What controls Turbo Core in Xeons?

[y723157510]

Hello there
I unlocked successfully
Also added bcfg v3.efi

But restart the default frequency

Cpu e5 2683 v3 r2 official version
Motherboard asus x99a usb3.1

Can I help me to manually load the problem every time?

 

[cekim]

Hello there
I unlocked successfully
Also added bcfg v3.efi

Can I help me to manually load the problem every time?

If you performed the bcfg driver add 0 fs<N>:\<path>\V3.EFI "v3 Turbo" step (where <N> is your drive and <path> is your path), then it should be loading (and will print a little message in the upper left) already. I copied the V3.EFI to my windows/linux EFI boot partition (on the main drive - fs0) and then called bcfg driver add with that location (rather than the thumb drive).

 

[Dingmel]

I went through the steps, and first time, it was fine. managed to get sustained 30x multipliers.

I then rebooted and went back to the EFI shell, and set the V3 driver to load up on boot after copying to the system EFI.
Following reboot saw the ratios max out at 25x again.
I tried loading V3.efi once again from the shell, this time, instead of loaded, it says locked.

Any ideas?

Thanks !

 

[cekim]

I went through the steps, and first time, it was fine. managed to get sustained 30x multipliers.

I then rebooted and went back to the EFI shell, and set the V3 driver to load up on boot after copying to the system EFI.
Following reboot saw the ratios max out at 25x again.
I tried loading V3.efi once again from the shell, this time, instead of loaded, it says locked.

Any ideas?

Thanks !

2 likely sources of the lock:
1. V3.EFI is being loaded (it will print a little message at the upper left)
2. BIOS was recovered to backup - double check that your uCode version is still "0" in the BIOS.

 

[Dingmel]

Thanks @cekim. I figured it out partially. It was because the driver was already loaded into and i went into the shell and tried to reload it.
issue now is, with the autoloaded driver, it still defaults to 25x. weird.

 

[cekim]

Thanks @cekim. I figured it out partially. It was because the driver was already loaded into and i went into the shell and tried to reload it.
issue now is, with the autoloaded driver, it still defaults to 25x. weird.

Some other variable you altered? Disable HT?
Also, keep in mind that anything that uses AVX will down-clock significantly. So, you need to check with something like cinebench that should give you an all-core turbo as good as it gets.

 

[Dingmel]

ok i've partially narrowed down the problem.

When i disable secure boot, and login to efi shell, manually load up v3.efi, then boot windows via the shell manually, it works. I'm able to get 30x on all cores.

problem was when i used bcfg to get the driver to load automatically.

running the command gave me a new window within the Rampage v extreme's efi which was an option to load the v3 turbo driver on boot.
with that enabled, somehow, the cores max out at 25x.

the bcfg command used was similar to the one offered in the thread. weird.


EDIT : bcfg driver dump - shows that the V3.EFI has been added, but still it doesn't self load

 

[GTZ]

Hi @fromax , and many thanks to @Dufus and many others who contributed.
I have a Xeon e5 2683 v3 - 2GHz -2,3GHz ES model F stepping 1.
MSI x99a raider bios p50
I did as you say, i changed the strings, and when i load to check with UBU, I see that the 306f2 is deleted from bios, and i flash it using the msi utility and / or ami flash utility afuwin64, after flash bios sees the cpu as 306f1, and then when i boot my computer, i check the current bios and it does show that the 306f2 is removed, but i think it is trying to use an even earlier version as a failsafe?
Anyway, when i load the v3.efi, it says wrong cpu. clearly, i am not able to remove the bios microcode ... am i right?
any suggestions from people who succedd doing is welcome, i was about to go buy an AMD ryzen for this, since my cpu only scores 1000 at single now, which is quite sad ... i know it has a potentialll...

as a side note, i can change the turbo ratio to 30 on bios (i was able to do that before) and cpuz still shows 26 max. which it never reaches that anyway.


thanks for the help

Try loading EFI driver V3_1, that dufus uploaded for another member for ES CPUs. It might work, it is good to see that your motherboard boots up with not microcode using an ES. You would be the first one becuase all other ES users were not able to post. Another thing I find odd is that you are able to increase the multi to 30. Usually Xeons let you change to the max multi allowed, in your case 26. If successful with the EFI driver for ES (F1 CPUs) and by chance you manage to get the CPU to 30.

