Default motherboard PL1 PL2 values for 65W CPU

[Gabriel02V]

I’ve recently purchased a MSI motherboard, model B365M PRO VH and I do not know if its behaviour is the designed one. More exactly I had installed a CPU Intel Core I5 8500, which has a rated TDP of 65 watts.

My concern is regarding the power draw. Using the default settings of the motherboard, the CPU is set automatically to 95W for long duration power limit (PL1) and 119W for short duration power limit (PL2). The cpu cooler I used is a Scythe Katana 4 which is rated for 130W, maximum dissapation. Which such settings, the fan climbs rapidly to 100% service and the CPU temp reaches 75-80C; which is quite hot, so I thought to lower the PL2 to 95W in order to decrease the heat as it can be notitced in the attached picture from UEFI/BIOS.

My questions are very simple:

• 1. If I have a 65W cpu, why the motherboards had set it to 95W for PL1 values and 119W for PL2? If I understood correctly from Intel web site the should be PL1=65 and PL2=1.25*PL1 which means aprox 81W
• 2. Is it safe for both CPU and motheroard to use this automatic setting of 95 for PL1 on long term?
• 3. Decreasing the PL2 from 119W to 95W CPU temp is stable at 70C and the fan goes below 100%.This setting should be safe, right?

With the previous system (Intel 1155 platform),I had never experienced such high temperatures using an I5 3470 (77W TDP).

 

[vick1000]

What are your temps in various situations and loads? What are you ambient temps?
With good case airflow and decent ambients, the Katana 4 with proper installation, should do fine on that CPU.

 

[Gabriel02V]

The temperatures I got were for the scenario of 100% cpu usage, considering Prime95 stress toool for an ambient temp is about 25-26C, with turbo boost activated. Obviously, depending on the power limit settings, the cpu temp can be hotter or not; as I said above with PL2=119W and PL1=95w I got 75-80C stable and all core boost frequency of 3900 mhz. Remark: making PL2=PL1=95W, the cpu temp drops to 70C.
But I'm not questioning if the temps are high or not because they are still within normal limits; I'm asking why my 65W cpu i5 8500 is recognized by the motherboard as 95W for PL1, respectively 119W for PL2 with the default settings and if it safe to keep these values for long term use.

Or perhaps I'm missing something. Did Intel changed the definition of the TDP? meaning that 65W TDP can be achieved only when turbo boost is off?; so it means that when all cores are 100% used, the maximum clock can not be higher than 3000 mhz, because in order to reach 3900 mhz using turbo boost then 95W-119 watts of power are needed to maintain the 3.9 ghz?. With the previous platform I had, an i5 3470 (ivy bridge, 77W tdp) with the all core boost activated was always less or equal to the TDP and the temp were 60-65C under the same stress scenario, using the same cooler.

 

[citay]

The mainboard makers largely disregard the official Intel guidelines for PL1 and PL2. In the past, they did so silently. The higher-end the mainboard was, the higher power limits they allowed. This looks better in benchmarks during mainboard reviews of course. Only on the lowest-end boards, they stuck a bit more closely to what Intel says, because otherwise the weak VRM would overheat, not to mention that some people might use the boxed CPU cooler there.

With the 500-series boards at least, MSI has added this sort of "Power Limit selection screen":

Of course, for all but the highest-end CPU models, Tower and Water cooler will be one and the same thing. Only a 10900K/11900K has a good chance of surpassing 288W of a power limit under full load such as Cinebench R23 Multi.

But the first option will always adhere to Intel, in this case (i5-11500) to 154W PL2 for 56 seconds Tau, then 65W PL1.

This setting influences the power limits on the 400-series boards:

I think on the 300-series boards you have to do it manually, there is no "CPU Cooler selection" option. MSI makes it seem like the only factor is the CPU cooling, but don't let that fool you, the VRM needs to be good enough too.

