What controls Turbo Core in Xeons?

Page 10 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

drchoi21

Junior Member
Feb 20, 2017
1
0
6
Has anyone tried Dual/Quad CPU Supermicro boards and posted the results? I have Supermicro X10DAi and would like to try out this mod.
 

TheSellHard

Junior Member
Jan 29, 2017
7
3
36
@GTZ no ratio overclocking for that CPU so this bug will not unlock the i5 -4460 unless perhaps one can write microcode. Best that you can probably hope for is all cores at 3.4GHz if set with a microcode version of 0x7 or less.

@fromax legacy mode should not matter, try running the shell driver in test mode first and from there exit to W7 or are you already able to do that? If it's a case that you are successfully booting W7 but stuck with base clock then please post a CPUZ txt report. First time I ever heard of RU is when The Stilt posted it so he may be able to give a few pointers. The MSR settings are already noted in the thread, V3.EFI was just meant as a crude way to apply those settings.

@vadimkara my microcode is not removed but modified similar to post #101. If it's not posting with default settings (no OC / FIVR defaults) then perhaps that CPU / BIOS combo relies on something being fixed from the microcode. Do note for that CPU it will be microcode with CPUID 306f1? not 306f2.

@Soarer while waiting you could go through the steps fromax took to modify his BIOS and possibly try something similar.
Dufus. Thx!
I installed two processors alternately and 2683 2695. One of the runs of the system with 2683 showed me monstrous figures. Photos in instagram. I tried many times to replicate it but to no avail. How is this possible? It's not photoshop!!! Corona Render showed 5670000, CPU-Z 23700 and CB R15 2332!!! =) Tell me how this could happen ???

Result CB R15 - https://www.instagram.com/p/BQv2q-WgY7i/?taken-by=thesellhard
 
  • Like
Reactions: Dufus

Student777

Junior Member
Feb 22, 2017
2
0
1
Did you only use UBU to remove the uCode? - That happend to me as well. After that I used MMtool to remove the ucode, and then afterwards open it in UBU and just straight away save it (without any other modifications, just saving it).

So:
1. Open bios in MMtool, remove ucode, save file/bios
2. Open saved file/bios in UBU, and save straight away (option 0 if I remember correctly) Don't do anything else.

3. Use the saved file/bios from UBU to flash.

Should solve the no-booting issue :)

Hi,
just signed up to the forum. First let me say, it's awesome what u guys have accomplished here. Great work!
My question is... Did anyone got it physically (not theoretically) realized to remove microcode from ASUS X99 (Sabertooth) motherboard with AMI Aptio V UEFI BIOS CAP FILE ver. 3402 installed and what's more importent, managed to get it flashed back to the mobo? Tried all methods mentioned here or at linked pages... but flashing it back to the board always failed. Tried all flashing methods too.
Please help!!! I got an Xeon E5-4667 v3 running on above mentioned mobo ASUS X99 Sabertooth with fw 3402. It's running great with a CPU-Z bench of 17800 but it's a shame not having it running at 'all cylinders'.
I really appreciate your help/support. Thanx in advance
 

strawberrydoll

Junior Member
Feb 22, 2017
1
0
1
With X99 ASUS motherboards, you can't flash the same version bios back to itself, so you'll need to downgrade first then flash the new modded bios according to sinders on win-raid
 

Dufus

Senior member
Sep 20, 2010
675
116
101
Seems CPU 306F1 can cause post problems with no microcode update. IOW the method shown on this thread is unsuitable and modification would likely have to happen in the BIOS boot block.

Remember V3.EFI uses BIOS NVRAM to get called so resetting to BIOS defaults or CMOS reset can clear that so it will no longer be loaded at POST.

@Ionstream the hexedit method is a possible way to avoid changes to file length while keeping simple checksums in check. Since some of those executable pointers may be hard coded then changing positions of code in the region would likely cause failure to POST. Of course if other checks are at hand there is no guarantee of success. Your registers show that turbo ratio changing is supported and not locked out, so should be good to go.

@TheSellHard in CBR15 base clock is shown as 2.3GHz which normally would means 1. BCLK at 115MHz, 2. a timer bug or 3. an option for cTDP up. In the video the benchmark takes 20 seconds to complete and such a time would be indicative of a score of around 2025. 2330 should take a little under 17.5 seconds if my calculation is correct. It would appear the timer(s) is showing a 15% boost.

