What controls Turbo Core in Xeons?

[DOC_Zenith]

Thank you, got that. I was not sure if this file includes those with -50mv and no in-system power consumption ones.

 

[bbm_]

1st post

first and foremost, thanks to all those who have provided the tools necessary for a crazy feat.

here are my current results using v3x2_50_vcc.efi w/ no microcode. was able to achieve x33 without microcode and only x32 with mc 39. also get spikes of 5ghz + on single core, looks very concerning. see cores: 3 & 19.
http://imgur.com/2hk0GcO

other scores:
passmark: 28159 http://www.passmark.com/baselines/V9/display.php?id=84834904332
geekbench: 37866 https://browser.primatelabs.com/v4/cpu/3201634
cinebench: 2920 http://imgur.com/XLWwqP1

 

[Dan136]

Does anyone have the source code for the EFI files? I have a ES-2695v3, currently my best result has been 3.2GHz under load with no microcode and 50mv offset. I think I am hitting power throttling. I would like to try and get microcode 27 working with a 50+mv offset, I can hopefully modify and recompile the code, but I doubt I will be able to write it from scratch. Currently with Microcode 27 I am at 3Ghz, but since I have the ES my max multiplier is 35, so I would like to get as close to that as possible.

*by 50+mv offset I mean I want to reduce vcore by 50mv or possibly more.

 

[abufrejoval]

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!

I always thought that the only difference between the ASrock Extreme4 and Extreme4/3.1 was the addition of a USB 3.1 card to the latter...
But when I look a the downloads section for both, I see totally distinct BIOS versions and release dates...

So would you kindly be able to do the same operation you did for the Extreme4 BIOS on an Extreme4/3.1 BIOS?

Because I did try doing it the UBU way (for both v1.2 and v3.4) and wound up with identical results: A BIOS which would just "scream" at me (one long beep, until you switch off the power) when I try to enter setup after posting (didn't even try to let it boot my non-uefi Windows 2016).

 

[ajschot]

Hey guys, i have a question...

I am tying to mod the bios of my secondary system, it is a ASUS ROG Strix X99 Gaming board.
I managed to mod it but how to remove the security check. When trying to flash the bios i get an error that the bios is not valid.
In this thread i saw people with asus boards, but how to get pass the check i forgot and i can not find it here in this thread...
Can anyboy help me?

EDIT: Have to use BIOS flashback, got it

But for a reason it gives still the mc! Really strange. Too bad it is so difficult to check if the bios flashback is running... the light flashes 3 times and stay on for about 10min but after that mc38 is still in de bios screen and i checked the bios file and it is removed i really don'tknow what is going on.
It is just not flashing the bios and i don't know why

EDIT2: Fault in the manual instead of after the 3 flashes it need to bensolid means it is not working the light must keep on flashing and when it stops it is flashed

 

[ajschot]

I always thought that the only difference between the ASrock Extreme4 and Extreme4/3.1 was the addition of a USB 3.1 card to the latter...
But when I look a the downloads section for both, I see totally distinct BIOS versions and release dates...

So would you kindly be able to do the same operation you did for the Extreme4 BIOS on an Extreme4/3.1 BIOS?

Because I did try doing it the UBU way (for both v1.2 and v3.4) and wound up with identical results: A BIOS which would just "scream" at me (one long beep, until you switch off the power) when I try to enter setup after posting (didn't even try to let it boot my non-uefi Windows 2016).

That is strange i did mine with Ubu without any problems.

 

[Sean J Sauve]

Wow guys i'm at the cusp of this experience I can feel it!!!

I am being throttle locked to 23x on all 18 cores of my Xeon e5 2699 v3.

Can any of the geniuses that exist among us here graciously guide me through the final step. I have my Asrock x99 Extreme11 essentially running the driver at the efi layer (can see it before windows loads from drive) yanked the system32 microcode file out. Modified the bios UBU and UEFITOOL to remove the microcode. Used VMware microcode loader tool to bring it back again. Tried to keep to the bare bootable settings.

AIDA64 is reporting max turbo boost of 38x on all multipliers !!! WOW but essentially that's just a pinto decal-ed like a Ferrari. I mean I can't get it to work

PLEASE HELP sending out a beacon be my hero explain the best way to fix this

FIRST POST BTW

 

[Sean J Sauve]

OK so I tried v3x2_50_39_vcc.efi instead of v3. My cores all ramp up to 32x !!! BUT I am only getting 7950 on CPUz on a 18 core E5 2699 v3 ... Something is way off, I mean it did boost a little but ultimately nothing too special. Like the others in this forum.. Though I must have missed something? I can't figure out what. The EFI driver DID say something about a mere 30x on UNCORE ? Which is something I don't understand to be honest.

Anyhow if anyone out there reading this is able to help guide me a touch I would really love it thanks.

 

[James Kim]

OK so I tried v3x2_50_39_vcc.efi instead of v3. My cores all ramp up to 32x !!! BUT I am only getting 7950 on CPUz on a 18 core E5 2699 v3 ... Something is way off, I mean it did boost a little but ultimately nothing too special. Like the others in this forum.. Though I must have missed something? I can't figure out what. The EFI driver DID say something about a mere 30x on UNCORE ? Which is something I don't understand to be honest.

Anyhow if anyone out there reading this is able to help guide me a touch I would really love it thanks.

I now have 2 x E5-2696 v3 on Supermicro X10DRL-i motherboard. I previously had 2 x E5-2658 v3 and it could reach max turbo speed (max x29, 2.9Ghz) on both processors. However, with E5-2696, I can only reach x32 on all cores. If I reduce # of cores in BIOS, say 10 cores, It seems that I can get max turbo speed on all cores. But, this will reduce # of cores available.

One interesting thing that I found out playing with BIOS, some settings makes different temperature +5/-5 C.

I don't think that efi files make much difference once mcode is removed in BIOS. It may make +/- 100mhz?

With my current settings,

CPU1 - temp: 40C, x32 multiplier on all cores, x38 on spike
CPU2 - temp: 38C, x31 multiplier on all cores, x38 on spike

CPUZ Bench: Single: 423, Multi: 7562
Cinebench R15: 5000

I may settle on this unless there is new tweak coming out.

 

[Sean J Sauve]

I now have 2 x E5-2696 v3 on Supermicro X10DRL-i motherboard. I previously had 2 x E5-2658 v3 and it could reach max turbo speed (max x29, 2.9Ghz) on both processors. However, with E5-2696, I can only reach x32 on all cores. If I reduce # of cores in BIOS, say 10 cores, It seems that I can get max turbo speed on all cores. But, this will reduce # of cores available.

One interesting thing that I found out playing with BIOS, some settings makes different temperature +5/-5 C.

I don't think that efi files make much difference once mcode is removed in BIOS. It may make +/- 100mhz?

With my current settings,

CPU1 - temp: 40C, x32 multiplier on all cores, x38 on spike
CPU2 - temp: 38C, x31 multiplier on all cores, x38 on spike

CPUZ Bench: Single: 423, Multi: 7562
Cinebench R15: 5000

I may settle on this unless there is new tweak coming out.

Hmm that means that our processors at 18 cores are quiet a deal slower than other models despite what gtx and others were claiming about these being the fastest. While sporting fewer processors and cheaper price tags at that. I think I saw a 10 core xeon receiving 16000 on the multi in cpuz bench. That is nothing to be scoffed at that's literally twice as fast and then some with a fewer cored processor. Something about how our processor is being handled at the power limiting level makes this not such a performance increase really. And at the loss of stability, C states and thermal for us what is the bother. Heck we should get a 10 core xeon too, Or am I mistaken did someone get 18 cores at 38 at least something more substantial with this mod and our processor. Personally I think its being throttle at the Max power being locked at 65 watts. Maybe i'm wrong this is a hopeless en-devour.

How did you get the cinebench 5000 test with 2 processors? Or each can do ?

 

[sk8fly]

Hy all! I'm new here!

system: Z10PE-D16 WS with 2x 2683v3 (306F2, C0/C1, QS)

i struggeling with the stability and performance of the hack!

i used a flashed bios with only one microcode inside (0EF6F11F 2016/10/07)
first it hangs at boot screen (CPU CATERR), i solved this with disabeling C6 state and limited to C2
after 30min the pc will freez again with CATERR.....

i also done some testing with vray benchmark which is relly good for testing with high end processors
0:45 sec BiosMod
0:46 original 3407 Bios
0:41 original 3407 + 104 Bckl (CPU-Z Multi: 34275)

i used Bios + V3.EFI from this thread https://www.hardwareluxx.de/community/f101/all-core-turbo-dual-xeon-e5-1163969.html

a note, my SSD is in BOOTMNGR mode not UEFI so i'm only tried it with the temp solution via shell

are there any suggestions?