So the choice is: Enforce the strict PL1 and PL2 power limits with strict Tau time (Greek letter T stands for the time period in physics) as specified by Intel, or fine-tune the limits yourself to adapt to your cooling. The latter is not risky per se. You simply go by the CPU and VRM/MOSFET temps.

The power limits themselves are found on this page which you posted already (this clean screenshot can be taken directly from the BIOS by pressing F12 and saving to USB stick):

Long/Short Duration Power Limit, and Duration Maintained (4096W is unlimited). Also here in "Advanced CPU Configuration", or in older boards, under "OC" > "CPU Features":
"Intel C-State" to Enabled.
"C1E Support" to Enabled.
"Package C-State Limit" to C10 or the highest C-number there is (C10, C8, C7, C6, depending on the board).
"Intel Speed Shift" to Enabled.

But those are just small things. The main thing is the power limits. Setting strict power limits will constrain the CPU somewhat, as it can't boost the frequencies so high anymore, but it will also keep temps in check.

Here you can see a test i did with a 9600K on my old system:

The AIDA64 at the bottom left is just monitoring the fan RPMs, which i set to a temperature-dependant curve.

What you see here is, the 9600K has a Long Duration Power Limit of 95W, and a Short Duration Power Limit of 118W, with a Tau of 28 seconds. It's actually not going very far above those limits if you disable them all. But observe the fan curves: After 28 seconds, the power consumption drops to exactly 95W, and the temperature-controlled fans follow suit. You can see the slight kink halfway through each of the (almost) three test runs, where the RPMs become lower.

Here is an illustration of the limits for reference:

From https://www.anandtech.com/show/13544/why-intel-processors-draw-more-power-than-expected-tdp-turbo

You can see, it sort of mimicks that graph, but in a flatter way, cause the 9600K, even with all limits disabled, doesn't go much above PL1.

For CPUs which are not too far over the limits anyway, you can disable the limits altogether without a big downside, or leave them in place, doesn't matter. Provided you have adequate cooling. There are not many cases where long multi-core load will happen, certainly not in games (the power consumption stays below the limit there), more in encoding tasks. But still it's not a huge loss in performance if you apply the Intel limits when the CPU doesn't go far above the limits if it's unrestrained.

For CPUs which are limited more by the Intel power limits, you have to look at it case-by-case. If you tell someone to disable the limits on a 10900K or 11900K, a high-end board and high-end cooling would be a must. Otherwise it will quickly run too hot and you'll have CPU throttling, VRM throttling, or both, which would lower performance. With cheaper boards, you would also worsen the longevity from all components around the VRM heating up too much. But if you unnecessarily restrict the CPU too much, you leave performance on the table.


However, you have another great option to reduce the temps: Select a lower "CPU Lite Load" setting, to use a lower frequency-vs.-voltage mapping for the CPU VCore (requires stability testing).

CPU Lite Load is basically the "silicon quality" setting, they are testing hundreds of CPUs for the variance in quality, i.e. how high of a VCore is required for stability, and then they put in a default value like 8, 12, or even 17 (first time i ever saw such a high value is on your picture), to be able to run 99.9% of the CPUs of varying VCore requirements. If you lower the setting, you are fine-tuning it down, to be more exact to your specific CPU. But you must ensure that it is Linpack-stable.

Linpack Xtreme: https://www.techpowerup.com/download/linpack-xtreme/

Run Linpack, select 2 (Stress test), 5 (10 GB), set 10 times, press Y to use all threads, and let it do its thing.
It's by far the best tool to detect instability, reacts even quicker than Prime95. Warning, this generates a lot of heat. So watch the temps using HWinfo64 Sensors.

