• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

Big temp diff between boards

Joined
Oct 10, 2019
Messages
56 (0.03/day)
Location
Malaysia
System Name LongWhite
Processor AMD Ryzen7 5800X3D
Motherboard MSI B550M Mortar
Cooling Scythe Fuma 2 with assorted fans
Memory 4x8GB G.Skill Trident Z Neo
Video Card(s) MSI RTX3070 Ventus 3x
Storage 2x XPG SX8200 1TB
Display(s) LG 34UC79G
Case Silverstone ALTA G1M
Audio Device(s) Behringer UMC204HD
Power Supply Corsair SF750
Mouse GPX Superlight /Pulsar Xlite
Keyboard GMK67 /Keychron V1
I had my 5800x3d on a Aorus x570 Pro Wifi inside a Fractal Torrent, underclock to -30 offset and idling on 45-ish celcius,

while the same CPU on a ASUS Strix B550i inside a Meshlicious, same underclock and it easily idles around 35 celcius.

Both using the same cooler (Galahad 240) intake. I know it's not apples to apples here, but the Torrent has a total of 9 fans and I thought it would not be this bad. What could be the reason for the huge temperature difference? The chipset alone will have that big a difference? (Just hoping to learn something here..)

Fyi I have not tried putting the b550 board into Torrent to test.. just too lazy :D
 
When idle most of the heat on Zen 2/3 comes from the I/O die (SOC). Did you change your IF/RAM settings?
 
When idle most of the heat on Zen 2/3 comes from the I/O die (SOC). Did you change your IF/RAM settings?
None that I'm aware of. I had 4x8GB G Skill Trident Z on Xmp 3600 on the Aorus, down to 2 in the Itx board.
 
None that I'm aware of. I had 4x8GB G Skill Trident Z on Xmp 3600 on the Aorus, down to 2 in the Itx board.

That's a rather open-ended question knowing how Ryzens idle

Pretty much as a rule across my last 6 Ryzen CPUs, Gigabyte boards push stock Vcore harder than other boards. Generally results in slightly better clocks benchmark scores, at the cost of temps and power.

As for idle, different story. But even if different boards apply the same default VSOC for 3600 XMP, SOC LLC can vary quite a bit resulting in different VSOC numbers, not to mention Asus and GB sometimes default to different values for CLDO_VDDGs and VDDP. Most LLC levels on my B550-I Strix seemed to be on the weaker side compared to Gigabyte ITX.

Ryzen idle is also extremely temperamental (especially if reading the default "CPU temp" in most software, which is Tctl/Tdie which acts basically as a chip-wide hotspot sensor). Hard to get accurate apples-to-apples comparison unless you use HWInfo on a fast polling interval, promising Windows/any other program is doing precisely nothing in the background, making sure Package Power remains reasonably similar, and nothing else is amiss.
 
Last edited:
The hotter X570 chipset and RAM will have an effect on internal case temperature (what might be reflected in the System or Motherboard temperature sensor reading) which ultimately impacts how warm the CPU will idle, however it's unlikely that the chipset and RAM generate enough heat for a 10 °C case temperature increase unless you have exceptionally poor airflow through a smaller case (which you do not).

I have a waterblocked 5900X (custom cooling loop) that idles at 33 °C, the same temperature as the ROG Strix X570-I motherboard at idle. This is the build described in my System Specs. The waterblocked graphics card idles at 31 °C. Unsurprisingly, the cooling loop's thermal sensor which measures coolant temperature of the graphics card's exhaust is also 33 °C so everything within 2 °C of each other at idle. Despite the hotter X570 chipset, the board temperature is fine with adequate case ventilation; the nine fans in this build run about 400 rpm (average) at idle, a light breeze not a gale.

My suspicion is the two Galahad AIOs. There is probably some manufacturing variance in units and quite possibly you did a better job mounting the one in the Meshlicious case (better thermal paste application, better tightening of the waterblock head to the CPU mounting bracket). The best way to check if this is the case is to swap the two AIOs between cases, but it doesn't sound like you'd be up to that.