Nice results from you all and great work in helping each other. :)
 
  • Like
Reactions: xeon_fan

TheSellHard

Junior Member
Jan 29, 2017
7
3
36
Seems CPU 306F1 can cause post problems with no microcode update. IOW the method shown on this thread is unsuitable and modification would likely have to happen in the BIOS boot block.

Remember V3.EFI uses BIOS NVRAM to get called so resetting to BIOS defaults or CMOS reset can clear that so it will no longer be loaded at POST.

@Ionstream the hexedit method is a possible way to avoid changes to file length while keeping simple checksums in check. Since some of those executable pointers may be hard coded then changing positions of code in the region would likely cause failure to POST. Of course if other checks are at hand there is no guarantee of success. Your registers show that turbo ratio changing is supported and not locked out, so should be good to go.

@TheSellHard in CBR15 base clock is shown as 2.3GHz which normally would means 1. BCLK at 115MHz, 2. a timer bug or 3. an option for cTDP up. In the video the benchmark takes 20 seconds to complete and such a time would be indicative of a score of around 2025. 2330 should take a little under 17.5 seconds if my calculation is correct. It would appear the timer(s) is showing a 15% boost.

Nice results from you all and great work in helping each other. :)

Hi Dufus ana thx.
My 2683v3 QS working on BCLK 100, but on 2.3 Ghz ;-)
I don't know how.
 

KAL2017

Junior Member
Jan 28, 2017
4
3
16
"
KAL2017 said:
I ended up going with a pair of E5-2667 V3 ES cpu's
What's cpuid of these processors - 306f2 ?"


No they are 306f1. They are 14 core cpu's with 6 cores disabled. HWinfo64 shows them to HCC cpu's with 8 cores LOL. They also have 35 mb of L3 cache.

With the id 306f1 of these cpu's makes me really leery to give it a try, I am going to contact Asrock Rack to see if they will send/sell me a bios chip with the latest bios since this mb came with the first version bios.

Oh and there is something wrong with the quote function, it keeps erroring out on me.
 
Last edited:

Anrentren

Junior Member
Feb 23, 2017
3
2
41
Hi.
I have asus x99-e ws/usb 3.1, and got stuck in UBU with error "Error in Replacing File". I run it as Admin and try on different PC, UBU/Windows/BIOS version.
Is there any solution to this, or maybe someone can help?
 

lucien_br

Member
Feb 13, 2017
37
7
41
Dufus. Thx!
I installed two processors alternately and 2683 2695. One of the runs of the system with 2683 showed me monstrous figures. Photos in instagram. I tried many times to replicate it but to no avail. How is this possible? It's not photoshop!!! Corona Render showed 5670000, CPU-Z 23700 and CB R15 2332!!! =) Tell me how this could happen ???

Result CB R15 - https://www.instagram.com/p/BQv2q-WgY7i/?taken-by=thesellhard
This is a fake result with timer hack or overclocked 15% bclk?
 

SZAO6

Junior Member
Feb 19, 2017
13
1
51
KAL2017 have you done this on Your dual CPU motherboard? I'm so curious that's possible on the dual CPU. I'm thinking about replacing my old machine with two E5 2670v1 with two 2683v3 (of course with another replacements) because I have two more of this CPU at home, and make it 3GHz on all cores on two CPUs.
 

GTZ

Member
Jan 25, 2017
36
12
41
Does anybody own a pre-qs 2695 v3? This chip has retail stepping but has a max turbo of 3.5 as appose 3.3 on QS and retail chips. I am wondering if it can sustain an all core turbo of 3.5 (3.6 with a little BCLK overclock), or will it have a similar story like the 2686 or 2699 where it drops a few bins.
 

SZAO6

Junior Member
Feb 19, 2017
13
1
51
So tell me, if I have two oem 2683v3 CPUs, next step - buy a dual CPU motherboard like Asus Z10 D8WS or maybe better (and cheaper) ASrock EP2C612 WS with AMI Bios (I'm not sure Asus Z10 have ami bios), will it work on dual CPU? Anyone tried?
 

TheSellHard

Junior Member
Jan 29, 2017
7
3
36
This is a fake result with timer hack or overclocked 15% bclk?
The results of measurements of the installed 2683v3 QS after 2695v3 QS. For some reason, frequencies for 2695 to apply for 2683. I don't know why. Dufus made a comment.

