how to reduce mainboard power consumption?

[charonme]

Are there any good ways of reducing the power consumption of MSI boards, especially if I don't use discrete graphics and only one nvme ssd? Mine is the PRO Z790-P wifi DDR4
Maybe something like reducing PCIe Lanes?

 

[citay]

Under "Advanced CPU Configuration" in the BIOS, i usually set the following:
"Intel C-State" to Enabled (Auto should equal Enabled, but anyway)
"C1E Support" to Enabled (clocks down in idle)
"Package C-State Limit" to C10 (this way, the CPU will enter C7 state in idle, instead of only C6. The C7 state lowers power consumption in idle by around 6W for me.)
"Intel Speed Shift Technology" to Enabled

If "Package C-State Limit" on C10 causes problems (quite rare, but it's theoretically possible with some configurations), then just leave it on Auto instead.

Now, you will know this, but others that may be reading: "CPU Lite Load" is a setting that is indivual for each CPU (not just the CPU model, but the individual CPU), and the best setting for that can be found out using my guide. Lowering this from the default mode can yield great improvements for power draw under load and efficiency.

Then under Settings\Advanced\Power, set ErP Ready to Enabled. Reduces power consumption in off state to less than 1W instead of 2-3W.

Under Settings\Boot, set both Fast Boot options to Disabled (Fast Boot is not good for much and is sometimes problematic).

Then go here:

You can experiment with PCIe Native Power Management and Native ASPM set to Enabled or Disabled for a test.
Also set these settings in the bottom menu like so:

Should there be any problems with the graphics card from this, set PEG0/1 ASPM back to Auto, or set "Native ASPM" to the opposite value.

It's also worth trying to bring down the so-called IMC-related voltages under the OC section (the voltages that power the Integrated Memory Controller), because once XMP is enabled for the RAM - depending on how aggressive of a profile that XMP is - the BIOS will raise some voltages way more than required. For DDR5, the voltages are CPU SA Voltage, VDDQ, and VDD2. As an example, for DDR5-6000 @ 1.25V, i recently could decrease the Auto voltages of SA Voltage 1.25V / VDDQ 1.3V / VDD2 1.3V down to VCCSA 1.0V / VDDQ 1.2V / VDD2 1.2V without a problem.

In Windows, make sure the power plan is set to "Balanced" (the only proper one).
Then set PCIe Link State Management to Maximum Power Savings in the Balanced power plan:

I get an idle power draw of only 22.5W (whole PC) with my Z690 DDR4 board that way, combining all the measures and using an efficient PSU.
As can be seen here on the table, Explained: How the new BIOS versions are causing higher temperatures.
This is a very low idle number for a modern system, and that is with eight SSDs in the system too (3x M.2 PCIe, 5x SATA).

Some of the improvements from these settings, you can only see if you hook up an energy monitoring device in front of your PSU on the wall outlet. This will show what really happens. It's more difficult to tell this just from the sensors you see with HWinfo and such.

 

[charonme]

the first half are CPU power related settings, are you sure it also affects how much the motherboard draws?
thanks for the ASPM pci settings, I'll try those
IMC settings are also mostly CPU related, although some of the RAM settings probably affect how much the RAM draws (probably mostly the ram voltage itself)
I'm already using SA=1.1V and VDDQ=1.24V, I think lower values weren't stable the last time I tried

 

[citay]

the first half are CPU power related settings, are you sure it also affects how much the motherboard draws?

Well yeah, each power draw reduction of the CPU also affects the motherboard. I would go so far as to treat the board and the CPU as a unit, but even if you want to view the board in isolation, the CPU VRM of the motherboard is the main area of power draw, due to the VRM's switching losses. You can easily add another 10% on top of the CPU power draw to arrive at what the VRM would draw from the PSU (and with inefficient discrete MOSFETs instead of powerstages, even more). For the minor CPU voltages you also have small VRMs on the board, for example there are VRM circuits below the CPU socket.

Actually, for the VRM, there is one further trick to bring down the power consumption slightly:

To reduce the overall switching losses in the VRM, set it to the lowest switching frequency here (for CPU and CPU GT switching frequency):

Depending on the board, "Auto" can sometimes already mean a pretty low setting. If the Auto setting is quite high for some reason, then the lowest setting can lead to worsened transient response, so with a power-hungry CPU, i'd do some Prime95 testing with a bit of random mouse movement in between to cause transient load. Usually, on a mid-range board using powerstages in the VRM, this switching frequency can be set low, the VRM can still deal with most CPUs without too much trouble.