Keep us updated.

 

[cekim]

This morning i tried again.
i flash with ubu, the bios again
i goes to bios to flash my new bios do with ubu
flash was ok
but i cant boot , the motherboard was blocked at 94 number on motherboard
you know, during the boot on the motherboard are displayed some numbers
mine was blocked at 94 and i failed!
can you tell me why?

94 is a PCIe device failure - often a video card that doesn't support UEFI.

 

[inventoroz]

Try loading EFI driver V3_1, that dufus uploaded for another member for ES CPUs. It might work, it is good to see that your motherboard boots up with not microcode using an ES. You would be the first one becuase all other ES users were not able to post. Another thing I find odd is that you are able to increase the multi to 30. Usually Xeons let you change to the max multi allowed, in your case 26. If successful with the EFI driver for ES (F1 CPUs) and by chance you manage to get the CPU to 30.

Keep us updated.

Thanks for the reply @GTZ .. CPUZ shows max multiplier 26 even if i change it in bios to a higher setting.
I really need help from someone who did it in an msi mobo.
I did exactly like @fromax said for the bios mod, I tried 5 different bioses, updating from windows or the within MBIOS option in the UEFI, non of them work to yield changing of the 0306F1, and I think we definitely need a guide for people to try this out.
It is the highest yielding mod, I will ever achieve if i can get it to work, I did many mods with Xeon X5650 and X5450, but this is quite a game changer since I will not need another cpu to do anything else if this one can game...

I will try the V3_1 when i get back home.

 

[Dingmel]

Sorted the problem out. Secureboot wasn't completely disabled... my own idiocy lol.

Edit : Looks like i spoke too soon. After i was trying out different bclk settings, the system crashed, and i needed to clear the CMOS.

Now, after adding the bcfg entry, and a startup.nsh entry (which intially worked), on top of disabling Secure Boot, it still doesn't auto load.

Any ideas?

 

[vermon]

Hello,

First of all thanks guys for all the time that was needed to make it work. Today I have also modded my bios, loaded V3.EFI, removed Win 10 microcode and then put x39 through VMware tool. All is working fine, Cinenebch score of 2000+.

My question though is ... while using watt-o-meter on stock bios/CPU settings I have 36W idle power consumption. While modded, it jumped to ~56-60W.

I guessed that Dufus in his script disables C states (?). I tried to remove the V3.EFI script, and windows boots fine, but when it goes into C'states it freezes. Wondering what is causing this behavior (MB with no microcode, doesn't know how to handle C-states?!). Because in theory we are not doing anything to the C-states (?).

In a nutshell - I wanted to ask if anyone played around to enable full multiplier without using the V3 script? And then manually disabled/enabled some C-states so it's 'stable'? (from what I remember C6 did not give much difference, maybe C3 or C1 is sufficient). What I am trying to achieve is to have the processing power available if needed, but still keep the idle power consumption low.

Any ideas ?

 

[citybobby]

Hello,

First of all thanks guys for all the time that was needed to make it work. Today I have also modded my bios, loaded V3.EFI, removed Win 10 microcode and then put x39 through VMware tool. All is working fine, Cinenebch score of 2000+.

My question though is ... while using watt-o-meter on stock bios/CPU settings I have 36W idle power consumption. While modded, it jumped to ~56-60W.

I guessed that Dufus in his script disables C states (?). I tried to remove the V3.EFI script, and windows boots fine, but when it goes into C'states it freezes. Wondering what is causing this behavior (MB with no microcode, doesn't know how to handle C-states?!). Because in theory we are not doing anything to the C-states (?).

