Unable to set power limit on Z490 Aorus Elite AC

[porina]

Mobo as subject, currently with 10600k in it. Looking to pre-test before replacing with Rocket Lake. One thing I wanted to do was set a power limit. I found the section in bios, set PL1 and PL2 to 65W, and rebooted. Fire up Prime95, 89W. Mobo did have settings for Package Power Limit and Platform Power Limit. I tried also setting Platform in addition to Package, no difference. Still 89W.

 

Running at mobo defaults without power limit, it took 114W. The limiter did reduce it, just not as much as expected. I don't have much experience on Gigabyte mobos, but similar setting has worked fine on Asrock for example.

 

Any ideas?

[Mister Woof]

Seems Gigabyte default is basically no power limit.

 

For my Z590, in Advanced CPU Settings, I only see one option for Turbo Power Limits: Turbo Power Limit - Auto/POR(idk what this means)/Enabled.

 

Auto I believe is unlimited, POR I have no clue, and Enabled allows you to set power limits.

 

There is probably some other setting somewhere that is set to "auto" and throwing you off somewhere.

 

EDIT:

 

Found this:

 

https://pcpartpicker.com/forums/topic/372841-how-to-manually-set-cpu-power-limits-on-gigabyte-z490-vision-d

 

• Billy-Bob_Kenobi said:

Quote

The setting you're looking for is called "Enhanced Multi-Core Performance" and is located in the "Tweaker" section of the advanced BIOS. It will be enabled by default. Disable that, and the board reverts to the stock turbo boost rules.

I believe Enhanced Multicore Performance Auto = normal multipliers, unlimited power; enabled = maximum multiplier to all cores + unlimited power; disabled = Intel spec for both multipliers and power limits.

[unclewebb]

@porina For most motherboards, the BIOS does not give you access to the Clamp option. You need to use Clamp for full control of the power limits.

 

Here is an example of 10 threads of Prime95 with both turbo power limits clamped to 65W. The CPU never exceeds 65W. Power consumption is clamped to this value the moment this test starts. The CPU slows down as much as necessary to keep power consumption in check. If the power limits are unlocked and Clamp is not checked, power consumption during the same test will easily go over 150W.

 

 

[Mister Woof]

4 minutes ago, unclewebb said:

@porina For most motherboards, the BIOS does not give you access to the Clamp option. You need to use Clamp for full control of the power limits.

 

Here is an example of 10 threads of Prime95 with both turbo power limits clamped to 65W. The CPU never exceeds 65W. Power consumption is clamped to this value the moment this test starts. The CPU slows down as much as necessary to keep power consumption in check. If the power limits are unlocked and Clamp is not checked, power consumption during the same test will easily go over 150W.

 

 

Thanks for this - do you know what POR stands for in Turbo Power Limit (above)?

 

Google has failed me and I can't for the life of me guess what it means.

[unclewebb]

30 minutes ago, Mister Woof said:

do you know what POR stands for

Not sure. I just have the basic settings on my Asus board.

 

[Staypuft]

On 3/22/2021 at 2:02 PM, Mister Woof said:

Seems Gigabyte default is basically no power limit.

 

For my Z590, in Advanced CPU Settings, I only see one option for Turbo Power Limits: Turbo Power Limit - Auto/POR(idk what this means)/Enabled.

 

Auto I believe is unlimited, POR I have no clue, and Enabled allows you to set power limits.

 

There is probably some other setting somewhere that is set to "auto" and throwing you off somewhere.

 

EDIT:

 

Found this:

 

https://pcpartpicker.com/forums/topic/372841-how-to-manually-set-cpu-power-limits-on-gigabyte-z490-vision-d

 

• Billy-Bob_Kenobi said:

I believe Enhanced Multicore Performance Auto = normal multipliers, unlimited power; enabled = maximum multiplier to all cores + unlimited power; disabled = Intel spec for both multipliers and power limits.

I have the Gigabyte Z590 Ultra with a 10700K.  Have been messing around with it looking for a way to just get power consumption under control (high temps).  Started with setting Enhanced Multi Core Performance to disabled.  This lowered the power consumption a bit, but it was not running within the Intel spec, so don't count on this setting being a one-click-fixes-all solution.  Basically the board is ignoring the Turbo Boost Power Limit Tau (time duration for PL2 before PL1 kicks in) with basically no TDP limit at stock clocks, regardless of the EMCP setting.  My particular chip settles out at ~196W steady state power during Prime95 torture test (as reported by HWinfo) with this setting disabled.

 

Next step was to set the CPU Internal AC/DC Load Line to Power Saving.  This worked to drop steady state power consumption to ~160W.  Temps entered the reasonable realm, but I wasn't happy that this was such a vague setting.

