What controls Turbo Core in Xeons?

[TLN]

I've finally registred on Anandtech because of that topic.
Wonder how it works in Linux/ESXi, is that possible to activate Turbo all cores? I guess I have to update BIOS and add driver in hypervisor, correct?

 

[Tritonk7]

@Dufus

Hello, is it possible to replace the code for the E5-2670 CPU to turbo worked on all 33 cor

 

[igan]

Basically Intel produced Haswell CPU's with an bug / errata which allows full turbo on all cores. This is fixed up with a microcode update.

have v1, v2 xeon this bug?

I have linked the file 'V3.EFI' (V3.zip 633bytes) which can be copied to a FAT32 USB flash drive.
...
Don't forget to have Windows or the vmware utility update the microcode, version 0x38 works well for me.

What about EFI and microcode VmWare files for V1(SandyBridge) and V2 (IvyBridge)?

 

[Ionstream]

After digging through the ucode in my bios, I've found two locations where the ucode is stored. I did this by looking up the hex values of the ucode in UEFITool.

According to the site, you're only supposed to mess with one ucode, if there are two present. Can anyone confirm this?

 

[Dufus]

@katlen the 2863v3 supports upto 240W for upto 40 seconds. Can also be bypassed as The Stilt has done.

@Gaha AFAIK this bug is only for Haswell (Xeon v3 processors).

@Ionstream RWEverything shows microcode version 0x36 already in effect so too late to change to all turbo mode. Most of what you see there are MTRR registers for describing memory types vs address range, if you want to see other registers then use the 'user' button. For instance add '0x1ad=0x1ad' for turbo values 1-8 or if you want a better description then 'Turbo 9-16 = 0x1ae'. What does MSR 0xCE and 0x194 show?

@Tritonk7 Personally I do not know how to change a non v3 or even if it is possible.

@igan AFAIK v1 and v2 do not have this bug. As far as microcode files for vmware are concerned for Intel it just uses the microcode.dat file that is provided and updated by Intel. Alternatively any microcode can be adapted to this format to be used by vmware.

 

[KAL2017]

I also registered because of this topic. Going to give this a go with the dual Xeon motherboard that I ordered. Should be fun. I still need to order the Cpu's and ram tomorrow. I have everything else already. This is an upgrade from my dual E5 2670 build.Yes it does have the AMI Aptio UEFI bios tho.

 

[katlen]

After digging through the ucode in my bios, I've found two locations where the ucode is stored. I did this by looking up the hex values of the ucode in UEFITool.

According to the site, you're only supposed to mess with one ucode, if there are two present. Can anyone confirm this?

How likely is by doing changes like this to brick the bios?
If you have a spare motherboard you could give it a try.

 

[fromax]

Hi! Much respect to @Dufus, @C-Power/Tw0tch, @The Stilt for this topic and to someEEguy for step by step instruction
I have MSI X99A SLI PLUS and xeon 2673 v3 and I would like to try this mod. Sadly UBU doesn`t support msi bios so have had to erase ucode from bios manually with MMtool (just erased the ucodes from Cpu path section and save bios). Flashed it and motherboard didn`t boot up (MSI has two bioses so no problem for experiments). So could someone give me a proper instruction how to earse ucode manually or make such modification for me and share bios https://msi.com/Motherboard/support/X99A-SLI-PLUS.html#down-bios
I would appreciate any help)

 

[Ionstream]

Hi! Much respect to @Dufus, @C-Power/Tw0tch, @The Stilt for this topic and to someEEguy for step by step instruction
I have MSI X99A SLI PLUS and xeon 2673 v3 and I would like to try this mod. Sadly UBU doesn`t support msi bios so have had to erase ucode from bios manually with MMtool (just erased the ucodes from Cpu path section and save bios). Flashed it and motherboard didn`t boot up (MSI has two bioses so no problem for experiments). So could someone give me a proper instruction how to earse ucode manually or make such modification for me and share bios https://msi.com/Motherboard/support/X99A-SLI-PLUS.html#down-bios
I would appreciate any help)

Using MMTool to remove uCodes will not work if it's an AMI Aptio V bios. You have to do it manually using UEFITool.

EDIT: MMTool cannot add in new bios modules, and it may have some problems with other bios modules.

 

[escksu]

Hi all, I just registered because I saw this thread.

Any chance of overclocking the 2699v4 or 2696v4 Xeons on Asus Z10PE-D8 WS? I bought 2 2696v4 but could not oc it. It always drop to 28x on all cores. IS there anyway to make it stay at max boost (37x) on all cores?

I am willing to try anything.

Thanks.

 

[The Stilt]

Doesn't work on Broadwell. It works on Haswell due to a bug, which was already fixed long before Broadwell was released.

 

[escksu]

Oh i see. Thanks for the reply!! Is there a chance that the bclk strap will work for broadwell?

 

[escksu]

Another question is how about 2699v3? will it work? Thanks!!

 

[GTZ]

Oh i see. Thanks for the reply!! Is there a chance that the bclk strap will work for broadwell?

From my personal experience, I was only able to clock both of the ES V4s I had to around 102 BCLK. I believe the busses on Broadwell are a lot pickier that Haswell E. On the other hand my 2683v3 I am able to do 104-105 completely stable. Both V4s I tested were early ES's, don't know if that was a factor. My 2683v3 is QS and is the same stepping as retail.