In a nutshell - I wanted to ask if anyone played around to enable full multiplier without using the V3 script? And then manually disabled/enabled some C-states so it's 'stable'? (from what I remember C6 did not give much difference, maybe C3 or C1 is sufficient). What I am trying to achieve is to have the processing power available if needed, but still keep the idle power consumption low.

Any ideas ?

Yes, that can be done and this is exactly what I did.

My system is 2683 v3 on Taichi. C1 is enabled.

I used the v3 script and renamed "mcupdate_GenuineIntel.dll" (for win10). It maintains x12 at idle; full load with HT disabled @ x30; enabled, x28.

 

[vermon]

Yes, that can be done and this is exactly what I did.

My system is 2683 v3 on Taichi. C1 is enabled.

I used the v3 script and renamed "mcupdate_GenuineIntel.dll" (for win10). It maintains x12 at idle; full load with HT disabled @ x30; enabled, x28.

I also have 2683 V3 but different MB - Asrock Extreme 4 with Bios 3.6.

With the V3.EFI script loaded I can have C6/C3 states enabled in bios and everything is running fine and stable. Even HT enabled gave me full x30 multiplier, with Intel Speedstep, all C states etc. - they only question is related to idle power consumption.

So to sum up:

a) Stock 3.6 Bios and everything enabled - 36W idle power consumption
b) Modified 3.6 bios and everything enabled (V3 / V3-2 EFI script) - 56W idle power consumption
c) Modified 3.6 bios and no efi script - freezes in Windows 20-40 seconds after booting up. If I run e.g. prime95 during that time, everything is OK (no freeze). As soon as I stop it, it freezes when going into C states

So in a nutshell I am wondering if it is possible to achieve x30 multiplier, with everything enabled (c states, speedstep, ht) and to have an idle power consumption of ~36W. In theory we are unlocking here max turbo frequency, we shouldn't (?) touch so much the lowest states of CPU (?). It must be probably connected with how MB is treating the CPU without microcode in BIOS.

 

[citybobby]

For me, c states, etc are all at their default (auto, mostly) settings. I experienced no freeze at all; power comsumption is about 25W at idle; about 90 W at full load.

 

[vermon]

For me, c states, etc are all at their default (auto, mostly) settings. I experienced no freeze at all; power comsumption is about 25W at idle; about 90 W at full load.

How are you measuring power consumption? Because apps like HWinfo can be far from accurate.

25W idle and 90W at full load is just hard to believe (unless you are writing CPU only, I meant whole platform). At 3.0 GHZ just CPU would be taking 90W or probably more - when you add to that RAM, GPU even at idle (as there in no integrated GPU), MB, HDDs, fans etc. I would expect much higher consumption. For me Prime95 with 8x4GB DDR4 was at ~176W stock and ~200W modified (from respectively ~36W and ~56W idle).

 

[Welsper]

It must be probably connected with how MB is treating the CPU without microcode in BIOS.

I've insert microcode update into efi driver. (V3x2_mc39.efi). So I've the boost unlock and 0x39 microcode in BIOS setup stage.
But this is useless for C3/C6 on my PC (EP2C612 WS, 2686v3) it still freezes during OS loading. With 0x27 microcode - same result.
However one man has write to me that my efi driver allows him to enable C6 on his PC (Asus z10pe d16, 2673v3) and it decreases power on idle: -5W per CPU.

 

[vermon]

I've insert microcode update into efi driver. (V3x2_mc39.efi). So I've the boost unlock and 0x39 microcode in BIOS setup stage.
But this is useless for C3/C6 on my PC (EP2C612 WS, 2686v3) it still freezes during OS loading. With 0x27 microcode - same result.
However one man has write to me that my efi driver allows him to enable C6 on his PC (Asus z10pe d16, 2673v3).

I was also testing your V3 file (had 4 - V3, V3_1, your older version without x39 and your newer version) - I was able to have C6/C3 enabled fine (no freezes) on all of them. But still watt-o-meter showed consumption of ~60W instead of ~36W while on stock BIOS.

