Hot Temps for 5800X3D

[sfjuocekr]

My 5800X3D get hotter wih more CO on stuff like games. Warzone is a good example. with -30 CO I get temps north of 70c.

With stock CO my temps are 60c.

Its a BIG misconception that CO helps temps, yeah it can, but in certain workloads the cores will reach higher frequency, thus, higher temperature.

Well yea, it is a perfect example!

During your playtime in the game the frequency will change, these changes cause the DLDO to overshoot more and because it was at a lower voltage with higher amperage now transitioning to a higher voltage higher amperage situation it will generate excess heat because it was allowed to overshoot that far because PPT was probably set high and there was no thermal constraint yet so EDC was still in effect.

That is why I specifically took 115/95/95 as an example to test the CO behavior, from my testing this seems to be the sweetspot where it can always boost to 4450 without reducing frequency under full load regardless of CO value. It will reach "max temp" faster with higher values, it will still reach the same temperature eventually with lower values.

At least from my testing, it seems to be "just" below 80 if I let it rip.

I dont completely understand what are you on about BCLK and multipliers but it seems to be working for me. CPU simply clocks higher with higher BCLK. Here's a SS from last month but I can try to get a new one if you wish. Or any other test really.

.View attachment 331068

That is remarkable, as I have not seen this behavior at all.

If you test between 100 and whatever you set, does the score actually increase?

 

[3x0]

If you test between 100 and whatever you set, does the score actually increase?

Some motherboards will limit the boost behavior/multiplier if you adjust the BCLK, some motherboards will keep the 45.5 boost multiplier. It's a BIOS issue AFAIK.

 

[Zach_01]

About heat desnity...
We all may not using the right scientific terminology but this is what we mean and we understand each other.

Example
You have 2 CPUs

1. Single CCD (5800X)
2. Dual CCD (5900X)

Both CPUs have the same default PPT limit (142W)
You stress both CPU with all core load and they both output the same heat energy (142W) to all directions (which they have contact to).
Lets say that on both CPUs the SoC Die outputs the same amount of heat (15W)
That leaves the rest 127W to come out from CCD/s.

1. SIngle CCD CPU outputs 127W from 80mm² area
2. Dual CCD CPU outputs 127W from 2x80mm² area
Ok on dual the distribution is not equal cause one CCD is doing more work than the second by going higher frequency, but still it could be something like 60/40.

So is it wrong to say that singled CCD CPU has higher heat density?

Better (larger area) distribution of heat means easier (better/higher distribution) transfer to IHS and to the whatever cooler.
From smaller area heat "stalls" because you can only transfer it by that much.
And this leads to higher temperature. Which by physics definition is "created" with "condensed" heat.

 

[A Computer Guy]

My 5800X3D get hotter wih more CO on stuff like games. Warzone is a good example. with -30 CO I get temps north of 70c.

With stock CO my temps are 60c.

Its a BIG misconception that CO helps temps, yeah it can, but in certain workloads the cores will reach higher frequency, thus, higher temperature.

I always thought CO was intended to be used with PBO. For example lowering EDC a bit is known to reduce heat then using CO so you can still hit high boost clocks. This is probably a bit apples and oranges but for example on my 5950x I can simply only reduce EDC (to 120) + CO and my boost clocks hit harder than at stock and the heat is generally lower for just a tad over the same performance as stock.

 

[adilazimdegilx]

Well yea, it is a perfect example!

During your playtime in the game the frequency will change, these changes cause the DLDO to overshoot more and because it was at a lower voltage with higher amperage now transitioning to a higher voltage higher amperage situation it will generate excess heat because it was allowed to overshoot that far because PPT was probably set high and there was no thermal constraint yet so EDC was still in effect.

That is why I specifically took 115/95/95 as an example to test the CO behavior, from my testing this seems to be the sweetspot where it can always boost to 4450 without reducing frequency under full load regardless of CO value. It will reach "max temp" faster with higher values, it will still reach the same temperature eventually with lower values.

At least from my testing, it seems to be "just" below 80 if I let it rip.


That is remarkable, as I have not seen this behavior at all.

If you test between 100 and whatever you set, does the score actually increase?

I ran it couple times to get the highest because 103BCLK was also my highest overall. This is what I get with -30 CO @100BCLK (also tested on about 19C ambient, cuz other one was also low ambient temperature)

 

[Zach_01]

This is what I get with -30 CO @100BCLK (also tested on about 19C ambient, cuz other one was also low ambient temperature)

I bet if you run it with CO 0 the effective speeds are dropping, max speeds stays the same. And score is lower...
And showing avg values help people to understant better this

 

[sfjuocekr]

I ran it couple times to get the highest because 103BCLK was also my highest overall. This is what I get with -30 CO @100BCLK (also tested on about 19C ambient, cuz other one was also low ambient temperature)

View attachment 331080

Only 400 points difference, that is the difference between having a browser open or not.

But it could very well be that on Windows the frequency scaler works differently, because on my system (Linux) it is controlled by AMD P-states and it will never go above 4550 single and with 4450 all core.

