What controls Turbo Core in Xeons?

Page 21 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

RamRock

Member
Apr 3, 2017
26
2
16
@UP
Or to be more exactly - user mistake. If it is really something wrong with motheboard - better go for replacement.
 

knowndragon

Junior Member
Apr 3, 2017
17
4
36
@knowndragon
Can you explain how V3_2 mod work? Because more I read less I understood... it set max turbo for all cores as max turbo for single core (limit is power consumption) or it set clocks just for some cores (that are set up in v3x2c).
I will look into it. The v3_2 for 2 cpu's or more. I know your running an Asus board. I am surprised it didn't have core optimization in it. Especially with the price paid for that motherboard. I think with that version of the v3 that some of the cores have to step down. I'll download it and look. Also not saying that your cooling isn't up to par. There is thermal throttle. Also make sure that windows didn't slip you an update if your on 10.
 

sciff

Member
Mar 6, 2017
136
52
71
Hello everyone!
Got my Xeon E5-2696 V3 a week ago, and tonight, after long struggles, I finally managed to unlock all core turbo. Below are my results.












One thing I can't understand. Why do even we need a microcode if everything works perfectly without it? Just wondering :)

As for 0x27 microcode, my system turned out to be unstable with it (which is no surprise, I know it is less stable than 0x39). The video driver dropped out during CPU-Z stress-test (the monitor turned off, then switched on and I got a notification that video diver has just been restored). There were no notable differences in performance, the same x33 multiplier and the same 3305 MHz. Only Cinebench performed a little better (2822 against 2789 with 0x39).

By the way, you can see on AIDA64 OSD-panel, that the video card is not present.

Therefore, I preferred 0x39 microcode, since I need stability for professional use.

P.S. Is the TDP of 93W shown on OSD-panel (AIDA64 Extreme) real? I can't believe that this processor doesn't even reach 145W with unlocked turbo boost and under load...
 
Last edited:
  • Like
Reactions: zzori

Ionstream

Member
Nov 19, 2016
55
24
51
@Ionstream
Did you turn off all C-states in bios ?


Best way is to use Asus flash back.
No you don't have to change anything except to start with UEFI, turn off C States.
Maybe less stable- but it's normal while you doing stuff like this :)
Nope, don't think it's gonna have an effect on my end.

Anyway, this is a SuperMicro board, so that ASUS flashback feature is not available to me.
 

GTZ

Member
Jan 25, 2017
36
12
41
Hello everyone!
Got my Xeon E5-2696 V3 a week ago, and tonight, after long struggles, I finally managed to unlock all core turbo. Below are my results.












One thing I can't understand. Why do even we need a microcode if everything works perfectly without it? Just wondering :)

As for 0x27 microcode, my system turned out to be unstable with it (which is no surprise, I know it is less stable than 0x39). The video driver dropped out during CPU-Z stress-test (the monitor turned off, then switched on and I got a notification that video diver has just been restored). There were no notable differences in performance, the same x33 multiplier and the same 3305 MHz. Only Cinebench performed a little better (2822 against 2789 with 0x39).

By the way, you can see on AIDA64 OSD-panel, that the video card is not present.

Therefore, I preferred 0x39 microcode, since I need stability for professional use.

P.S. Is the TDP of 93W shown on OSD-panel (AIDA64 Extreme) real? I can't believe that this processor doesn't even reach 145W with unlocked turbo boost and under load...
This is why I posted my findings earlier, your post validates my post.

When you run Cinebench R15, with no ucode causes uncore to downclock to 2300 just like on my setup. Netting you a decrease of 100 points on your score. It only downclock under heavy load though. You can check it with HWinfo64. It it looks like ucodes fixed clocking issues with the uncore.
 
  • Like
Reactions: sciff

sciff

Member
Mar 6, 2017
136
52
71
GTZ, interesting, I didn't even notice this difference!

By the way, it turns out having no microcode loaded, causes certain applications to actually run faster. Take Corona benchmark, which utilizes AVX and SSE4.2 instructions. Normally in my case it runs at a maximum of 2.8 GHz, whereas if you uninstall the microcode, it runs at 3.3 GHz


 
Last edited:

randir

Junior Member
Mar 19, 2017
15
13
41
@Ionstream ,
>this is a SuperMicro board

MMTool won't work with it. I've successfully modded X10DRi-T by an algorithm described at the beginning of this thread, when you manually modify 2 bytes of the microcode signature so it is no longer recognized as 306F2. There're four f2 06 03 00 sequences in this bios, and you need to change 2nd and 3rd ones (but always re-check bios validity with all tools available after this, yours may have different structure).

@sciff ,
>having no microcode loaded, causes certain applications to actually run faster
This has already been discussed - Intel changed 256-bit registers throttling during development. You can try to lower uncore clock speed and that you'll get higher base clock, as it's hit by TDP limits pretty hard.

I personally won't run any production enviroment without any microcode, and probably with anything earlier then the microcode from the times of the first QS released - Intel doesn't share what they've actually patched, only some really outstanding bits, like TSX bug.

@Welsper, I've researched some updates for the v3_2.efi, can I use your sources under GPL to start a github project for more modifications?
 