 

[citybobby]

Yes, I refer to cpu consumption only. Compared to the stock bios, it eats about 10W more at idle - these readings are from Core Temp (can't tell whether they are accurate or not).

 

[Welsper]

But still watt-o-meter showed consumption of ~60W instead of ~36W while on stock BIOS.

Hmm. It may be uncore CPU block. V3.efi changes multipliers for boost on cores (0x1AD..1AF registers) and also as I guess multiplier for uncore (register 0x620 - I've no description). As I remember uncore multiplier changes to lower value on idle but now it is fixed at max value (3GHz in my case).

 

[vermon]

Hmm. It may be uncore CPU block. V3.efi changes multipliers for boost on cores (0x1AD..1AF registers) and also as I guess multiplier for uncore (register 0x620 - I've no description). As I remember uncore multiplier changes to lower value on idle but now it is fixed at max value (3GHz in my case).

Do you have perhaps any ideas how this could be fixed?

 

[Welsper]

Do you have perhaps any ideas how this could be fixed?

Try this one: https://ufile.io/e664a1

It changes cores boost + has some power limit experiments + updates microcode to 0x39, but doesn't touch uncore.
Looks like it works. But it may have several penalty to performance somewhere in comparision with V3.efi where uncore on FULLMAX everytime.

 

[vermon]

Try this one: https://ufile.io/e664a1

It changes cores boost + has some power limit experiments + updates microcode to 0x39, but doesn't touch uncore.
Looks like it works. But it may have several penalty to performance somewhere in comparision with V3.efi where uncore on FULLMAX everytime.

I just tried it. It freezes with C6 enabled, but not on C3. On C3 idle usage is ~50W. So 6W less than previously, but still much higher than 36 on stock. Cinebench was ok at ~2040. Wondering why C6 makes it freeze on a modified bios. Any ideas what could be causing it/how to fix it? I can test the solution Maybe lowering the voltage causes it? Maybe we could try to leave it changeable so it can be changed from Windows (?). Because it seemed a random freeze (it went to C6 state for a bit, but after X seconds it just froze).

 

[Welsper]

Hmm.
I didn't check C3/C6 yet. W/o Cx it looks stable, I've been using it for a couple of hours.
And I've got less than 50W per CPU... closer to 30W in balanced mode. And about 30-40W in Idle with "High performance".

On C3 idle usage is ~50W.

How much power do you have in "DRAM Power" ?

Changing from windows - It's problem I think. 1) microcode present. 2) MSR inaccessible from application level (ring3).

Spoiler: With UnCore Max (original V3, windows power options: Balanced)

Spoiler: Win UnCore default (windows power options: Balanced)

 

[vermon]

Hmm.
I didn't check C3/C6 yet. W/o Cx it looks stable, I've been using it for a couple of hours.
And I've got less than 50W per CPU... closer to 30W in balanced mode. And about 30-40W in Idle with "High performance".

How much power do you have in "DRAM Power" ?

Changing from windows - It's problem I think. 1) microcode present. 2) MSR inaccessible from application level (ring3).

Spoiler: With UnCore Max (original V3, windows power options: Balanced)

Spoiler: Win UnCore default (windows power options: Balanced)

I was speaking about total system power consumption (measured with watt-o-meter in power socket). In HWInfo all the time it shows me (in idle) CPU around 6-8W, DRAM around 2W. But the funny thing is it was showing similar values when I had ~56W consumption from the 'wall' and ~36W ... so I do not trust it so much

Edit: Now I tried to do a screenshot from HWInfo - I was able to work ~5 minutes with c6 enabled and then it randomly crashed. So maybe it is the matter of lowering the voltages? (at least in my case). Is there any chance for you to make EFI file without changing/locking voltages?

C6/C3 difference for me: https://1drv.ms/f/s!AsjK6p42qStil2qP71T-qb99SdWL ... C3 - 50W, c6 - 42W, stock 36W.