You could try repasting and remounting the Galahad in the Torrent case to see if the temperatures improve.

You also did not mention whether or not the two AIOs' fan curves are set identically in UEFI on the two motherboards so that's another thing to check (radiator fan RPM at idle).
 
Last edited:
It could be the default voltages the boards run at. Gigabyte boards are known to pump power. Asus are more reserved in their stock settings for voltages.
 
It could be the default voltages the boards run at. Gigabyte boards are known to pump power. Asus are more reserved in their stock settings for voltages.

That's a thought, I suppose the easiest way to check this would be to run HWiNFO on both systems at idle and compare readings.
 
That's a thought, I suppose the easiest way to check this would be to run HWiNFO on both systems at idle and compare readings.

The closest you can get is either through the Vcore side of the PWM controller's sensors (which on the Strix aren't exactly reliable or accurate due to ASP1106), a separate die-sense Vcore circuit (not available on Asus below [real] ROG), or eyeballing it in Ryzen Master (probably most reliable).

SVI2 TFN Vcore in HWInfo/any software not Ryzen Master does not properly display idle Vcore. It will show 1.4V+ at idle.
 
The closest you can get is either through the Vcore side of the PWM controller's sensors (which on the Strix aren't exactly reliable or accurate due to ASP1106), a separate die-sense Vcore circuit (not available on Asus below [real] ROG), or eyeballing it in Ryzen Master (probably most reliable).

SVI2 TFN Vcore in HWInfo/any software not Ryzen Master does not properly display idle Vcore. It will show 1.4V+ at idle.

Okay, so Ryzen Master not HWiNFO for checking CPU voltages at idle with the caveat that the readings still might not be terribly exact.

He can mostly learn whether or not the motherboards are pumping out the same voltage at idle with this.

Would a voltage difference be reflected in PPT or CPU Package Power at idle if I were looking at HWiNFO? Or do people need to be looking at Ryzen Master again for this.

I'm not inclined to run five different monitoring apps. Sure, I'll give up a little accuracy but I have other things I need to be doing on my PCs rather than staring at sensor readings.
 
Okay, so Ryzen Master not HWiNFO for checking CPU voltages at idle with the caveat that the readings still might not be terribly exact.

He can mostly learn whether or not the motherboards are pumping out the same voltage at idle with this.

Would a voltage difference be reflected in PPT or CPU Package Power at idle if I were looking at HWiNFO? Or do people need to be looking at Ryzen Master again for this.

I'm not inclined to run five different monitoring apps. Sure, I'll give up a little accuracy but I have other things I need to be doing on my PCs rather than staring at sensor readings.

RM is the most exact. It's just harder to interpret because there's no min/max column or logging like HWInfo, only observing with Eyeball Mk.1. SVI2 Vcore in HWInfo isn't slightly less accurate, it's completely incapable of showing idle Vcore. SVI2 vs die sense at the same moment:

svi2 vcore.png

die sense vcore.png


Regardless, VSOC and the CLDOs should matter more since properly idling Ryzen draws a lot more power from VSOC than Vcore since Vcore should dip down to 0.2V/0.9V and basically negligible power draw. The whole Gigabyte thing is more for load than idle. HWInfo reads VSOC fine unlike Vcore, and Zentimings gives you all 4 in one convenient place.

VSOC is much more likely to be reflected in idle Package Power than Vcore.

It all comes back to the fact that scientifically observing idle characteristics on Ryzen is extremely difficult as one would expect from a CPU that might come out of idle several times a second at the slightest provocation from some obscure background process that isn't even visible to the user. CPPC updates something like every 2ms
 
Last edited:
RM is the most exact. It's just harder to interpret because there's no min/max column or logging like HWInfo, only observing with Eyeball Mk.1. SVI2 Vcore in HWInfo isn't slightly less accurate, it's completely incapable of showing idle Vcore. SVI2 vs die sense at the same moment:

View attachment 260256
View attachment 260255

Regardless, VSOC and the CLDOs should matter more since properly idling Ryzen draws a lot more power from VSOC than Vcore since Vcore should dip down to 0.2V/0.9V. The whole Gigabyte thing is more for load than idle. HWInfo reads VSOC fine unlike Vcore, and Zentimings gives you all 4 in one convenient place.