CPU Lite Load Mode 1 applies the lowest voltage, which can be unstable. The default value will be quite high, to be able to run the worst silicon (highest VCore required in testing) without stability issues. But most CPUs have better quality and you can go lower, decreasing power draw. Stability must be verified. For my 9600KF, it was unstable at CPU Lite Load 3, while 4 was stable, then for additoonal headroom, i went to 5. Rock stable and considerably less power draw than the default. So, work your way down until you hit a setting that is unstable, test the next highest mode, if that's stable, go one mode higher for headroom. When it's hot in the summer, the CPU can be more prone to errors, so set it one mode higher than the lowest stable one, to have some headroom for extreme circumstances. Only if your CPU is stable all the way down in Mode 1, like mine, you can keep it on Mode 1.

As you can see there on the picture (old screenshot from a beta BIOS), in my system right now, i set "CPU Lite Load" to Mode 1 for my 11500. So, no added VCore in any load situation, which is stable for me (good quality CPU). But each CPU has its own lowest mode. You also see all the other options i set, like Package C-State Limit C10 (biggest number C-state available).

 

[Gabriel02V]

Ok, it is more clear now. I can conclude the following:
- In regards to "cooler type selection menu", I've checked the UEFI settings and indeed, there is no such thing on B365M PRO VH model.
- To get the boost feature more stable at higer clocks, for non-K CPUs, the power required will surpass the 65W TDP rating; basically non K CPUs with the unlocked power limits, will be quite close in terms of power consumption as K models.
- The energy efficieny and lower heat are achieved only with the base clocks when using the official intel TDP rating for that specific cpu.
- As B365M PRO VH has included on the supported list the model 9900KS which is a 127W TDP PL1, it means, that I can use safetely the default settings for i5 8500 of 95W for PL1 and 119W for PL2 with a mainstream aftermarket cooler as Scythe Katana 4 in the same way as you presented above the default settings for tower cooling when using i5 10500 (65w rated TDP) when PL1=PL2=288W

 

[citay]

Right, your 300-series board has no cooler selection prompt, instead it will just remove the power limits if they are left on Auto, by setting them to 4096W.

Yes, the non-K CPUs will be almost exactly like the -K CPUs once you remove the power limits (minus the 100 or 200 MHz difference between them in frequency, and minus the OC capability). The base clocks are almost irrelevant, because most of the time, the CPU will use the turbo to boost to higher frequencies according to the type of load. It's not like setting the official power limits will restrict the CPU to the base clock either, but it can make it not boost as high for the turbo clocks. How severe this effect is depends on how low the power limits are set and how much your CPU model is restrained by that. The latest 65W parts with a high core count would be severely restricted in certain situations by setting the power limits exactly to spec. Older 65W parts a bit lower in the model range would not be restricted as much. Your i5-8500 is much more on the tame side. I would guess that it's unrestricted power draw under full multithreaded load is no more than 100-130W.

Energy efficiency is mostly a calculation of the energy spent for the completion of a certain task. This image from the article i linked is interesting in this regard:

Unrestrained, the CPU will be the fastest to complete a given task. With a PL1 slightly below its power consumption when unrestrained, it doesn't lose that much performance for shorter computation tasks, but will stay cooler. With a PL1 a lot below, it will stay much cooler again but also take longer, and this performance disadvantage will only increase with longer tasks (like long rendering runs).

Now, from this, you would have to go one step further and calculate the total energy spent to complete the task. It's possible that the CPU with the lower/stricter PL1 is more efficient. But you also may want a certain level of performance. People who earn money with their PC by doing rendering work etc. will not care about efficiency, only performance. Same often goes for people who overclock. Other people may want to optimize efficiency more. This depends a lot on how far the CPU has been taken on the voltage-per-frequency curve.

For each CPU model, there is a curve such as this:

The newer and the more high-end a CPU model (both from AMD and Intel) is, the steeper the increase of the curve, so the 100% mark (stock setting) will further towards the right. With almost all the most modern CPU models, certainly the higher models of the 10th/11th generation for Intel, overclocking is pretty much useless, since they use 99.9% of their capabilities with all the different turbo modes, and use very agressive frequencies at stock. Meaning, they're at a place on the curve which was previously only reached in an overclocked state. But that's just an aside. Again, your CPU is far from being set up that aggressively at stock, it's an older kind.