You can't see the effects of this easily on any sensors, except maybe the MOS temperature will be lower afterwards, but usually you have to look on the energy meter to see the effects.

For the rest of the board, all you can do is enable all the PCIe power saving as described. Unless you want to turn off onboard chips, which is what i have also done, like WLAN/BT (don't use it), or onboard audio (my speakers are active USB ones). Or for example if you use WLAN instead of LAN, you could disable the onboard LAN.

You can't really seperate CPU and board fully from each other, but from all these settings, you will see idle power draw go down quite notably on the energy meter. So i thought i would just mention pretty much everything i use. I even experimented with RAM power saving settings, see here, but those can't really achieve that much in the big picture.

 

[charonme]

Well they are somewhat separated now because the board is supplied with the ATX connector and CPU (with its VRMs) is supplied via the EPS cable which I can measure with a clamp ammeter. I probably should experiment with the PWM settings (I originally assumed Auto was the lowest setting) because my EPS measurements are quite higher than the package power reading from hwinfo, sometimes even 38% higher and that seemed very suspicious to me since I got the clamp ammeter, but I haven't been able to resolve it.

 

[charonme]

I tried some measurements and it looks like the default ("Auto") PWM setting was the highest! Although the frequencies I was detecting slightly varied with different loads, they were never exactly what the bios setting said

[edit] for some reason the EMG noise spectrum on the back of the VRMs indicated the "Auto" setting was equal to the highest "677kHz" setting, however now that I've done power measurements it looks like the "Auto" setting draws as much power as the lowest "238kHz" setting (281W on the EPS cable with PL1=210W) even though the "677kHz" setting only draws around 1.7% more (if that's even above the measurement inaccuracy) and after 24 minutes of P95 the reported VRM temperature was 66.5°C and then it crashed with a BSOD. I seem to remember the highest setting also tended to crash a couple of months ago when I was trying it out.
Next I switched back to 238kHz and after more than an hour of P95 the max VRM temperature was at most 63.0°C and sometimes it drops below that.
Then I tried the Auto setting and it crashed after it reached 62.5°C in a few minutes even though I thought this setting was already established as stable by much longer tests. Am I seeing degradation here? Let's try the 238kHz setting again...

 

[citay]

Yes, sometimes (especially recently) they really seem to err on the side of caution with some BIOS settings, aiming to achieve the highest stability even in the worst-case scenario. Yesterday i updated a B660 board, what was the new default for CPU Lite Load? Mode 18 again.

I haven't looked into wether they might be pulling a similar move on the switching frequencies, and i can only measure it indirectly by comparing the full system's power draw under load (where the differences are the most apparent) with different switching frequencies, but right now, nothing would surprise me anymore. With higher switching frequencies, you get better transient response from the VRM, but it's a game of diminishing returns after a certain point. The VRM gets hotter and hotter, the higher the switching frequency (from the accumulating switching losses), but you don't get similarly improved transient response. Plus, some other settings may actually work better (aka be tuned for) a lower switching frequency.

So usually i do a quick baseline power draw test with the default switching frequency, then set it right to the lowest setting and see if that can lower the power draw by a good amount. If so, i keep it on the lower setting, if not, i might as well keep it on the default, in case MSI actually put some thought behind their default selection. But with recent developments like selecting a sky-high default mode for CPU Lite Load - which is only benefitting them and Intel, not the end user - i am starting to question any and all defaults. They are really not what is best sometimes.

 

[charonme]

Right now it looks like the default Auto setting is equal to the third lowest one ("349kHz") but it still doesn't seem to affect the EPS power and the VRM losses are independent from the PWM frequency and are above 31% and noticeably rising with VRM temperature

[edit] oh finally OK after seeing this chart (from this test) of various motherboard EPS cable measurements in CB R23 with 125W power limits I'm much less worried about my EPS measurements. My board seems to be somewhere in the middle of the table with 157.2W @ 125W PL1
Although I now regret that I haven't realized this comparison existed, I'd probably have chosen the MAG Z690 Tomahawk DDR4 instead of my PRO Z790-P wifi DDR4 (although I'm not exactly sure what the Z690 lacked compared to Z790)

 

[citay]

I recently built a PC with this very board model and a 12400F for a budget-oriented system, and i have to say, the VRMs got quite toasty (especially the upper part where the heatsink is a bit small, difficult to see here). But then again, that's not too unusual to see on anything than a higher-end board.