VSOC is much more likely to be reflected in idle Package Power than Vcore.

It all comes back to the fact that scientifically observing idle characteristics on Ryzen is extremely difficult as one would expect from a CPU that might come out of idle several times a second at the slightest provocation from some obscure background process that isn't even visible to the user. CPPC updates something like every 2ms

Ultimately though, it's not really about one measurement (Vcore, VSOC, whatever) that affects CPU power consumption though, right? Otherwise motherboard manufacturers would fix that sensor data and simply use that as the source.

One reason I like HWiNFO is the fact that you can turn on graphing and compare the current sensor reading with recent performance. During idling, I can see if the current number is a spurious spike or something indicative of normal idle performance by comparing with the graph. Especially with CPUs, there are always occasional loads from some background process doing something.

Now that you have described Ryzen Master's behavior, I have even less desire to use it, particularly for monitoring idle performance. So thank you for saving me the time and bandwidth to download it, play with it and scratch my head trying to figure out how to turn on graphing because it's not there.

:D

Hell, with the AIO pump in my daily driver, I see occasional momentary speed dips from 850 rpm down to 680 and I discount those as either erroneous sensor data since the speed returns immediately (no gradual ramp up or down). The only way I identified this was through HWiNFO's realtime graphs.

And I don't see how idle voltage settings in two motherboards is going to result in a 10 °C difference in identical CPU parts. A couple of degrees Celsius, yeah, okay maybe.

I have two 5600X CPUs in different builds and the idle temperature is 11 °C different. One is cooled by a 140mm AIO and the other by a 240mm AIO. And the former doesn't even really have the cooling capacity of a true 140mm radiator since the pump takes up the center of the radiator (the pump is not in the cooling block). My guess is if I swapped CPUs between the two machines, the idle temperature variance would still be in 9-11 °C range.
 
Last edited:
As far as my cooler and fan setting, they are pretty much the same. AIO is the same Galahad 240, I just move it from one system to another. Actually the same story happened to my 3700x (in the same setup before this), the cpu temp is just higher on the Gigabyte.

I did noticed at idle my Vcore is always higher on the Gigabyte. On Aida64 the 3700x often shows over 1.5v.

So if I mostly just gaming, didn't notice any performance difference and wanted less heat, next time I should opt for vendor other than GB?

I'm also actually a bit fed up with my Aorus board, the specs are great for the money, but I had many bios issues and until now it stuck every time on boot the first time. Asus is more pleasant so far, but it's generally expensive and Armory Crate is really a POS.
 
As far as my cooler and fan setting, they are pretty much the same. AIO is the same Galahad 240, I just move it from one system to another. Actually the same story happened to my 3700x (in the same setup before this), the cpu temp is just higher on the Gigabyte.

I did noticed at idle my Vcore is always higher on the Gigabyte. On Aida64 the 3700x often shows over 1.5v.

So if I mostly just gaming, didn't notice any performance difference and wanted less heat, next time I should opt for vendor other than GB?

I'm also actually a bit fed up with my Aorus board, the specs are great for the money, but I had many bios issues and until now it stuck every time on boot the first time. Asus is more pleasant so far, but it's generally expensive and Armory Crate is really a POS.

Like I said, 1.5V does not reflect reality, and idle temps are nigh impossible to compare properly. But if under load (ie. gaming), then yes, it's generally always a safer bet to stay away from Aorus if you don't want aggressive stock Vcore. Been for a couple generations of CPUs already, anecdotally same on the Intel side.

