on AGESA 1207 with its clock penalty in R23 (which isn't the case as you're on 1206 now).

Just to confirm, BIOS 2604 yeah?

 

[Mussels]

Tbh if the paste is covering your entire IHS, you don't see any areas with noticeably poor contact

I'm going to 100% agree with this except that over long periods of time (which clearly has not happened yet) it'll dry out faster

I'm beginning to suspect it's primarily the corsair cooler, if OP has been running in quiet mode

 

[LordsnLadies]

Just to be clear: You're on AGESA 1.2.0.7 (latest BIOS), you've done the "load optimised defaults" and you have no settings altered regarding CPU voltages (offsets) other than the PBO settings and curve optimiser?

Just to confirm, BIOS 2604 yeah?

Yeah this is running 2604 with AGESA 1.2.0.6b. It gets 100mhz less at the same undervolt on AGESA 1.2.0.7.

I run DOCP profile for my memory and resizable bar, and I added some PBO limits so that I don't have to keep reentering them. I tried upping the voltage by 0.025v on the VDDCR through the bios last night but it didn't produce any noticeable in results.

Your "90/60/90 @ -30mv with pump/fans extreme profile" looks pretty much perfect, you could add 10-20 to each value and probably reach the full 4.45GHz without overheating, and then work out what fan/pump settings you need

The highest clocks i can reach on cinebench r23 is 4.3ghz with no limits and -30mv undervolt on agesa 1.2.0.6.

No limits @ -30mv

 

[Mussels]

Yeah it's the cooler, you're thermal throttling

It's not maxing out any those values, but the thermal limit reduces the boost before a 'true' throttle

 

[tabascosauz]

Yeah this is running 2604 with AGESA 1.2.0.6b. It gets 100mhz less at the same undervolt on AGESA 1.2.0.7.

I run DOCP profile for my memory and resizable bar, and I added some PBO limits so that I don't have to keep reentering them. I tried upping the voltage by 0.025v on the VDDCR through the bios last night but it didn't produce any noticeable in results.

Added some PBO limits.....are you running a modded BIOS, or is this just something Asus forgot to hide in the BIOS?

RMA might be the right call. There's a few things that I can't explain from your screenshots:

• We have similar ambient temps, but I finish R23 with almost the same temps (71C). Similar performance between the C14S and TC12DX I can understand, but not to the AIO with max speeds.
• Vcore of 1.07V is kinda low for -30? On 1206, I clock about 4.35GHz at 1.10-1.12V. Granted, every CPU is different and CO does not define fixed steps in terms of mV, but still......
• Clocks are just very low for -30 on a 1206 BIOS. Thermals should not be the cause for that.
• Major differences for SOC power draw at 1.0V between two CPUs on the same board.......would have expected the difference to come from a different board not CPU.

The only things left to suggest are to try a larger positive offset (+0.05V), and see if clocks improve with a slightly less aggressive undervolt (-20, -25).

The clock stretching is less of a concern, it seems Core and Effective just don't line up very well on 5800X3D.

If anything, -30 with stock limits looks a lot more normal. If R23 scores are between 14000-15000, that's probably close to normal.

Yeah it's the cooler, you're thermal throttling

It's not maxing out any those values, but the thermal limit reduces the boost before a 'true' throttle
View attachment 269377

That ain't it. At 88W and 71C, his numbers look normal (75-80% thermal limit).

The stock runs going up to 90C are understandably throttling, stock thermal limit is 90C and it happens to everyone. Well, except the friend's 5800X3D, apparently, who apparently has the coldest running sample in existence.

 

[Mussels]

Tabasco: he's throttling, its downclocking

You're seeing the voltages for the effective clocks

I'm looking at the most recent image where thermal limit is at 95.8%, it's going to downclock to prevent reaching that - the prior images it was hitting the PBO limits instead


In all the images, it's one of two things:
1. The PBO limits are reached and its cold and fine, but lower than expected performance

2. the PBO limits are not reached, and thermals are 85C+ in the downclocking range

 

[tabascosauz]

In all the images, it's one of two things:
1. The PBO limits are reached and its cold and fine, but lower than expected performance

2. the PBO limits are not reached, and thermals are 85C+ in the downclocking range

We don't disagree.......

The clocks (4.3GHz) look the most normal at stock power limits than lower limits. The early thermal throttling is not that drastic on X3D. Somewhere between 70 and 80C (75C?) it starts dialing back the clocks slightly. Even with it, the 142W -30 results don't look that off.

Vcore is Vcore. 1.06-1.07V is just plain low.

Something does seem off with the AIO performance-wise if it can't even beat a TC12DX, but the point is that clocks simply shouldn't behave this way, the TC12DX produces the same results. On 1206, at 70C, 90W and -30.

 

[LordsnLadies]

Added some PBO limits.....are you running a modded BIOS, or is this just something Asus forgot to hide in the BIOS?

Official bios, its a good thing they did leave it in.

Vcore of 1.07V is kinda low for -30? On 1206, I clock about 4.35GHz at 1.10-1.12V. Granted, every CPU is different and CO does not define fixed steps in terms of mV, but still......

Looks like mine needs 1.16v~ and 75-80tdc to hit 4.29-4.3ghz effective on agesa 1206.

The only things left to suggest are to try a larger positive offset (+0.05V), and see if clocks improve with a slightly less aggressive undervolt (-20, -25).

No limits @ -20mv and +0.05 vcore

I ran a couple tests at different vcore offsets (0.18, 0.25, 0.31) and the conclusion is that it just decreases clocks because of the extra heat.

In all the images, it's one of two things:
1. The PBO limits are reached and its cold and fine, but lower than expected performance

2. the PBO limits are not reached, and thermals are 85C+ in the downclocking range

Yeah it is a thermal limit, but I don't think its my cooler. Aside from just being a terrible sample, my theory is that the TIM isn't making proper contact so after a certain point it doesn't matter what cooling you have, it simply won't improve. But without testing another cooler I can't really prove that.

 

[tabascosauz]

Official bios, its a good thing they did leave it in.


Looks like mine needs 1.16v~ and 75-80tdc to hit 4.29-4.3ghz effective on agesa 1206.


No limits @ -20mv and +0.05 vcore
View attachment 269415

I ran a couple tests at different vcore offsets (0.18, 0.25, 0.31) and the conclusion is that it just decreases clocks because of the extra heat.

There's not much point adding voltage at stock limits. Your 142W -30 clocks already weren't that bad, and here you're just thermally limited so more Vcore just works against you.

What I was curious about is whether +0.05 helps at 90W, where Vcore just looks low. Or, if your clocks improve when going from say 90/60/90 to 120/65/90.

+0.18 and more is just too much. +0.31 is like, 1.45V. For safety Vcache has a limit of like 1.35V theoretical and 1.3V practical.

What's your mobo temp when you wake up in the morning and start up?

Also, I've been meaning to ask, 11GB of RAM use while doing nothing is a LOT are you sure nothing's running in the background?

 

[LordsnLadies]

What I was curious about is whether +0.05 helps at 90W, where Vcore just looks low. Or, if your clocks improve when going from say 90/60/90 to 120/65/90.

90/60/90 @ -30mv and +0.05vcore

So it didn't really help, only made things worse. Without the vcore offset it can maintain 4.1-4.125ghz.

What's your mobo temp when you wake up in the morning and start up?

I'll check tomorrow. I would imagine like 30-35c, atm its sitting at 38c.

Also, I've been meaning to ask, 11GB of RAM use while doing nothing is a LOT are you sure nothing's running in the background?

Windows 11 uses a lot of memory. Just idling with nothing running it uses 8-9gb. But yes, I have a few apps like steam or spotify in the background.

I did try testing on a windows 10 bootable OS, and my cinebench scores on there were like 1000 points higher at the same clock speeds. Just wish it improved thermal/clock behaviour too.