Your MSI B365M Pro-VH is severely restrained by its VRM (voltage regulator module), the area to the top and left of the CPU, which transforms the 12V coming from the EPS cable into the ~1V for the CPU to use. It uses cheap components throughout and it doesn't have any heatsinks. Cheap VRM controller, few VRM phases, cheap discrete (seperate) Lo- and Hi-MOSFETs per phase instead of integrated Powerstages, no heatsinks, that makes it only suitable for entry-level CPUs.

The technical difference of discrete MOSFETs vs. Powerstages is explained a bit here, https://www.gigabyte.com/MicroSite/312/images/PowIRstage.html

Basically, seperate MOSFETs have considerably higher switching losses, so you get way more heat from the VRM. Each time one of the MOSFETs switches on and off, you have energy turned into heat, and it's way more heat than with a single powerstage which combines the components. And the more power your CPU model draws, the hotter the VRM. So if you were to put a 9900K on there, no matter how good your CPU cooling was, it would now be the VRM that is the limiting factor and would overheat quite quickly. Then you would have VRM throttling or the board would turn itself off.

That is the misconception with the CPU support list. It is not a "CPU suitability list", it is strictly a list of CPUs that the board would recognize with a certain BIOS version. Nobody in their right mind should put a high-end CPU on an entry-level mainboard model, that is just asking for trouble. Best case, you would have to restrict it severely by setting way too strict power limits, making it perform like a much cheaper CPU.


Keeping all that in mind, in your situation, 95W/119W are still reasonable. I would highly advise to try that trick with "CPU Lite Load" i mentioned. I bet you can lower power consumption by a bunch, which can increase performance if the turbo was limited by the power limits somewhat. And without any downsides (provided you verify stability).

 

[Gabriel02V]

I will consider lowering the PL2 of 119W to 95W and let the default values of PL1 to 95W. In this way it will act as 95W TDP CPU and I think it would be the best balance between performance and vrm health. There have been for decades 95W TDP CPUs used on normal motherboards without the gaming features such as aluminum heatsinks and no problems were noticed as long the manual overclock was not applied to stress the vrms.
In most of the cases the CPU power draw will be below PL1 because the 100% usage is somehow valid in stress scenarious and I do not expect a power draw of more than 60-80W even when gaming.

 

[citay]

Yes, in gaming the CPU power draw will not be as high, your graphics card will be the limiting factor.

 

[Gabriel02V]

I come back to this topipc with an updated. It seems that not the PL1 and PL2 values are decisive factors for the higher temperatures I reported in my first post. CPU Lite Load (CLL) feature is much more efective but unfortunately I didn't understand much how it does adjust the values of cpu voltage according to Intel design. In UEFI settings I found to be set automatically to value 17 whereas the minimum is 1 and maximum 20.
Currently I set it on 5 and the temperatures dropped significantly by 5C (also the fan speed way down to 1700 RPM) and all core turbo is still 3.9 GHz and 1.098V average voltage per core (please refer to the printscreen from below for the results). Also if I set the CLL to 1, the all core turbo boost is the same 3.9 GHz without noticing any instability. Before with CLL 17 the voltage was at least 1.12V temp 70c, and 2000+ rpm fan speed.

Can anybody explain what's the default value/the recommended settings for I5 8500 cpu?

CLL 5 results

 

[citay]

It's like i explained before, CPU Lite Load is a sort of "additional CPU voltage" (additional VCore) from MSI, which aims to make all CPUs of varying silicon quality run stable. They have tested many different CPUs and determined a setting that will ensure stability, even if your individual CPU doesn't have a good quality and needs a higher VCore than other CPUs of the same model. Now, if you lower CPU Lite Load (while ensuring stability), you are fine-tuning it down to the exact VCore mapping that is sufficient for your specific CPU. You are taking off some of the additional VCore that MSI normally adds, because a lot of CPUs are still running 100% stable with less additional VCore than the high average value that MSI has determined.