The most egregious Gigabyte boards seem to exploit power underreporting to boost power/scores/temps (HWInfo probably has a good writeup of what Power Reporting Deviation means somewhere on website). My B450I Aorus from a couple boards ago was the most blatant offender, ~70% power reporting deviation under load. My more recent Gigabyte B550 boards aren't nearly as bad, but still err on the aggressive side. I think all 3 of my Asus boards (B550-I Strix, B550M TUF, Impact) lean to the opposite side (100%+, technically overreporting). Nowadays, AGESA has means to mitigate this by editing power telemetry in most BIOSes, but I'm not sure how well it works.

Ultimately though, it's not really about one measurement (Vcore, VSOC, whatever) that affects CPU power consumption though, right? Otherwise motherboard manufacturers would fix that sensor data and simply use that as the source.

And I don't see how idle voltage settings in two motherboards is going to result in a 10 °C difference in identical CPU parts. A couple of degrees Celsius, yeah, okay maybe.

I have two 5600X CPUs in different builds and the idle temperature is 11 °C different. One is cooled by a 140mm AIO and the other by a 240mm AIO. And the former doesn't even really have the cooling capacity of a true 140mm radiator since the pump takes up the center of the radiator (the pump is not in the cooling block). My guess is if I swapped CPUs between the two machines, the idle temperature variance would still be in 9-11 °C range.

SOC power does scale visibly with VSOC, and since it's largely constant that's visible at idle. But correct, unlikely 10C idle temp difference from Vcore.

Much more likely load is just more spiking frequently / some background process when OP just happened to be observing. It's exceedingly difficult to get Windows into a state where it's calm enough to eliminate most idle spikes. Disabling services and programs, update, indexing, disabling all internet + rebooting, etc.

You can't really compare two different CPUs cooled by two different coolers though. Cooler contact is never the same, no two IHSes are the same, and AMD has some significant core delta variation in its chiplet products (10C+ at times)
 
Different boards = different behavior. Even if AGESA is the same, telemetry reports are very different between boards, As the CPU relies its operation on the board's feedback.
PCs are not really idling now days, even if their utilization (%) is low. A lot of different small things can be happening in background on different Windows installations.

In order to just try to come to a conclusion you need to monitor a lot of staff.
Core voltage, core powers, Core VID(effective), SoC voltage and power, active core counts, C-states sensors like "Package C6 Residency" and individual core C0/1/6 residencies, CPU PPT(avg), effective clocks might show something..., discreet clocks can or may not ....etc

We have to understand that these CPUs can self sample function parameters 1000times/sec (1ms intervals) and can alter those parameters 50times/sec (20ms intervals). The best monitoring software samples all those sensors at 1000ms = every 1sec. If you make software to sample faster (like 500ms or less) disturbance of the CPU state sky rocket and you driven further away from what it should've been. So no software can really monitor accurately these pieces of hardware.

Let alone that most software disturb CPU states even at 1000ms polling sample rate in their try to poll sensors.

The best you can do is to leave them as long as possible to the system state you want to observe. Like 1-2hours if not more. The longest you leave it the more accurate (average)values you will/may see. You can't draw any conclusion by observing current, min and max values.
1. In case of current you cant see all related parameters at once, you only see 1 and there is that 1-20ms vs 1000ms issue.
2. In case of min/max you cant really say that every min or every max value happen the same time and under same conditions.

In case of HWiNFO the "Snapshot CPU Polling" option is a must as this disturbs CPU the least (minimizes the "observer effect").

Power Reporting Deviation value has no meaning in anything else than 100% CPU load, so that goes down the drain also for observations close to idle, low or middle loads.
Just for the record, the same board with same BIOS version can report different telemetry deviations on different CPUs (I saw this with R5 3600 and R9 5900X).

RyzenMaster tells what ever AMD wants to tell. It uses AMD proprietary (and unknown) methods to sample and report values. And those few people that may know something (not necessarily everything) are under permanent NDA.

As for CPU voltage, I believe that the "CPU Core Voltage SVI2 TFN" is the most accurate one as this is polled straight from CPU and it matches very closely with "CPU Core VID (effective)" which is the only and final voltage request(to VRMs) from the CPU.
 
Last edited:
Back
Top