Another question is how about 2699v3? will it work? Thanks!!

It should work on all Haswell E Xeons.

 

[escksu]

http://www.intel.com/content/dam/ww...cification-updates/xeon-e5-v4-spec-update.pdf

Hi, I was reading up on this. There is one error which states that cores may be unable to reach Maximum Turbo Frequency. It affects cpus with more than 10 cores. There is no fix for this and workaround is through BIOS.

Since turbo frequency is adjusting the voltage and multiplier, I am thinking it may be possible for BIOS to override the default settings. But I don't know if it can increase beyond the default values.

 

[The Stilt]

Another question is how about 2699v3? will it work? Thanks!!

It will work, but it won't sustain the full 3.6GHz on all cores. Something between 2.8GHz - 3.3GHz depending on the used microcode & workload should be pretty accurate.

 

[GTZ]

Just wanted to make sure to thank everybody that contributed to this. This is truly innovative and thought it would never be possible. Thank you Dufus for not keeping this to yourself and sharing it to the world. Learned a lot from this thread, I am sure glad I decided to keep my V3.

I just wanted to point out a few things that happened to me and don't know if it applies to everybody or isolated to me.

If you only remove the microcode from the BIOS and have MCE options in the BIOS and set all cores turbos to the maximum the OS will freeze and hardlock. This happened to my AsRock Extreme 4 on any BIOS 2.X (Versions 3.X took away MCE controls). This still happened with the latest microcode installed on Windows. This was easily solved with running the latest BIOS and the UEFI script provided. Regardless if your BIOS has MCE controls you should still run the script. Solved all my issues.

Also for the past few days I could not figure out why the automated script would only work if I press a key (any key) during boot. In my case I had to disable fast boot.

Again thanks

 

[escksu]

Hi, is it possible to load another microcode into the bios?

I saw this MEF406F1_0B000017 has the BDF67 error. So just want to give it a try.

Thanks

 

[mrpiggy]

I want to give a huge shout out and THANK YOU to all you guys (especially Dufus)!!! This is awesome! Works great on my E5-2690v3 (12-core 3.5Ghz Turbo, also pushing bclk to 105) with an ASRock x99 Extreme6.

The CPUz bench went from low 17000-ish to 21000-ish. Even more huge is my Cinibench score went from 1907 to 2175! Huge increase!

Was quite the learning curve in getting it to work, but jumping around everyone's instructions eventually worked (although I had the added step of converting Windows 10 Legacy/MBR to EUFI/GPT)..

 

[TLN]

It will work, but it won't sustain the full 3.6GHz on all cores. Something between 2.8GHz - 3.3GHz depending on the used microcode & workload should be pretty accurate.

Wonder what is "max" CPU that can run "all core Turbo" without dropping?
2683 runs 14cores @ 3.0.
Can you run 2686 18 cores @ 3.5? Probably will drop to 3.0, same way as 2699.

 

[mrpiggy]

Well for a point of reference, my 12 core E5-2690v3 will keep all the cores at full 35 turbo speed in something like Cinibench (or CPUZ's multproc benchmark) and you can see in perfmon that all the cores are pegged. That being said, when I run something like Prime95 torture test, every core instantly drops to a 29 multiplier, and then starts quickly dropping lower and lower to bottom out at about 26-27 mix multiplier. . Oddly enough, after 3-5 minutes at 26-27 mix and the core temps have crept up to a max of 58C (I could run it cooler, but I like my fan silent), the multiplier actually goes back "up" to constant 27-28 mix multiplier.

 

[C-Power/Tw0tch]

@Dufus Is the real hero of this thread, thanks to his driver/script
Obviously @The Stilt as well for major contributions

Anyway as far as my 2683v3 goes, it sticks at 30x multi (and a 105.1 bclck) whatever the load is I throw at it, and I LOVE it

This entire thread inspired me to dig deeper into how microcodes and drivers work, apart from the "simple" overclocks I usually do with unlocked i7 CPU's.

Unfortunately for v1 - v2- v4 users, it seems this "bug" only works for V3 (Haswell) Xeons.

I am seriously inspired now into writing my own drivers/scripts/EFI boot code - and hopefully I can contibute more to this thread soon (tm)

 

[srsparky32]

@Ionstream

Noticed you were working on this with a C612 chipset. I'm running a Supermicro X10SRA with C612, but with a v4 14c xeon. so I'm not able to utilize this change, but I'm considering picking up a 10C v3 to try this out with, and would love to hear about any progress you make.


@C-Power/Tw0tch

Respectfully, how do you know if the bug only works for v3 xeons? has someone else already been trying with v1/v2/v4 xeons?

 

[escksu]

@Dufus Is the real hero of this thread, thanks to his driver/script
Obviously @The Stilt as well for major contributions

Anyway as far as my 2683v3 goes, it sticks at 30x multi (and a 105.1 bclck) whatever the load is I throw at it, and I LOVE it

This entire thread inspired me to dig deeper into how microcodes and drivers work, apart from the "simple" overclocks I usually do with unlocked i7 CPU's.

Unfortunately for v1 - v2- v4 users, it seems this "bug" only works for V3 (Haswell) