What controls Turbo Core in Xeons?

Page 12 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

knopflerbruce

Junior Member
Mar 5, 2017
15
1
41
could you please share the modded bios for asrock x99 oc formula, I have the same motherboard. I am new to this and unable to edit. Thank you

I haven't modded anything yet due to a lack of xeon v3 cpus. There's a 2686 on the way here from Japan, will test stuff as soon as it arrives.
 
  • Like
Reactions: Cata40

ajschot

Member
Feb 16, 2017
65
2
16
Hi all,

2 things
1) Were to find other microcode's? I want to try 0x38 maybe it gives me better results.


2)Is there a way to get this turbo working on Hackintosh?
I tried a lot of times but get it only stable on basic 2Ghz with some poweremanagement driver. other thing do you guys use also microcode in Ubuntu? Is there also an app for?
 

Zladimir

Member
Apr 14, 2011
34
3
71
Is there anything hard-coded in there that might need to be changed for the 2696? The max boost on this chip is 3.8, not 3.6, but I'd assume any assumptions about 2699 would just result in it never getting to 3.8 on single core operations...

As each cpu unit type has an unique id there shouldnt be a switch in system. However the Turbos are written by Dufus and I dont know how he wrote them for 2696v3. You can check that by using ru.efi after you injected dufus v3.efi, as you can read out your turbo in cpu msrs than.
By myself I wish Dufus gave us a little bit more information about how he created or modify v3.efi, that we could make our own adaptions.
Also there is a way to fake cpuid's. But Im not fond of this either, but would love to try it if someone would know more. Clover FakeCPUID for expample uses a kind of bootloader to fake cpuid (which even results in different clocks) before the general system starts. Mostly its used for increasing support on OS X, but clover also supports Win10.

Still Im struggle to increase my cpu ratio above 28x. Doesnt matter what I try my best performance I have clearly without any microcode and only injected v3.efi. I tried now microcodes 27, 29, 2D, 36, 38 and 39. Its at least all I found. Would like to try any pre-27 microcodes, but don't know where to get them. Anyway they all performed by me exactly the same.

Quite confusing for me that we all got the same MB and as I could see from pictures have even the same stepping of our 2686v3, but got such different results. Thats why I still believe that there is something else we didnt noticed which has influence. Sadly Lucien didnt share yet his setup configs (voltage, c states, turbos, if he used dufus efi or an own, etc.)

1) Were to find other microcode's? I want to try 0x38 maybe it gives me better results.

Intel download center is your source, but as I said, I didnt found any pre-29 mc, but the others I mentioned are there definately.
 

custom90gt

Member
Feb 9, 2017
38
7
41
Just had a quick question for you guys.

I have my 2683 v3 running at 3ghz but my Cinemark R15 scores are only 1850 (stock I was getting 1720ish). Any input for me there? I can see it's running at 3ghz while running the test according to CPU-Z. I have an asrock x99m motherboard, running V3.efi (not sure where people see V3_1.efi), and 27 via the VMWare driver.

Appreciate all the help!
 

lucien_br

Member
Feb 13, 2017
37
7
41
How you managed to get ratio of 29x ? Did you used Dufus v3.efi ? You reprogrammed MSR registers ? Or did you played with vcore or whatever ? Thanks for any clarification!
Knowing what is different may give us a clue how to influence.


What does these options do exactly ?
Hi! Simply I used v3.efi, ucode 0x27 or 0x39 (make no difference), Bios Modded:
https://drive.google.com/file/d/0B_ZQtaxTWt-QdGNGSnA5U0UtMkU/view?usp=sharing
, unpark all cores, speed-step off, vcore offset +0,02, BClk 103,2, C-States on (default) and nothing special configuration.
Try any thing above!
With HT on x29 all 18 cores and HT off x35 all 18 core. See my video with BF1 and general tests for more details and CPU-Z prints!
https://www.youtube.com/watch?v=Faae3_Tdtfw
 
Last edited:

lucien_br

Member
Feb 13, 2017
37
7
41
Someone with an 2686v3 could confirm if the experience this issues?
So it goes like this , i got an x99 TAICHI (bios 1.4 + v3.efi) and some problems persists:
- I have only 1/2 successful boots to windows, half of the time it freezes
- Sometimes CPU overheats and goes to 90degrees celsius (normally on full load it does not exceed 55*C) and have to wait 1-2min to turn on PC again (its very rare it happends once 2weeks)
- Mosfet area extremely hot when using V3.efi (80*C or more / some speedfan reports on some AUXTIN1/2/3 over 100*C ,i put a cooler over the mosfets and seems to be cooler now , AUXTIN is not related to mosfet)
- When it wakes up from sleep the PC freezes (i did not disable any power-state from bios, i remember reading some C State needs to go off , this might be it , alternatively i disabled Sleep/Turn off monitor in windows)

Tried:
OS: windows 7/10
Microcode 27,39 and latest microcode from windows (mcgenuine_intel.dll)

Event viewer does not report anything on crash, just sudden system stop.
I had this problems from the beginning of doing this, any ideas what can cause this?
I do same thing with you. Simply I used v3.efi (Dufus), ucode 0x27 or 0x39 (make no difference), Bios Modded:
https://drive.google.com/file/d/0B_ZQtaxTWt-QdGNGSnA5U0UtMkU/view?usp=sharing
, unpark all cores, speed-step off, vcore offset +0,02, BClk 103,2, C-States on (default), disable secure boot, and nothing special configuration.
Try any thing above!
With HT on x29 all 18 cores and HT off x35 all 18 core. See my video with BF1 and general tests for more details and CPU-Z prints!
https://www.youtube.com/watch?v=Faae3_Tdtfw
Edit: I had 13ºC increase. HT on, x29 all core, full load with Prime95 hits 66ºC. I thing it is normal.
No freezes and all ok!
 
Last edited:

lucien_br

Member
Feb 13, 2017
37
7
41
Hey @lucien_br
how you managed to get such constant Turbos and high clocks? How your cpu clocks in BF1 is awesome! Also at minute 6:08 your clockrate is awesome. Can you explain please what you did differently ?
Im especially interested in the microcode you use.
Please, share ! :D


As here are so many Taichi users, did someone of you successfully installed nvme raid0 ? I injected myself an updated IRST driver and flashed, but still cant get a nvme raid0 to setup. Any advice ? In the win-raid forum I didnt found any who has a Taichi board, too.

What C-states you guys have activated on Taichi board ?
I do same thing with you. Simply I used v3.efi (Dufus), ucode 0x27 or 0x39 (make no difference), Bios Modded:
https://drive.google.com/file/d/0B_ZQtaxTWt-QdGNGSnA5U0UtMkU/view?usp=sharing
, unpark all cores, speed-step off, vcore offset +0,02, BClk 103,2, C-States on (default), disable secure boot, and nothing special configuration.
Try any thing above!
With HT on x29 all 18 cores and HT off x35 all 18 core.
I had 13ºC increase. HT on, x29 all core, full load with Prime95 hits 66ºC. I thing it is normal.
No freezes and all ok!
Raid 0 in bios config only on ports 0 and 5. See the board manual.
 

lucien_br

Member
Feb 13, 2017
37
7
41
Mine is running very fine on Windows 10 also never loaded microcode with the VMware tool just did the bios mod and loading V3_1.EFI (don't know what is the difference between the normal an the 1 version) but it worked in Windows 10 and is very stable, only having problems with OS X but I think it is an OS X thing so I have to find out what is going wrong, it is always running @ 2GHz then it is stable but if it runs from 1,2 - 3Ghz it freezes after 4minutes

Otherthing, where to find microcode? 0x038
Where can I get v3_1.efi to test here?
 
  • Like
Reactions: Murilo900

lucien_br

Member
Feb 13, 2017
37
7
41
As each cpu unit type has an unique id there shouldnt be a switch in system. However the Turbos are written by Dufus and I dont know how he wrote them for 2696v3. You can check that by using ru.efi after you injected dufus v3.efi, as you can read out your turbo in cpu msrs than.
By myself I wish Dufus gave us a little bit more information about how he created or modify v3.efi, that we could make our own adaptions.
Also there is a way to fake cpuid's. But Im not fond of this either, but would love to try it if someone would know more. Clover FakeCPUID for expample uses a kind of bootloader to fake cpuid (which even results in different clocks) before the general system starts. Mostly its used for increasing support on OS X, but clover also supports Win10.

Still Im struggle to increase my cpu ratio above 28x. Doesnt matter what I try my best performance I have clearly without any microcode and only injected v3.efi. I tried now microcodes 27, 29, 2D, 36, 38 and 39. Its at least all I found. Would like to try any pre-27 microcodes, but don't know where to get them. Anyway they all performed by me exactly the same.

Quite confusing for me that we all got the same MB and as I could see from pictures have even the same stepping of our 2686v3, but got such different results. Thats why I still believe that there is something else we didnt noticed which has influence. Sadly Lucien didnt share yet his setup configs (voltage, c states, turbos, if he used dufus efi or an own, etc.)



Intel download center is your source, but as I said, I didnt found any pre-29 mc, but the others I mentioned are there definately.
Tomorrow I post my BIOS config images, but I'm warning you that there is nothing special and I'm using v3.efi (Dufus) and ucode 27 and 39.
I do same thing with you. Simply I used v3.efi (Dufus), ucode 0x27 or 0x39 (make no difference), Bios Modded:
https://drive.google.com/file/d/0B_ZQtaxTWt-QdGNGSnA5U0UtMkU/view?usp=sharing
, unpark all cores, speed-step off, vcore offset +0,02, BClk 103,2, C-States on (default), disable secure boot, and nothing special configuration.
Try any thing above!
With HT on x29 all 18 cores and HT off x35 all 18 core. See my video with BF1 and general tests for more details and CPU-Z + HWinfo screens!
https://www.youtube.com/watch?v=Faae3_Tdtfw
I had 13ºC increase. HT on, x29 all core, full load with Prime95 hits 66ºC. I thing it is normal.
No freezes and all ok!
 
Last edited:

cekim

Member
Mar 6, 2017
87
19
41
Some updates and surprises (X99Ex4 + 2696v3 + linux + v3.efi):
1. Some of my issue appears to be good-old fashioned overclocking - needs more voltage - added 0.020 to VCCin and Vcore and it is _more_stable

2. I am able to see 3100MHz now on 18 cores with HT disabled (running stressapp here):
Every 1.0s: grep -i mhz /proc/cpuinfo Wed Mar 8 00:23:29 2017
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968
cpu MHz : 3099.968

3. There is some fundamental problem with this board and stripped BIOS - it is near 50/50 that shutting down/rebooting results in BIOS code 79 which is unrecoverable in any manner that I've found other than CMOS clear.

- This happens with BOTH 2696v3 chips I've tried now
EDIT: ok, I've now seen this with 2 chips and the stock bios... This board might have something wrong with it...
EDIT: This 79 code even survives switching to bios B, booting successfully, switching back to A (now I see 79 again and only a clear of A fixes it... )
 
Last edited:

katlen

Junior Member
Jan 27, 2017
10
7
51
So I managed to fix my issues with x99 taichi that system would freeze/not boot all times, I disabled BCLK speed spectrum , lock the bclk at 100mhz, disabled powerstate c3, and lock the ram to 2133MHZ (i use ddr4 ecc ram) and it's all stable now and works perfectly 100% successful boots to windows.
Also without HT disabled i get 28x when cpu usage is 100% and 35x when usage on all cores is lower than 50% which is good improvement for games/single thread programs. 1600/22000 single/multi CPU-z benchmark.

A note on bclk overclocking/downclocking:
In my experience with windows7/10 i know one common issue with bclk decreasing/increasing, if bclk is higher or lower than 100MHZ your audio will get desynced or scrambled (atleast via hdmi with nvidia gpu) (i know this because i had hyper-v installed which downclocked bclk by 3mhz and when i overclock to 105MHZ i get same result, its common issues on other processors too i seen it on my friends i7 4770k with hyper-v installed)
 

cekim

Member
Mar 6, 2017
87
19
41
Ok, i've managed to do it. If anyone needs a modified asrock extreme 4 bios, here it is: https://www.sendspace.com/file/9jm96i
Thanks for such a great thread and all of your instructions guys!
My X99 E4 refuses to recognize this image as a bios - "no image file detected".
The diff shows a lot of changes from the 3.60 bios on the web. Can you describe what you did to this and how you load it?
 

mtnew10gm8

Junior Member
Mar 6, 2017
8
2
41
No I'm on Win10x64. Microcode and "howto" I've taken here. Just look at post #39.

Have you tried RU to program MSR? Attempted just to view (no modification) - for some reason the latest version of ru.efi hangs when I select CPU1
 

TLN

Junior Member
Jan 27, 2017
21
6
41
Have you tried RU to program MSR? Attempted just to view (no modification) - for some reason the latest version of ru.efi hangs when I select CPU1
I wanna look into that. Because of dual cpu mobos - people reporting they got only one CPU unlocked.
Wish Dufus gave up some hints on v3.EFI or even source. As I understand it's far from rocket science, but if you're not doing that before looks like it.
I'll try RU.EFI on my board today, will see what I got. I've already done my own EFI application(helloworld) but haven't tested it yet.
 

TLN

Junior Member
Jan 27, 2017
21
6
41
I've made my own V3x2.efi
https://ufile.io/541182
( Unlock boost for E5-26xx-v3 multiprocessor systems, based on original V3.efi ).

PS: It is my first experience with uefi programming... it works, on my PC at least... but why I have got x28 on the first and x30 on the second CPU... I'll check tomorrow.

Could you please share some details on that?
I was looking into EFI programming past couple of days. Made my Hello world.efi lol.
Wonder why your efi is 8KB and original one is about 1KB. Is that ASM vs .c difference? From my understanding we're changing several registers only, so it should be pretty small app.
 

Welsper

Member
Mar 5, 2017
25
18
81
Yes. I've written it on C so it larger than original that was made on ASM (I think so, I don't have source).
Also my current code contains really ugly (unoptimized) translation ASM to C, where several simple ASM operations with registers become several function calls.

Source code:
https://ufile.io/eac1d
 
Last edited:
  • Like
Reactions: cekim

mtnew10gm8

Junior Member
Mar 6, 2017
8
2
41
Yes. I've written it on C so it larger than original that was made on ASM (I think so, I don't have source).
Also my current code contains really ugly (unoptimized) translation ASM to C, where several simple ASM operation with registers become several function call.

Source code:
https://ufile.io/eac1d

It works perfectly - 30x on both CPUs
CPU-Z: Single: 1361 and Multi: 40178
Thank you very much.

Aside from benching, my "real-world" app (vray rendering) ends up at 29x across all cores at full load.

Also NOD32 keeps quarantining the v3x2.efi "because of a threat present" :)
had to add exclusion manually.
 
Last edited: