What controls Turbo Core in Xeons?

Page 63 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Methanoid

Junior Member
Mar 16, 2017
13
3
81

Thanks for posting these.. would you be able to mod the Gigabyte X99P-SLI bios too, please? https://www.gigabyte.com/Motherboard/GA-X99P-SLI-rev-10#support-dl (latest is F24)
 

custom90gt

Member
Feb 9, 2017
38
7
41
Since your using throttlestop that should show why with limit reasons.

You're right, I checked the throttling reason and with 18 cores enabled it says that it's throttling due to EDP current and Max Turbo. I wonder if there are any ways around that.
 

kjboughton

Senior member
Dec 19, 2007
330
118
116
Maybe VRM voltage sags a little when dual CPUs are used.

Shouldn't matter as the VRM input is further regulated downstream.
The loadline shouldn't change with C3/C6 disabled/enabled either.
On my system, I have no problem with C6 (non-retention) enabled. C3 is a no-go. Not sure it's supported by the E5 Xeons. Also, there's a difference between core C6 and package C6. Package (retention) C6 on these system seems to be problematic, core not so much.
 

kjboughton

Senior member
Dec 19, 2007
330
118
116
You're right, I checked the throttling reason and with 18 cores enabled it says that it's throttling due to EDP current and Max Turbo. I wonder if there are any ways around that.

A: Increase Core (domain 0) negative dynamic voltage offset

One of my E5-2696 v3 scale as such:
0 mV = 3.2GHz max all-core turbo in heavy workloads (including AVX2)
-50 mV = 3.3GHz " " "
-70 mV = 3.4GHz " " "
-100 mV = 3.5GHz " " "

The other (the better of the two):
0 mV = 3.3GHz max all-core turbo in heavy workloads (including AVX2)
-20 mV = 3.4GHz " " "
-50 mV = 3.5GHz " " "
-90 mV = 3.6GHz " " "

My theory is that on init, the PCU reads the programmed VIDs (not the actual Vcore, but the programmed VIDs), and crosses this with the current (internally measured, I believe, not linked to VIN from the VRM) to calculate power. So, lower VID -> less power usage calculated -> higher allowed multi to consume the available power margin, as controlled by internal ARM PCU.
 
  • Like
Reactions: custom90gt

xcimo

Junior Member
Oct 19, 2009
3
0
61
I've gone through many pages on this thread, but found no step by step guide as to what to do / set.

Maybe someone could help me a bit

I have a ASUS Z10PE-D16 WS and two Xeon E5-2678 v3 running Windows Server 2016 x64 fully patched.

E5_2678v3.png


I've found the following post with a modded Z10PE-D16 bios:

Okay, so I got it running, I'm using v3x2_50_39_vcc
Dual 2683V3 C3&C6 Disabled, Cinebench score 3700, but I had to use High-Powered-Scheme, in balanced the score was "only" 3400, which is still resonable.

Is there a efi file, for better overall performances in balanced mode?

Also, I would like to make my own contribution. I searched the 50 post for all the usable efi and bios files for z10ped16ws that I could find.

I have them hosted here https://skydrev.org/index.php/s/5rvI5pVwddsXLZT
password is anandtech.

Maybe I should clean that folder, so it's more user friendly.

It would be cool, if the original modder of Z10PE-D16 WS made a new modded bios without microcodes from the newest available bios 3407.

It contain BIOS 3307.

Which step do I take from there?

Is there a newer modded bios than 3307 somewhere?

I have read about micro-code, disabling C3 and C6 state in the bios and also about renaming dll in windows (or uninstalling KB..)

Do I need all of that? Or just flash the BIOS with ASUS ez-flash and disable C3/C6 will be enough?

Thanks you!!
 

junkim

Junior Member
May 12, 2017
19
0
11

Hallo CANONKONG

Would you please mod the new bios for ASUS X99-E WS/USB 3.1 (Version 3601) just removing MC

http://dlcdnet.asus.com/pub/ASUS/mb/Socket2011-R3/X99-E_WS_USB31/BIOS/X99-E-WS-USB31-ASUS-3601.zip

thanks
 

Dufus

Senior member
Sep 20, 2010
675
119
101
You're right, I checked the throttling reason and with 18 cores enabled it says that it's throttling due to EDP current and Max Turbo. I wonder if there are any ways around that.

What current limit are you running? I only see EDP with AVX2 and then drop just one bin. Unfortunately my chip's top bin is only 30x so might be the reason I'm not seeing what others are. Until I can try with one of those then I'm pretty much stuck.
 

kjboughton

Senior member
Dec 19, 2007
330
118
116
Didnt work. :(
IWyGaxH.jpg
Ok. This is good. I can fix this. This was the first run on a single-CPU system. Stand by...

EDIT: DONE... go grab the newest release candidate (v3x4.efi) from the Release section at the GitHub page and replace existing file.
 
Last edited:
  • Like
Reactions: MOF

MOF

Member
Jul 31, 2017
118
33
101
Ok. This is good. I can fix this. This was the first run on a single-CPU system. Stand by...

EDIT: DONE... go grab the newest release candidate (v3x4.efi) from the Release section at the GitHub page and replace existing file.