Thanks for your help

phil

 

[James Kim]

Hmm that means that our processors at 18 cores are quiet a deal slower than other models despite what gtx and others were claiming about these being the fastest. While sporting fewer processors and cheaper price tags at that. I think I saw a 10 core xeon receiving 16000 on the multi in cpuz bench. That is nothing to be scoffed at that's literally twice as fast and then some with a fewer cored processor. Something about how our processor is being handled at the power limiting level makes this not such a performance increase really. And at the loss of stability, C states and thermal for us what is the bother. Heck we should get a 10 core xeon too, Or am I mistaken did someone get 18 cores at 38 at least something more substantial with this mod and our processor. Personally I think its being throttle at the Max power being locked at 65 watts. Maybe i'm wrong this is a hopeless en-devour.

How did you get the cinebench 5000 test with 2 processors? Or each can do ?

Cinebench R15 score (5000) was done with two CPU. With 2 x E5-2658 v3 QS, I had 2800 without mod and 3300 with mod. As you mentioned about power usages, I checked mine and it said that 60W per CPU. And, I know that this CPU is rated 145W. Even fully loaded, it said that 60W is being used. I have no idea why it say that.

Also, the reason you believe that you have lower number on CPUz is because you are comparing the score from old version. The website's change logs for new release are below.

• Intel X299 (1.79.1).
• AMD ThreadRipper (1.79.1).
• AMD Ryzen 5 and Ryzen 3.
• Fix lockup at loading on Ryzen systems with RAID.
• New benchmark version (17.01)

And, I ran two versions on my system and saw the different numbers due to different benchmark method.

http://postimg.org/image/bp7feard9/

 

[Zladimir]

efi with vccin

Hey randir,

please send us your source files. I would really like to look at them. Anyway, what you think how you managed to get another +1 to the multiplier? I tried by myself to adjust vccin even with an offset of -120mV, but even as it changed, multiplier kept the same. So I seriously ask myself what you did different than me.
Any hint or clue would be helpful, maybe helping to push out more of our Xeons.

I highly doubt it is an tdp or tjmax or whatever limit, as Ive the the small 18core cpu "2686v3", and I am far away from my tdp levels or tjmax's (I use an oil cooling keeping whole environment on stress on max on 30°C, and even with "hacked mc" I keep under 105 Watt at full load).

Thanks in advance !

 

[lucien_br]

Dear all,
not really new to forum but really registed today to post this.
follow page 3's steps post by C-Power (which a total info gathering steps of both two genius, Dufus and stlit)
and my 2699 v3 can finally meets 6950x on single thread bench(cpu-z and 10-cores enable, 3.6g tb), happy with both my compute and gaming usage just with a click on bios settings(thx asus for my fav profiles LOL)
my workstation configuration:
MB: Asus X99-DELUXE-II
CPU: Xeon E5-2699 v3 ES (with 306F2h rev)
and other not so important things to boot up.....

since post imgs is not so easy, so I'm just ignore the screenshots....

wish everyone happy mod

Hi! I bought a X99-DELUXE II too. Please, can you share your modded bios? I'm afraid do it myself. Thanks!

 

[lucien_br]

Hello Sciff, Awesome, that's great news! Thanks so much for all the contributions everyone!

So far, using VERCETY's instructions I was able to remove the Haswell-E microcode from the x99 deluxe II bios.
Cooler has not arrived yet.

Sciff, so you are saying no microcode in the bios, but as far as the vmware driver goes: 0x27 microcode for best performance or the 0x39 driver for better stability?

Or did you find a way to eliminate the microcode completely?

Also are there options to change turbo MSR's on the fly in windows 10 or is there a workable "idle" setting possible from the bios when there is no CPU load?
I think I read about someone who was able to use idle state for no load?

Same chip E5-2696 V3 on Asus Deluxe II Motherboard.

Thanks very much!

B9C

Hi! I bought a X99-DELUXE II too. Please, can you share your modded bios? I'm afraid do it myself. Thanks!

 

[Sean J Sauve]

Hy all! I'm new here!

system: Z10PE-D16 WS with 2x 2683v3 (306F2, C0/C1, QS)

i struggeling with the stability and performance of the hack!

i used a flashed bios with only one microcode inside (0EF6F11F 2016/10/07)
first it hangs at boot screen (CPU CATERR), i solved this with disabeling C6 state and limited to C2
after 30min the pc will freez again with CATERR.....

i also done some testing with vray benchmark which is relly good for testing with high end processors
0:45 sec BiosMod
0:46 original 3407 Bios
0:41 original 3407 + 104 Bckl (CPU-Z Multi: 34275)

i used Bios + V3.EFI from this thread https://www.hardwareluxx.de/community/f101/all-core-turbo-dual-xeon-e5-1163969.html

a note, my SSD is in BOOTMNGR mode not UEFI so i'm only tried it with the temp solution via shell

are there any suggestions?

Thanks for your help

phil

Try the the other v3 variants composed of different optimization settings that supposedly depending on your individual processor gives different results; with respect to the work of the authors ask for more details as I too am not fully settled on the what I perceive to be the most optimized settings for my composition.

posted here :

https://www.sendspace.com/file/21y5mp

 

[Sean J Sauve]

Cinebench R15 score (5000) was done with two CPU. With 2 x E5-2658 v3 QS, I had 2800 without mod and 3300 with mod. As you mentioned about power usages, I checked mine and it said that 60W per CPU. And, I know that this CPU is rated 145W. Even fully loaded, it said that 60W is being used. I have no idea why it say that.

Also, the reason you believe that you have lower number on CPUz is because you are comparing the score from old version. The website's change logs for new release are below.

• Intel X299 (1.79.1).
• AMD ThreadRipper (1.79.1).
• AMD Ryzen 5 and Ryzen 3.
• Fix lockup at loading on Ryzen systems with RAID.
• New benchmark version (17.01)

And, I ran two versions on my system and saw the different numbers due to different benchmark method.

http://postimg.org/image/bp7feard9/

YOUR RIGHT ; CPU-z has indeed changed the performance scheme. and this highlights how I was mislead thanks and sorry but best get the message out that people should actually be seeking different values now.

Although has anyone any idea how to correct the somewhat sagged single core performance ? I seems to always be coming in soft from the litmus from proper version comparison.

Sincere that you to yourself and the guru's in this community, hats off

 

[traderjay]

Cinebench R15 score (5000) was done with two CPU. With 2 x E5-2658 v3 QS, I had 2800 without mod and 3300 with mod. As you mentioned about power usages, I checked mine and it said that 60W per CPU. And, I know that this CPU is rated 145W. Even fully loaded, it said that 60W is being used. I have no idea why it say that.

Also, the reason you believe that you have lower number on CPUz is because you are comparing the score from old version. The website's change logs for new release are below.

• Intel X299 (1.79.1).
• AMD ThreadRipper (1.79.1).
• AMD Ryzen 5 and Ryzen 3.
• Fix lockup at loading on Ryzen systems with RAID.
• New benchmark version (17.01)

And, I ran two versions on my system and saw the different numbers due to different benchmark method.

http://postimg.org/image/bp7feard9/

Hey James - thanks for sharing your experience and I am wondering how did you mod the bios for your supermicro motherboard? I have the following system and wondering if it is possible to mod the bios to achieve higher turbo:

Motherboard: X10SRL-F (latest bios)
CPU: Intel XEON E5 2696 v3

 

[sciff]

Hi! I bought a X99-DELUXE II too. Please, can you share your modded bios? I'm afraid do it myself. Thanks!

Here: https://yadi.sk/d/aTzUlTl43KbFXy (it's version 1701, use BIOS Flashback only)

I highly doubt it is an tdp or tjmax or whatever limit, as Ive the the small 18core cpu "2686v3", and I am far away from my tdp levels or tjmax's (I use an oil cooling keeping whole environment on stress on max on 30°C, and even with "hacked mc" I keep under 105 Watt at full load).

That's impossible. I have a slightly more powerful E5-2696 V3, and it consumes up to 170-200 W in regular applications and up to 240 W when AVX2 extensions are being used. You are clearly looking at a wrong reading, look somewhere else.

 

[sciff]

I now have 2 x E5-2696 v3 on Supermicro X10DRL-i motherboard. I previously had 2 x E5-2658 v3 and it could reach max turbo speed (max x29, 2.9Ghz) on both processors. However, with E5-2696, I can only reach x32 on all cores. If I reduce # of cores in BIOS, say 10 cores, It seems that I can get max turbo speed on all cores. But, this will reduce # of cores available.

So when you write "only x32", you think that it's a bad result? You're greedy, man

By the way, if I had your motherboard, I would be reluctant to try unlocking two E5-2696 V3 on it. They consume a lot of power (up to 240W in AVX2 applications if you don't use any microcode and up to 200W in regular applications) and this can contribute to VRMs overheating. Supermicro server boards don't seem to have decent VRM cooling including your board, and sometimes their VRMs don't even have a radiator sitting on them (isn't that so in your case?)