 

Next I tried various combinations of Turbo Boost Power Limits with the AC/DC set back to Auto.  In this process I first tried POR.  This appears to be the setting for "Intel Standard" and it set my PL1 to the proper 125W, no idea what Tau was, but PL2 was 156W... IDK, makes no sense.  The processor downclocked to 125W (@4.1GHz) after a number of seconds at 156W, overall temps were great.  Not happy with 4.1 so I kept going.

 

Next I tried the Turbo Boost Power Limit set to Enabled and messed around with various settings for PL1, PL2, and Tau until the temps were under 80C and the lowest steady state clock in Prime95 was 4.5GHz for PL1.  I'll run with this until deciding to overclock.  Based upon my test results, I'll need a stronger CPU cooling solution for that.

 

 

For the OP, I would recommend experimenting a bit more with your power limits, keeping in mind that the 10600K PL1, PL2, and Tau are 125, 182, and 56, respectively.  Not sure how the board will handle the 65W lower limit.  It may be setting it's own lower limit.  Try 100W and see if it works, then you'll know 65W is too low for that board (not to say too low for the chip, the board might just be doing its own thing on the lower limits).

[Mister Woof]

25 minutes ago, Staypuft said:

I have the Gigabyte Z590 Ultra with a 10700K.  Have been messing around with it looking for a way to just get power consumption under control (high temps).  Started with setting Enhanced Multi Core Performance to disabled.  This lowered the power consumption a bit, but it was not running within the Intel spec, so don't count on this setting being a one-click-fixes-all solution.  Basically the board is ignoring the Turbo Boost Power Limit Tau (time duration for PL2 before PL1 kicks in) with basically no TDP limit at stock clocks, regardless of the EMCP setting.  My particular chip settles out at ~196W steady state power during Prime95 torture test (as reported by HWinfo) with this setting disabled.

 

Next step was to set the CPU Internal AC/DC Load Line to Power Saving.  This worked to drop steady state power consumption to ~160W.  Temps entered the reasonable realm, but I wasn't happy that this was such a vague setting.

 

Next I tried various combinations of Turbo Boost Power Limits with the AC/DC set back to Auto.  In this process I first tried POR.  This appears to be the setting for "Intel Standard" and it set my PL1 to the proper 125W, no idea what Tau was, but PL2 was 156W... IDK, makes no sense.  The processor downclocked to 125W (@4.1GHz) after a number of seconds at 156W, overall temps were great.  Not happy with 4.1 so I kept going.

 

Next I tried the Turbo Boost Power Limit set to Enabled and messed around with various settings for PL1, PL2, and Tau until the temps were under 80C and the lowest steady state clock in Prime95 was 4.5GHz for PL1.  I'll run with this until deciding to overclock.  Based upon my test results, I'll need a stronger CPU cooling solution for that.

 

 

For the OP, I would recommend experimenting a bit more with your power limits, keeping in mind that the 10600K PL1, PL2, and Tau are 125, 182, and 56, respectively.  Not sure how the board will handle the 65W lower limit.  It may be setting it's own lower limit.  Try 100W and see if it works, then you'll know 65W is too low for that board (not to say too low for the chip, the board might just be doing its own thing on the lower limits).

Luckily I fully intended to ignore power limits/turbo duration anyway - I can see the frustration.

 

I am actually getting much better thermals and clocks with a manual overclock than with the 10900k at "stock".

 

ASUS apparently is the only one that follows the guidance.

 

It's unfortunate for those who want a plug and play experience.

[unclewebb]

6 minutes ago, Staypuft said:

it set my PL1 to the proper 125W, no idea what Tau was, but PL2 was 156W

Intel used to recommend that PL1 be set equal to the TDP (125W) and PL2 should be set 25% higher than PL1.

 

125W X 1.25 = 156W

 

That number makes sense but it is not the Intel recommended value. The datasheet shows this.

For the 8 core 10700K, Intel recommends PL1 = 125W, PL2 = 229W, Tau = 56 seconds

 

10th Generation Intel® Core™ Processors, Datasheet Volume 1 of 2

https://ark.intel.com/content/www/us/en/ark/products/199335/intel-core-i7-10700k-processor-16m-cache-up-to-5-10-ghz.html

 

 

9 minutes ago, Staypuft said:

Not happy with 4.1 so I kept going.

And that is why not many enthusiasts run their Intel CPUs at the default settings. This sacrifices a lot of performance, especially on Intel's CPUs that have a 65W TDP rating. I recommend buying a top tier cooler or AIO so you can run these powerful (power hungry) CPUs indefinitely at their full rated speed when necessary.

 

[Staypuft]

9 minutes ago, unclewebb said:

Intel used to recommend that PL1 be set equal to the TDP (125W) and PL2 should be set 25% higher than PL1.

 

125W X 1.25 = 156W

 

That number makes sense but it is not the Intel recommended value. The datasheet shows this.