It works very well

f4RnRC0.jpg
 
  • Like
Reactions: kjboughton

xcimo

Junior Member
Oct 19, 2009
3
0
61
Question: If you flash a BIOS without microcode update. Will windows 10 apply the missing update every time it boot? Otherwise it seem there could be important bugfix missing (other than the bug in xeon v3 that is used here).

I'm just a bit worried about that part.
 

Cata40

Member
Mar 2, 2017
156
6
81
Question: If you flash a BIOS without microcode update. Will windows 10 apply the missing update every time it boot? Otherwise it seem there could be important bugfix missing (other than the bug in xeon v3 that is used here).

I'm just a bit worried about that part.
only if you update windows, it will update microcodes.
 

xcimo

Junior Member
Oct 19, 2009
3
0
61
only if you update windows, it will update microcodes.

So everytime you patch you need to remove the microcode somehow? Or do they only apply in windows (bugfix the cpu) but will be not overwrite for the cpu max hyper boot on all core?

I should have been more clear, I am wondering if I will be missing important other cpu bugfix (microcode update). I patch windows regularly. In this case it's Windows Server 2016.

Can you elaborate a bit more?
 
Last edited:

KW8cSgF53G

Junior Member
Oct 4, 2017
9
3
51
Ok. This is good. I can fix this. This was the first run on a single-CPU system. Stand by...

EDIT: DONE... go grab the newest release candidate (v3x4.efi) from the Release section at the GitHub page and replace existing file.

Could you build a full turbo&downvoltages .efi driver for me please, i really wanna try this!
(I messing its required IDE for a day, and i still getting errors when compiling it...)

Dual (way) cpus, and offset -50mv which apply to both IA, LLC and SA voltage. thanks!
(I knew the potenial risk of system unstable!)

I am now using C_Payne's .asm full turbo&downvoltages .efi driver,
but i want to lower the temperature more on idle status or full loading!
(He's now got IA and LLC downvoltage working)

(Sorry for my bad English.)
 

Cata40

Member
Mar 2, 2017
156
6
81
So everytime you patch you need to remove the microcode somehow? Or do they only apply in windows (bugfix the cpu) but will be not overwrite for the cpu max hyper boot on all core?

I should have been more clear, I am wondering if I will be missing important other cpu bugfix (microcode update). I patch windows regularly. In this case it's Windows Server 2016.

Can you elaborate a bit more?
there is nothing to elaborate on
only if you update to windows will reappear
otherwise no
 

kjboughton

Senior member
Dec 19, 2007
330
118
116
The requirement for programming to be successful is that NO MICROCODE PATCH BE LOADED at time of programming (during boot). As Windows is booting, it will auto-load Haswell-E/EP microcode patch revision 0x36. Personally, I prefer 0x1F for a number of reason which I won't go into right now. It is possible to defeat the Windows auto-load feature and load whatever you want, or nothing at all. MAJOR updates(i.e. new major build releases, RS1, RS2, etc.) would require you to re-apply the work-around to prevent Windows for auto-loading microcode. MINOR updates (hot fixes, security patches, etc.) will not require the re-application of the workaround. The workaround takes about 10 seconds and a re-boot to apply. Again, microcode update present, no microcode update present... Windows works either way. Just make sure there is no microcode patch loaded at POST (requires you to remove the microcode built into the BIOS).
 
  • Like
Reactions: glitchman

Cata40

Member
Mar 2, 2017
156
6
81
The requirement for programming to be successful is that NO MICROCODE PATCH BE LOADED at time of programming (during boot). As Windows is booting, it will auto-load Haswell-E/EP microcode patch revision 0x36. Personally, I prefer 0x1F for a number of reason which I won't go into right now. It is possible to defeat the Windows auto-load feature and load whatever you want, or nothing at all. MAJOR updates(i.e. new major build releases, RS1, RS2, etc.) would require you to re-apply the work-around to prevent Windows for auto-loading microcode. MINOR updates (hot fixes, security patches, etc.) will not require the re-application of the workaround. The workaround takes about 10 seconds and a re-boot to apply. Again, microcode update present, no microcode update present... Windows works either way. Just make sure there is no microcode patch loaded at POST (requires you to remove the microcode built into the BIOS).
can you explain how to do to apply new drivers efi in windows or bios, the new method?
If the microcode is deleted in bios with modded bios, it is possibility to load at boot, again?
 
Last edited:

MOF

Member
Jul 31, 2017
118
33
101
?
Same as before.

You can install any efi driver with efi shell and "bcfg driver add ......" command.
Also you can install any microcode with VMWare Microcode Updater
 

Cata40

Member
Mar 2, 2017
156
6
81
?
Same as before.

You can install any efi driver with efi shell and "bcfg driver add ......" command.
Also you can install any microcode with VMWare Microcode Updater
Are you referring to the new kjboughton driver?
So, from what I understand, it can not be inserted into bios so far, right?
Edit: sorry for my inattention
with new efi driver of kjboughton i have:
with microcode i have 27, 28 or 29 all core
whithout microcode 31 max 32 at all core
 
Last edited:

MOF

Member
Jul 31, 2017
118
33
101
Do not rush. First compiled version of v3x4 is a not much different other efi drivers at this time. Thats mean you cant get better result.

Inserting the driver into bios is must be a last thing to do. If everything works perfect an you get good results then you may inject efi driver to your bios.