 

[adilazimdegilx]

I bet if you run it with CO 0 the effective speeds are dropping, max speeds stays the same. And score is lower...
And showing avg values help people to understant better this

This is with -0 CO. Basically stock 5800x3D. And yes, effective speeds drop quite a lot. Lowest I got was 14455 so about 50 point variation occurs.

 

[sfjuocekr]

About heat desnity...
We all may not using the right scientific terminology but this is what we mean and we understand each other.

Example
You have 2 CPUs

1. Single CCD (5800X)
2. Dual CCD (5900X)

Both CPUs have the same default PPT limit (142W)
You stress both CPU with all core load and they both output the same heat energy (142W) to all directions (which they have contact to).
Lets say that on both CPUs the SoC Die outputs the same amount of heat (15W)
That leaves the rest 127W to come out from CCD/s.

1. SIngle CCD CPU outputs 127W from 80mm² area
2. Dual CCD CPU outputs 127W from 2x80mm² area
Ok on dual the distribution is not equal cause one CCD is doing more work than the second by going higher frequency, but still it could be something like 60/40.

So is it wrong to say that singled CCD CPU has higher heat density?

Better (larger area) distribution of heat means easier (better/higher distribution) transfer to IHS and to the whatever cooler.
From smaller area heat "stalls" because you can only transfer it by that much.
And this leads to higher temperature. Which by physics definition is "created" with "condensed" heat.

If we are talking scientific terminology, there is no such thing as heat density. There are:

Thermal mass;
Thermal conductivity;
Thermal resistivity.

If you have two dies, you are simply increasing surface area not increasing density.

 

[Zach_01]

If we are talking scientific terminology, there is no such thing as heat density. There are:

No we are not... I never said that, and no one else did.
Thia is what I said...
"We all may not using the right scientific terminology but this is what we mean and we understand each other..."

Did you even try to understand what I wrote?

2 dies are dropping "heat density" actually, because the same amount of heat is "produced" and distributed from more material, from more area.
No one said or thinks differently in here.
Double CCD CPUs with same PPT as 1CCD ones are easier to cool for a reason.
The same amount of heat distributed to more material and area results to lower temperature. This is physics actually

How can you not understand what we are saying with heat density even if its not a scientific term.

This is with -0 CO. Basically stock 5800x3D. And yes, effective speeds drop quite a lot. Lowest I got was 14455 so about 50 point variation occurs.

Excellent example of CO, thank you!
On 1st (-30) you have max Vcore(SVI2TFN) 1.20v, max speed 4.45GHz, max effective speed 4.44GHz, max temp of 72C and higher score
On 2nd (0) you have max Vcore(SVI2TFN) 1.28v, max speed 4.45GHz, max effective speed 4.22GHz, max temp of 75C and lower score
Both have about the same CPU power consumption (PPT)

This looks like undervoltage to me that helps cores to sustain highest boost (4.45GHz) longer within the same power envelope.
Effective speed is always the key factor.

So on CPUs with capped speed like the 3Ds undervoltage really works through CO.
On non3D CPUs that their max boost clock is not reached by all core clocks all you do with CO is making them work higher speed with lower voltage than the one that would've need to go there. Even go higher power if they have the headroom. Thermally too...

Eventually it is an optimizer after all. A V/F curve optimizer.
Better score (+7%) with better operating conditions.

 

[freeagent]

You dont need to overclock the CPU to hit 15K+..?

But as soon as you start to lower PPT/TDC/EDC you lower overall CPU performance. Plus you are using a negative curve on top of that?

My R23, nothing fancy at all, and not my best score with this CPU..

145w CPU package power, 142w PPT from Linpack Xtreme:

 

[AusWolf]

I've had air coolers (dark rock slim) out perform custom water loops. They're fussy with how coolers are mounted, rather than needing bigger coolers.


I sold that dark rock slim, and i'm still sad about it.

Truth!

My 7800X3D throttles with a 280 mm be quiet! Silent Loop 2, but has about 10 °C headroom with a Dark Rock 4 that costs less than half of the AIO.

 

[sfjuocekr]

My Fuma 2 also performs slightly better under full load, I just can not stand the howl of the fans to achieve that!

No we are not... I never said that, and no one else did.
Thia is what I said...
"We all may not using the right scientific terminology but this is what we mean and we understand each other..."

Did you even try to understand what I wrote?

2 dies are dropping "heat density" actually, because the same amount of heat is "produced" and distributed from more material, from more area.
No one said or thinks differently in here.
Double CCD CPUs with same PPT as 1CCD ones are easier to cool for a reason.
The same amount of heat distributed to more material and area results to lower temperature. This is physics actually

How can you not understand what we are saying with heat density even if its not a scientific term.


Excellent example of CO, thank you!
On 1st (-30) you have max Vcore(SVI2TFN) 1.20v, max speed 4.45GHz, max effective speed 4.44GHz, max temp of 72C and higher score
On 2nd (0) you have max Vcore(SVI2TFN) 1.28v, max speed 4.45GHz, max effective speed 4.22GHz, max temp of 75C and lower score
Both have about the same CPU power consumption (PPT)

This looks like undervoltage to me that helps cores to sustain highest boost (4.45GHz) longer within the same power envelope.
Effective speed is always the key factor.

So on CPUs with capped speed like the 3Ds undervoltage really works through CO.
On non3D CPUs that their max boost clock is not reached by all core clocks all you do with CO is making them work higher speed with lower voltage than the one that would've need to go there. Even go higher power if they have the headroom. Thermally too...

Eventually it is an optimizer after all. A V/F curve optimizer.
Better score (+7%) with better operating conditions.

It is not an optimizer, it is a relatively simple regulator that they put a "fancy" name on.

What I don't understand is why someone would look at "effective" clock speeds, the max clock you can get on a single core load is generally always shown and you should just look at your current clock speed with a load applied.

There is no reason for the CPU to clock down to 4.22GHz when it is below 75 degrees. So there is something else limiting it from boosting or the reported values are wrong.

 

[freeagent]

The values are right, he was manipulating his power limits.

 

[A Computer Guy]

It is not an optimizer, it is a relatively simple regulator that they put a "fancy" name on.

Using CO you are optimizing your power usage per core along the voltage/frequency curve are you not? If your cores don't require as much power (compared to the default settings) and you limit your total power (via PBO) you optimize your CPU to favor a new balance in power, thermals, and performance.

 

[freeagent]

For Zen 3, on anything other than X3D, more power, plus negative curve= higher boost. On Zen 3 X3D if you want the boost, your only option is via CO. On Zen 3 X3D if you lower power limits, it is easy to see performance taper off accordingly in something like Linpack Xtreme. So, imo if you are turning power down to decrease temps, you have an actual cooling issue, be it the cooler isn't up to snuff, or maybe something else.

 

[Zach_01]

My Fuma 2 also performs slightly better under full load, I just can not stand the howl of the fans to achieve that!


It is not an optimizer, it is a relatively simple regulator that they put a "fancy" name on.

What I don't understand is why someone would look at "effective" clock speeds, the max clock you can get on a single core load is generally always shown and you should just look at your current clock speed with a load applied.

There is no reason for the CPU to clock down to 4.22GHz when it is below 75 degrees. So there is something else limiting it from boosting or the reported values are wrong.

I thought this is common knowledge by now
From 3000series onwards (chiplet design) Ryzens have become even more high dynamic chips and a simple core clock speed from traditional software cannot tell the truth about the core behavior.
They can monitor core parameters as speed, voltage, current x1000/sec and alter those x50/sec (every 20ms)

You may see a "steady" speed of 4.5GHz under full load and under traditional core clock for example but in reality speed may vary 200~300MHz or more as self internal management monitors other aspects as voltage, current and thermals and adjusts accordingly. Even HWiNFO set on 1000ms polling interval cannot represent the full truth. I don't have the knowledge of how exactly effective speeds are calculated but I know for sure it has to do with core states (C0/C1/C6) also.
Also, readings are more inaccurate for Ryzens if "Snapshot CPU Polling" is disabled from HWiNFO main settings

For more information about it you can ask the author (Martin) of HWiNFO at software's forums

 

[Mussels]

Just going to debunk this "CO" being an undervolt, it is not.

It's an offset, it's undervolting vcore while keeping VID the same.

Since the CPU bases its clocks on reported VID and VID can now be higher when PBO limited (less volts = more amps) it's an undervolt. It's not as easy to see as it would be with a CPU that had no wattage limits, but we cant disable the 142W limit.

For Zen 3, on anything other than X3D, more power, plus negative curve= higher boost. On Zen 3 X3D if you want the boost, your only option is via CO. On Zen 3 X3D if you lower power limits, it is easy to see performance taper off accordingly in something like Linpack Xtreme. So, imo if you are turning power down to decrease temps, you have an actual cooling issue, be it the cooler isn't up to snuff, or maybe something else.

Depends on the board - mine also has a working voltage offset, I can use CO and the offset together for even lower voltages and a flat 4.45GHz all core no matter the load.

Too low on the Vcore and the CPU becomes aware Vcore and VID don't match up and the clocks start to lower, so CO comes first and then the VCore offset needs to be dialed in to the edge of that trigger limit, so no performance is lost.

 

[RJARRRPCGP]

If my 5800x is anything to go by, IF/RAM stability tanks if the CPU passes 75c - that took ages to figure out (in the end I confirmed a 75c throttle in the BIOS solved it and slower cooling speeds made it crash faster, finally figuring it out)

Sounds like something my 3700X would do, it seemed more likely to have a fatal machine check exception with "Cache Hierarchy Error" as reason, if the room was merely at or around 75 F.

 

[Chicken Patty]

damn, I see this thread is still going lol. I have an AM5 rig now, but my 5800X3D/3090Ti rig is now running in my living room as a media PC and occasional gaming. 5800X3D is still a toasty chip but much better than when I made this thread. I ended up reapplying thermal paste and switching thermal pastes and it did much better.