BCLK - 100!!!
 

Dufus

Senior member
Sep 20, 2010
675
116
101
@TheSellHard Sounds like an issue with Windows using the TSC (Time Stamp Counter) of the CPU as the performance clock and not updating properly after the swap, perhaps a fast boot issue.

Try setting Windows to use the High Precision Event Timer (HPET) instead by enabling it in BIOS if not already enabled and adding the BCDEdit option "useplatformclock yes".
 

TheSellHard

Junior Member
Jan 29, 2017
7
3
36
@TheSellHard Sounds like an issue with Windows using the TSC (Time Stamp Counter) of the CPU as the performance clock and not updating properly after the swap, perhaps a fast boot issue.

Try setting Windows to use the High Precision Event Timer (HPET) instead by enabling it in BIOS if not already enabled and adding the BCDEdit option "useplatformclock yes".
Thank You Dufus. You're the best ;-)
 

ajschot

Member
Feb 16, 2017
65
2
16
So got my E5-2683v3 (OEM) wow great work with my E5-2658v4 QS I got multi score of 17000 in Z-cpu now with this methode and the E5-2683v3 I got around 21000 multiscore, with Geekbench 4 I got 26343 on multi core and 3200 single core. I think it is pretty good.... only on OS X my system freezes so I have to find out how I can get it to work, but without the V3.EFI loaded it also freezes so it could be an other problem. I see better score's here but I think I have to tweak some more to get it better.

Thank you all for all the work (specially Dufus)
 

Zladimir

Member
Apr 14, 2011
33
3
71
Hi there,
First of all, thanks to everyone for contribution and special thanks to Dufus and Stilt !

Now I got my E5-2686v3 at constant ratio of 28x (+500MHz) on allcore!


Here are version 0x27 & 0x39 microcodes for Haswell-Ex (0x306F2) in VMWare driver / Linux compatible format: https://1drv.ms/u/s!Ag6oE4SOsCmDhFnET3uw9wHeV4EA
Rename the desired version to microcode.dat, and proceed as instructed by VMWare.
From where you got them ? And how to get other microcode versions in .dat format ? I would like to try others.


All Haswell-EP HCC SKUs should be unlocked up to 240W.
Are you sure ? Afaik only the max limit is 240W, but the Power Limit 1 is already forced at 120W. In my Taichi board Ive the ability to change them, but the new values are not taken. The only settings which I can successfully change are Thermal Design Current and the time for Long Duration of Power Limit 1.
240W is impossible to reach in my opinion, cause already Power Limit with 144W kicks in after 7,81ms (at least as HWinfo64 says)

What tools should i use, to programm MSR registers(maybe in small manual).
Yeah, a small manual on programming MSR would be cool.


By my own observation mc 27 and 39 perform the same. With both I get only an allcore ratio of 27x and have a CPU consumption of 105-106 W. CPUz gives me an increase from 20.000 points to ~24000. Quite well of course.
Using no microcode (0) in Windows gave me yet best performance. On constant 98W consumption I get constant ratio of 28x (two, three cores jump even to x29) and stable 25.000 points in CPUz.
Furthermore I tried to play with vcore offset. Strangely it has on microcode 0 on my Taichi-board a reversed effect. When I decrease my vcore in UEFI my actual voltage increases. But when I increase my offset by +0,100V as reversed it decreases, consumption jumps down to 94W with same CPUz result and less heat production (quite nice!).
Sadly increasing or decreasing Vcore offset by 0,100V didnt had any influence on my maximum ratio. Also setting my BCLK to 100 or 102 (my two m.2 nvmes dont tolerate more) didnt had any influence on my max ratio.
Checking HWinfo64 I can say for sure there is no throttling of thermal or current in effect. So does anyone found yet out why the CPU doesnt clock on a higher ratio ?


Does anyone know if its possible to trick the computer believing there is sitting another cpu inside than I have ? For example faking a 2696 or 2699 v3 cpu. Just as a thought.

My 2686v3 stayed at 2.9Mhz on all 18-core with the micro code 0x27. Is there any way to make 3.5Ghz in the 18-core?
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.


@TheSellHardTry setting Windows to use the High Precision Event Timer (HPET) instead by enabling it in BIOS if not already enabled and adding the BCDEdit option "useplatformclock yes".
What does these options do exactly ?
 

ASK THE COMMUNITY