  • Like
Reactions: Ionstream and sciff

RamRock

Member
Apr 3, 2017
26
2
16
Hello everyone!
Got my Xeon E5-2696 V3 a week ago, and tonight, after long struggles, I finally managed to unlock all core turbo. Below are my results.












One thing I can't understand. Why do even we need a microcode if everything works perfectly without it? Just wondering :)

As for 0x27 microcode, my system turned out to be unstable with it (which is no surprise, I know it is less stable than 0x39). The video driver dropped out during CPU-Z stress-test (the monitor turned off, then switched on and I got a notification that video diver has just been restored). There were no notable differences in performance, the same x33 multiplier and the same 3305 MHz. Only Cinebench performed a little better (2822 against 2789 with 0x39).

By the way, you can see on AIDA64 OSD-panel, that the video card is not present.

Therefore, I preferred 0x39 microcode, since I need stability for professional use.

P.S. Is the TDP of 93W shown on OSD-panel (AIDA64 Extreme) real? I can't believe that this processor doesn't even reach 145W with unlocked turbo boost and under load...
Mainly microcode making less IDLE power consumption.
Not too real another thing that microcode fix :)

How did you increase multiplayer to 33?
 

RamRock

Member
Apr 3, 2017
26
2
16
I will look into it. The v3_2 for 2 cpu's or more. I know your running an Asus board. I am surprised it didn't have core optimization in it. Especially with the price paid for that motherboard. I think with that version of the v3 that some of the cores have to step down. I'll download it and look. Also not saying that your cooling isn't up to par. There is thermal throttle. Also make sure that windows didn't slip you an update if your on 10.
Bios on Z10 D16 WS it's have just a little part of desktop boards....
I have 2x Noctua u14s they are getting job done (less than 70 C).

When I'm using less stress apps (70% Cinebench power consumption) It's still doesnt turbo to max single core turbo.
 

custom90gt

Member
Feb 9, 2017
38
7
41
@vermon - any luck getting C6 to run well? I have the same issue with it freezing loading windows and having a much higher idle power usage.
Thanks!
 

sciff

Member
Mar 6, 2017
136
52
71
Mainly microcode making less IDLE power consumption.
Not too real another thing that microcode fix :)
Are you sure? AIDA64 OSD-panel on my screenshots doesn't show that there's any less consumption when e.g. 0x39 microcode is used.

Well, it would be great if there was a microcode allowing Corona Benchmark running at maximum turbo frequencies...

How did you increase multiplayer to 33?
I had to make some bios settings changes on my ASUS X99-E WS/USB 3.1. But of course I didn't set the multiplier manually.
 
Last edited:

rottwag

Member
Apr 2, 2017
77
11
41
I had to make some bios settings changes on my ASUS X99-E WS/USB 3.1. But of course I didn't set the multiplier manually.
As i own an Asus x99-A, may I ask which options you changed? would be great!

Thanks
 

sciff

Member
Mar 6, 2017
136
52
71
No problem, Andy! Here are my BIOS screenshots, where I made changes:

























 

RamRock

Member
Apr 3, 2017
26
2
16
He didn't turned off C6 that's a reason.
@ByAuS read last 3 pages plz :)


@sciff
Same options but I got just x31 :(
Did you load some efi?

Yes I messaure from a wall power consumption. Things in hwmonitor are not so good
 

sciff

Member
Mar 6, 2017
136
52
71
@sciff
Same options but I got just x31 :(
Did you load some efi?
Yes, without efi it won't work.

I just happened to reinstall windows, I had to actually format the whole half of the HDD with the system partition and a small additional one which isn't visible in windows, and because of that this efi stopped loading (even though I repeated the procedure of copying the files manually into C:\EFI\BOOT folder). The unlock disappeared.

Therefore I had to reinstall the V3.EFI and also make sure it loads every time. Some of the BIOS adjustments I made were specifically targeted at that. So that it loads at every system startup, no matter what, whether you turn on the computer, restart it or load Windows from BIOS.

Before those changes, sometimes it wouldn't load. I think Fast Boot option is the main culprit.

P.S. By the way, as I understand, the Sleep mode disables the unlock?
 
Last edited:

RamRock

Member
Apr 3, 2017
26
2
16
@sciff
Yes... but my problem is you got multiplayer 33 and I have 31 - same CPU, similar config of the bios (I have Z10PE-D16 WS).

PS
C6-states can freez PC.
 
  • Like
Reactions: knowndragon

sciff

Member
Mar 6, 2017
136
52
71
RamRock, I was told by TheSellHard, that dual CPU mainboards have weaker CPU power supply, therefore you can't boost a CPU or CPUs as much on C612 chipset.

On my ASUS X99-E WS/USB 3.1 I've got just one CPU but two CPU power connectors. You see what I mean? In all core turbo my E5-2696 V3 starts to consume power like crazy, hence increased temperatures and cooler FAN RPM. I can even start hearing a slightly noticeable weird sound, as if my PC had a mini-turbine inside :D

It's unrealistic to expect the same results.

And why do you think x31 multiplier is bad? It's good enough, in my opinion, especially if you have two CPUs running at this speed.

P.S. Which revision is your CPU? Mine is R2.
 
Last edited:

ASK THE COMMUNITY