However, instead of waiting if you notice instability, you must ensure that it is Linpack-stable.

 

[Gabriel02V]

Ok. So, If I understand it right, if the entire scale from 1 to 20 represents additional voltage, then the closest stock value should correspond to level 1 from CPU Lite Load setting. So it is not considered undervolting if the CPU Lite Load is set below 10.

 

[citay]

Almost. The fact is, every mainboard gives a bit of additional VCore to what the CPU requests (VID). For example, on ASUS mainboards the relevant option should be "CPU SVID Behavior".

They all have evaluated many CPUs in-house and determined a range from best-case to worst-case for different quality CPU specimen, in regards to how much voltage they need. If your specific CPU is a worst-case example, then it actually needs a good bit of additional voltage to run stable in all circumstances. So if you then lower "CPU Lite Load" too much, it equates to a too low voltage for that specific worst-case CPU, and the CPU will be unstable. But most people don't have such a bad CPU, so they can lower the mode by several steps and it will still be stable. Then, instead of undervolting, it would really just be "taking off the unneeded additional voltage". But you could also call it undervolting, since you are lowering a voltage that MSI has determined to be a good default setting, so all CPUs (even bad ones) run stable.

Doesn't really matter what you call it. I call it fine-tuning to your specific CPU, and since the effect can be so enormous in a good way (lower power consumption and heat), it's definitely worth doing.

 

[Gabriel02V]

Alright, I lowered CPU Lite Load to level 1 and run 15 mins of Linpack stress test and it didn't reset, or freezed in this period.. The maximum power draw registered was 91W, so it was below than the default values for PL1 (95W) and PL2 (119W). The maximum temp was 70C temp at 80% fan speed and you were right: Linpack is more relevant than Prime95.

The default setting CPU Lite Load=17, in my opinion it somehow overestimates the quality of this cpu's wafer. For example in Prime95, where I tested it initially, the maximum power draw climbed to 109W for 16s (PL2) and then to 95W (PL1), and also all core boost varied between 3.9 and 3.7 GHz having a temp of 80C. Now, the new setting CPU Lite Load=1, I get 85W maximum power draw, so it is below PL1, temp 63-64C (60% fan speed) and all core boost stable to 3.9 GHz.

The thing is I noticed a similar behavior few months ago, on i3 10100 (65W TDP) with asus h510m-d motherboard which I was suprised when it run a bit hotter (3-4C) than my previous cpu i5 3470 (77W). Now, I understand why: cpu wafer's quality is kind of a lottery, and best way for Intel Corp. to sell the "defective" lots is to apply more voltage from the motherboard' settings.

The bottom line: the differences in terms in terms of power consumption and heat are really noticeable between level 1 and level 17 from CPU Lite Load setting.

 

[citay]

Yes, there can be a huge benefit from adjusting CPU Lite Load to the optimal setting for your specific CPU. Many CPUs don't need nearly as much additional voltage as the average value that MSI set.
Not many people know about this setting, but i like it a lot, and everyone who's tried it only gave me positive feedback about big savings in power and heat.

 

[Gabriel02V]

It would have been less confusing if Intel would have mentioned for each cpu a specific voltage range; because we know that a higher voltage reduces the lifespan of the product overtime.

 

[citay]

I don't think it matters too much, most people will have bought a new CPU after less than 10 years anyway. While you can immediately degrade a component by applying extreme voltage, for a gradual degradation with that comparatively "safe" extra voltage from the mainboard to take effect over time, a CPU will likely run problem-free for much longer than 10 years. Hardware like PSUs, with their electrolytic capacitors, ages much faster than a CPU, so i wouldn't worry about the CPU lifespan in particular.

Reducing the extra voltage has more of an immediate effect in that you lower power consumption and heat output, not so much to prolong the CPU lifespan. Of course, less power consumption also means lower temps in the VRM as well as potentially slightly lower temps in the PSU, depending on its fan control, so it's always good.