For the 8 core 10700K, Intel recommends PL1 = 125W, PL2 = 229W, Tau = 56 seconds

 

10th Generation Intel® Core™ Processors, Datasheet Volume 1 of 2

https://ark.intel.com/content/www/us/en/ark/products/199335/intel-core-i7-10700k-processor-16m-cache-up-to-5-10-ghz.html

 

 

And that is why not many enthusiasts run their Intel CPUs at the default settings. This sacrifices a lot of performance, especially on Intel's CPUs that have a 65W TDP rating. I recommend buying a top tier cooler or AIO so you can run these powerful (power hungry) CPUs indefinitely at their full rated speed when necessary.

 

Thanks for the info.  I found a similar table but didn't know about the 1.25x legacy calculation.  Makes sense now.  At least we know what POR does.

 

For today, I settled at PL1=168, PL2=198, Tau=104.  Works great for now.

 

I thought my cooler choice would get me better thermals (Dark Rock Pro 4 with both fans maxed).  Looking at AIO solutions now.  Also considering a re-TIM of the DRP4 just to make sure it's correct.  Not my first rodeo with building a PC, but it is my first time using this cooler and the included be quiet! branded TIM.

[Mister Woof]

15 minutes ago, Staypuft said:

Thanks for the info.  I found a similar table but didn't know about the 1.25x legacy calculation.  Makes sense now.  I settled at PL1=168, PL2=198, Tau=104.  Works great for now.

 

At least we know what POR does now.

 

I thought my cooler choice would get me better thermals (Dark Rock Pro 4 with both fans maxed).  Looking at AIO solutions now.  Also considering a re-TIM of the DRP4 just to make sure it's correct.  Not my first rodeo with building a PC, but it is my first time using this cooler and the included be quiet! branded TIM.

Thank you for testing and finding out what POR means.

 

I looked all over the internet and cable up with nothing. I'm the manual was equally useless at explaining it.

9 minutes ago, Staypuft said:

As is usual for any motherboard manual...  

As for my 10900k, even with a Deepcool Assassin III, the thermal density even at only 1.26v drooped vcore overwhelmed the air cooler in more strenuous stress tests within a few minutes.

 

CPU intensive games had fan speeds ramp up all over the place to keep up with the instantaneous heat spikes. 

 

I reseated my cooler three times thinking it was that - but I came to find that with any amount of overclock and/or without an undervolt, these 14nm 10 core high boost chips just run away with heat quickly.

 

The 360 and the larger radiator helped to buy enough time as a buffer to heat soak before thermals got out of control for regular day to day CPU intensive tasks, and shorter 30-60 minute stress tests.

 

I keep seeing that there's not much difference between a large air cooler and a 360, and maybe that's true after hours of letting the water heat soak, but in my experience with these monolithic high core going chips, there's a huge difference in real world use (well unless you're uses are rendering)

 

By the time whatever cpu intensive thing I'm doing is done, the 360 is still just starting to warm up whereas the air cooler almost immediately ramps up to maximum and struggles.

 

The 360 buys you time.

[Staypuft]

1 minute ago, Mister Woof said:

Thank you for testing and finding out what POR means.

 

I looked all over the internet and cable up with nothing. I'm the manual was equally useless at explaining it.

As is usual for any motherboard manual...  

[unclewebb]

@Staypuft Keep an eye on how much voltage your CPU is getting. If you can reduce CPU voltage without losing stability, you can significantly reduce power consumption and heat. 

 

[Mister Woof]

7 minutes ago, Staypuft said:

As is usual for any motherboard manual...  

I edited my post with more information but I'm having a hard time fixing it on mobile

[Staypuft]

1 minute ago, unclewebb said:

@Staypuft Keep an eye on how much voltage your CPU is getting. If you can reduce CPU voltage without losing stability, you can significantly reduce power consumption and heat. 

 

Good tip.  I watched it a bit and really struggle to figure out which of the half dozen VIDs in HWinfo are the correct one.  My last test at current settings have me pulling VID=1.15 on most of the sensors.  I don't believe it (maybe the board is too new?).  CoreTemp tells me it's more like 1.25.  Either way, it seems really low.  My temps don't seem to line up. 

[unclewebb]

8 minutes ago, Staypuft said:

VID

VID is a meaningless number. Ignore that and ignore the VID that Core Temp reports.

 

VID is requested voltage. The important number is how much voltage is actually going to your processor. I like to lower the voltage to see what I can get away with without losing stability. On an Asus board, I have used the VCore number reported by HWiNFO to guide me. 

 

[Mister Woof]

For gigabyte boards, my understanding is the vr vout is the best sensor for vcore

[Staypuft]

3 minutes ago, Mister Woof said:

For gigabyte boards, my understanding is the vr vout is the best sensor for vcore

Great, I saw that one on the list and will check it out.  Step 1 is to find a good AIO

[Mister Woof]

4 minutes ago, Staypuft said:

Great, I saw that one on the list and will check it out.  Step 1 is to find a good AIO

They are all pretty similar outside of fans, rgb, and software.