I may settle on this unless there is new tweak coming out.

If I was you, I would use V3x2_cup39.efi which doesn't have -50mV vccin offset and has microcode built into it, therefore one CPU's TDP can never exceed 200W and probably won't even reach 180W

 

[sciff]

OK so I tried v3x2_50_39_vcc.efi instead of v3. My cores all ramp up to 32x !!! BUT I am only getting 7950 on CPUz on a 18 core E5 2699 v3 ... Something is way off, I mean it did boost a little but ultimately nothing too special. Like the others in this forum..

It's like when you overclock a desktop CPU. They differ from each other, it's always a lottery. x32 is still very good, it's +4 multiplier increase from x28 all-core.

Though I must have missed something? I can't figure out what. The EFI driver DID say something about a mere 30x on UNCORE ? Which is something I don't understand to be honest.

Uncore means everything except cores in a processor. Ring buses, L3 cache, memory controllers, etc. They work at x30 multiplier in your case.

By the way, which CPU do you actually have? Is it 2696 or 2699?

 

[Zladimir]

That's impossible. I have a slightly more powerful E5-2696 V3, and it consumes up to 170-200 W in regular applications and up to 240 W when AVX2 extensions are being used. You are clearly looking at a wrong reading, look somewhere else.

Readings are real. Beside monitoring current demand I also use an external plug which gives me the same increase in current demand compared without having deleted mc, but also in comparison to idle current demands. Its all about numbers. Your model is just an bumped up version of mine, that current demand explodes towards up is pretty normal.

However Im not here to prove or argue something. Im more interested of the source files from randien. Really want to look into them.

 

[sciff]

Zladimir, I also use a socket monitoring device. What's your CPU's power consumption in idle at all-core x12? Mine is 35W with v3x2_50_vcc.efi

Didn't randir express plans to start Github project on this?

 

[Zladimir]

@sciff, I sit in LFM on ~14-15 W. Also should mention OS changed consumption. Now sitting on win10 64x cu2 (thread sheduler and power saving model was changed). With pre-cu1 I was on 18-19 Watt. Also using no microcode. Noticed constant 10-15 Watt usage increase with mc (look my first post in this forum). Dont know if board (x99 taichi) plays a real role here too, but might. All C-states are activated.

Have to mention that with randirs I dont get real readings on monitoring app, as its broken with its efi. Apps reading are from first v3.efi. Current demand from plug varies only in decimal place compared to randirs efi.

Anyway are such difference and fluctuations inside of normal tolerance. I wouldnt bother about it, as only your heat dissipation is more of an interest. You cant even tell how valuable the readings of an plug are, especially as the fault tolerance drastically increases to a lower value of measurement. Thats a basic of measurement engineering.

However, back to randirs efi. Yes he said he would want to start a Github project, but he waited for the allowness of the previous creator of our efi mod. This was already 6 weeks ago and since then randir didnt show any presence in this forum, sadly. Like me he came only for this project.

 

[James Kim]

Hey James - thanks for sharing your experience and I am wondering how did you mod the bios for your supermicro motherboard? I have the following system and wondering if it is possible to mod the bios to achieve higher turbo:

Motherboard: X10SRL-F (latest bios)
CPU: Intel XEON E5 2696 v3

Hi traderjay,

I have currently have X10DRL-i motherboard. And, I am sure that X10SRL-f is not much different from it. And, I am sure that you can unlock your e5-2696 v3 chip to 3.2Ghz(Spike to 3.8Ghz) all cores with modded BIOS and a EFI file instead of 2.8Ghz.

I have plan to upgrade my Lenovo TS440 system to X10SRL-f (that you have) soon and I have all parts and modded BIOS file. I didn't flash it yet since I just received the motherboard (Newegg was selling it for $175, Open box) yesterday. And, I will use e5-2658 v3. I was able to achieve 2.9Ghz all cores (12C, 24HT) with that chip on X10DRL-i motherboard.

If you can wait a couple of day, I will test the modded BIOS on my X10SRL-f and I will let you know. Or, you can mod it yourself using the tools available in this forum. Just grab the latest UBU and mmtool.exe. X10DRL-i and XSRL-f BIOS files should be very easy with the tools without any hiccup. Other Supermicro BIOS files seems to have issue with the tools and require manual mod using Hex editor.

Also, you may need to change CPU power settings to get optimal performance.