31% losses does seem high though. Here's from the datasheet of the AOZ5016QI DrMOS powerstages (Driver + Hi-/Lo-MOSFETs in on package) used on this:

fSW is the switching frequency, so from 600 KHz to 750 KHz, even this small step you see the efficiency drop. Lowest efficiency is at about 82% with low currents being pulled. But of course there are also losses in other components like the chokes. If i roughly calculate for 210W CPU power draw, then in these DrMOS parts alone (14 of them), we have maybe 3-4W losses per DrMOS, so that's around 40-50W already, then with the chokes etc. we can have over 50W combined VRM losses for sure under full load within the power limits.

Also, the leakage current within the DrMOS rises exponentially with rising temperature. So it is doubly important to have heatsinks with a large enough surface area that are well-oriented for the airflow (and there actually being some airflow in the area), to keep the VRM as cool as possible for lower overall losses. Same goes for the CPU or any integrated circuit of course, there is a correlation between the temperature and the leakage current / higher resistance. The cooler you can keep a circuit, the better it will perform, in general.

 

[charonme]

31% losses does seem high indeed, that's why I was so worried about this, but seeing that comparison table (that's with a 125W power limit) it looks like that's not that uncommon and that some other factor plays a big role besides the VRM datasheet efficiency. Perhaps there is a huge loss on the powerplane and the power traces between VRMs and the socket in the motherboard itself, those get pretty hot too. Or maybe the way the power is lost on the VRMs is much more complicated in the way they're used on motherboards than their datasheets would suggest

 

[citay]

Yes, it's never as simple as in the datasheets, where it's just a single DrMOS / powerstage in isolation on an "infinite" PCB. Once you have them tightly packed in a VRM, and once the inductors/chokes and capacitors come into play, and depending on how the PCB is built, how the VRM is designed and tuned, it can go all kinds of directions. This board is certainly not the model of efficiency.

I think my own MAG Z690 TOMAHAWK WIFI DDR4 is a positive outlier here, even buildzoid was excited about the powerstages it uses when testing the sister board model. And by the way, this sister board model, the Z690 Torpedo EK X, thermally connects the VRM to the watercooled CPU block. See here for a very interesting article, https://skatterbencher.com/2023/05/30/2-phases-1-raptor-lake/

 

[charonme]

MAG Z690 TOMAHAWK WIFI DDR4 is a positive outlier here,

Yeah it's at the top of that list
are you seeing any downsides of Z690 compared to Z790?

 

[citay]

are you seeing any downsides of Z690 compared to Z790?

No, other than IA CEP not being exposed as a setting (but it's not interfering when i lower CPU Lite Load by a lot), and some other settings like 100 MHz FSB lock are also not there. But no dealbreakers of any kind. Z790 is only a slightly modified Z690 anyway.

 

[charonme]

I tried blowing air at the back of the motherboard to check if the EPS->CPU package inefficiency is caused by poor cooling of the traces in the motherboard. The MOS and CPU socket temperatures went down considerably (by around 21°C), but the EPS->CPU package losses went down only by around 12 watts (approx 15%). Fun fact: the fans consumed more than that

anyway another "fun" statistic: when playing a video the entire PC eats 62W (cpu package shows 6W) and 55W in idle

 

[persiz154202d8]

The CPU switching frequency is an interesting option. It may slightly increase the power consumption at higher frequency but also may help to lower the CPU voltage keeping the same stability. So at the end of the day you ending up with a more power efficient system by using high switching frequency. My energy meter shows negligible difference between Auto and 677 KHz(the maximum one on my MB) if any, but helps me to use lower CPU Lite Load Mode.

 

[charonme]

may help to lower the CPU voltage keeping the same stability

Yeah I theoretically expected this, but wasn't successful in confirming it with measurements (yet?). I was able to confirm that the frequency proportionally really changes, so it's not like there is a bug in the bios that would ignore the setting, but I didn't see any change in voltage spikes, stability, power draw or efficiency