 

I only have the one I have because of best buy credit and gift cards. It's expensive; $190, but it also comes with the rgb Commander controller. I wouldn't necessarily have got it if I wasn't limited to best buy. It works well though outside of price.

 

I had the EVGA one which actually works pretty well, and it's cheap, but the fans are terrible and the software doesn't really work.

 

The top AIO I see recommended is the Arctic Liquid Freezer II 280/360. Its well reviewed and also has a little VRM fan on it that supposed to help 

[porina]

I think every Intel mobo I've had since Haswell era has honoured user set power limits, with the single exception of a Gigabyte Z97 board. Is this just a Gigabyte thing? No problems with Asus, Asrock and maybe MSI in that area, but thinking more I might not have tried it on MSI since their boards have a load of different problems.

 

What I had done in the past on other systems was to dial in PL1=PL2=whatever I wanted and I could test performance at a fixed power limit. I guess this is not happening easily on a Gigabyte board. I really don't want to go the software route to do this, although XTU I guess is an option.

[unclewebb]

4 hours ago, porina said:

although XTU I guess is an option.

Intel XTU does not give you access to the Clamp option within the power limit register so XTU is not an option for this problem.

 

What the Clamp option does is it forces the CPU to slow down below the base frequency if necessary so it does not exceed the turbo power limit that you have set.

 

When Clamp is not checked, the CPU will only slow down to the base frequency. Power consumption might still be above the power limit you have set when running at the base frequency but even so, it will not slow down any further. 

[porina]

11 minutes ago, unclewebb said:

When Clamp is not checked, the CPU will only slow down to the base frequency. Power consumption might still be above the power limit you have set when running at the base frequency but even so, it will not slow down any further. 

Hmm... I don't recall now if it was at base frequency or lower. Certainly the observed power of 89W is close to the TDP-down value of 95W. Do some mobos/manufacturers default to clamp-on and Gigabyte don't? Again, this is the first time I've seen this happen.

[unclewebb]

4 hours ago, porina said:

Do some mobos/manufacturers default to clamp-on and Gigabyte don't?

Certainly a possibility. I have not tested enough motherboards to know.

 

TDP-down does not seem to lower the power limit on my Asus board. The regular turbo power limits are still in control.  

 

To test TDP-down, I disable Turbo Boost and watch the multiplier. The 10850K has a default 36 base multiplier. In TDP-down mode, this drops to 33. A slower CPU speed reduces power consumption so that seems to be how it achieves the lower TDP power limit.

 

 

 

[Staypuft]

On 3/27/2021 at 12:54 AM, unclewebb said:

VID is a meaningless number. Ignore that and ignore the VID that Core Temp reports.

 

VID is requested voltage. The important number is how much voltage is actually going to your processor. I like to lower the voltage to see what I can get away with without losing stability. On an Asus board, I have used the VCore number reported by HWiNFO to guide me. 

 

After a few days messing with it, looks like Vcore ~1.26V typical with the set power limit PL2 of 198W, drops to 1.18V under constant load at 168W.  Auto voltage.  

[porina]

Dropped in a 11700k yesterday and now starting to play with power limiting on that, which could use it! Running unlimited AVX-512 code I've seen in excess of 220W. Just set a 166W power limit (random number I picked) and it seems to be holding it. None of the funny business I saw with the 10600k. Haven't tried lower yet, maybe that's where unexpected behaviour starts.

 

edit: 

 

New problem but doesn't seem worth starting a new thread on it. Power limit seems to work, at least down to 142W as my next test level.

 

I also wanted to try AVX offsets, but setting these did nothing to load clocks. Still boosted to same level as before. I was hoping reducing clocks would be a more predictable way to limit temperatures than a power limit, since if few cores are active they still get a large power budget.

[Staypuft]

Never tried the AVX offset.

 

Another way to get your power down is to drop the voltage manually.  I ended up bumping power limit (PL1 and PL2) to 229W and then playing with manual voltage until I achieved stable operation in multiple scenarios (16 thread P95 small FFT and 2 thread P95 small FFT).  The thing is with these CPUs, if you run a full 16 thread load the clock drops to it's preset value (4.7GHz on my 10700K).  When running two threads it will bounce around between 5.1GHZ and 4.9GHz, then immediately (within 5 minutes) crash with a low voltage that runs 24/7 stable on 16 threads.  I played around raising manual Vcore and enabling values of LLC to reach the lowest voltages while being stable at full load and at low load.  My end result was Vcore=1.200V w/ LLC=Low.  Temps now never go above 80C on P95 small FFT.  Power consumption hovers around 185W with the 16 thread P95 small FFT load.