What controls Turbo Core in Xeons?

Page 112 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

topmysteries5

Member
Jan 31, 2019
179
17
61
My motherboard is Huanan X99-8M, and the CPU is also 2683V3.
Using 70 50 50 1.8v FFS, running Cinebench R20 up to only 2.9Ghz

X99-AD3 -> 4 DIMM DDR3
X99-8M -> 2 DIMM DDR4

View attachment 8591View attachment 8592
You're running CB R20, it uses AVX2. Try running R15, 2683v3 gets ~2200CB @ 3ghz on all core. I get ~5300CB in R15 with dual 2686v3.
These motherboards are good quality. I'm still using their X79 delux from 2017. Almost 14-15 hours/day as render rig.
 
Last edited:

pamelafx

Junior Member
Jul 10, 2019
6
0
11
Hi,

It has a microcode fix - does that decrease performance?

In order for this hack to work, the microcode has to be removed for Haswell E. When Moff or someone makes you the BIOS, there will literally be *no* microcode for Haswell in there. Then you can put whatever microcode you want after you boot the OS. In fact, it seems like Win 10 will put some fresh microcode on it's own:
support.microsoft.com/en-us/help/4465065/kb4465065-intel-microcode-updates

You should read page 3 of this thread carefully. All the best understanding of this hack starts on that page!!!
 
Last edited:

pamelafx

Junior Member
Jul 10, 2019
6
0
11
Hi @MOF

I injected the ffs driver from you post #1955 (https://forums.anandtech.com/threads/what-controls-turbo-core-in-xeons.2496647/page-79#post-39186745) in this thread into my BIOS, and it's working, but not quite at full speed.

The efi driver that worked best for me was FreeCableGuy's v3x4-0.10b-i306f2-rc9_90_50_50.efi - this one gives me the expected 3.4GHz under Prime95, and 3.5GHz under real world render loads...

Do you by chance have any other versions of your ffs that would give me the 90/50/50 that lets my Asus X99-A and E5-2690 v3 run at full speed and never have to worry about re-installing the efi driver if my Windows need a re-install?

Thanks!
 

MOF

Member
Jul 31, 2017
118
33
101
  • V3_MOF_505050.efi
  • V3_MOF_505050.ffs
  • V3_MOF_605050.efi
  • V3_MOF_605050.ffs
  • V3_MOF_705050.efi
  • V3_MOF_705050.ffs
  • V3_MOF_805050.efi
  • V3_MOF_805050.ffs
  • V3_MOF_905050.efi
  • V3_MOF_905050.ffs

If you want you can load these efi drivers directly from the efi shell for testing.

use "load" command
load fs5:\V3_MOF_505050.efi (loads efi driver)
fs1:\EFI\Boot\bootx64 (loads windows)
something like this, you get it. :)

https://drive.google.com/file/d/1aBzuUedZkrQKlrnLHVekXhSEGfAx1Rb3/view?usp=sharing
 
Last edited:

jordanpchome

Junior Member
May 28, 2017
6
0
36
  • V3_MOF_505050.efi
  • V3_MOF_505050.ffs
  • V3_MOF_605050.efi
  • V3_MOF_605050.ffs
  • V3_MOF_705050.efi
  • V3_MOF_705050.ffs
  • V3_MOF_805050.efi
  • V3_MOF_805050.ffs
  • V3_MOF_905050.efi
  • V3_MOF_905050.ffs

If you want you can load these efi drivers directly from the efi shell for testing.

use "load" command
load fs5:\V3_MOF_505050.efi (loads efi driver)
fs1:\EFI\Boot\bootx64 (loads windows)
something like this, you get it. :)

https://drive.google.com/file/d/1aBzuUedZkrQKlrnLHVekXhSEGfAx1Rb3/view?usp=sharing

Are these FFS and EFI files VCC 1.82Volt Locked?
 

pamelafx

Junior Member
Jul 10, 2019
6
0
11
Thank you MOF, I can confirm that V3_MOF_905050.ffs works beautifully on an Asus X99-A motherboard, so it would probably work nice for that whole family (X99-A II, X99-pro, X99-deluxe, etc).

One thing I will say for anyone else with an Asus X99 board (hope this helps someone else)...

You MUST set in your BIOS:

Ai Overclock Tuner - MANUAL

Asus MultiCore Enhancement - DISABLED

CPU Core Ratio - SYNC ALL CORES

If you don't do this, then shortly after boot your OS (Windows), the Xeon will revert to STOCK speed... and you don't want that !!
 

MOF

Member
Jul 31, 2017
118
33
101
Thank you MOF, I can confirm that V3_MOF_905050.ffs works beautifully on an Asus X99-A motherboard, so it would probably work nice for that whole family (X99-A II, X99-pro, X99-deluxe, etc).

Good but -90mv core offset still too much. Heavy test doesn't show real stability, cpu can pass any stress test with -90mv but can be freeze while idle or single core under load . Because core voltage may drop the below 0.6 volt when the processor is idle.
 

Micrornd

Golden Member
Mar 2, 2013
1,279
178
106
Good but -90mv core offset still too much. Heavy test doesn't show real stability, cpu can pass any stress test with -90mv but can be freeze while idle or single core under load . Because core voltage may drop the below 0.6 volt when the processor is idle.
I finally got me setup working :)
And as you say 90-50-50 is not stable for me either, but I did find something interesting in trying it.

With 90-50-50 without a microcode Cinebench R15 runs all cores on my pair of 2696's @ 3.5ghz.
With 90-50-50 without a microcode
Corona 1.3 (AVX) runs the pair @ 3.4ghz, just a 1ghz differential.
All other offsets without microcode have a 1.25 to 2 ghz differential.
With microcode (I use 43), the differential is a minimum of 2ghz and a maximum 3ghz.
From past post here I didn't think a differential of 1ghz was possible.

BTW - my 2696s run @ 3.5 with almost any offset without microcode, and 3.4 with microcode.
 

MOF

Member
Jul 31, 2017
118
33
101
Without microcode you can get higher frequencies but without microcode higher frequencies doesn't mean higher performance in all circumstances. Some application doesn't work well without mcode.
 
Last edited:

Micrornd

Golden Member
Mar 2, 2013
1,279
178
106
Without microcode you can get higher frequencies but without microcode higher frequencies doesn't mean higher performance in all circumstances. Some application doesn't work well without mcode.
That's not what I'm seeing.
Performance is higher in all apps I've checked (mostly video encoding and image manipulation) and in all benchmarks I've run.
So it appears that the apps I use work well without microcode.
Just luck :)
I'm currently running your old 80-30-50_vcc with no microcode.
Could I talk you into making a 70-30-50_vcc and 90-30-50_vcc to test?
 

MOF

Member
Jul 31, 2017
118
33
101
Could I talk you into making a 70-30-50_vcc and 90-30-50_vcc to test?

Yea sure but it no sense. Actually cache and SA offset doesn't change anything, we don't have any clue that the cache or SA voltage makes a difference.
 

Micrornd

Golden Member
Mar 2, 2013
1,279
178
106
I don't know what to tell you.
Maybe it's my board (which no one else seems to use) or maybe my CPUs.
But I can use your 80-30-50_vcc with or without microcode and it's perfectly stable.
Your 80-50-50_vcc with or without microcode is not stable for me (neither was 90-50-50_vcc but performance jumped while it worked).
That's why I was curious to try those other variations to check performance and stability.
Also what program do you use to edit or make .efl files like those, I'm interested in learning how to.
 

NoturnoJpop

Junior Member
Jul 25, 2019
4
0
6
Hello could someone help me modify the GA X99 UD3 rev 1.0 F23c bios to overclock the Xeon 2650 -v3 please?
Thank you !
 

KenSoftTH

Junior Member
Jun 29, 2018
9
0
11
I only managed to downvolt to -0.060 on Asus X99 Pro USB3.1 with Xeon E5-2686 v3 ES (not QS) more than that and it's not stable. My clock only goes up to 2.7 GHz on all core and I can only overclock BCLK to 103. Am I getting bad chip or just not enough tuning? Been using the mod for two years now.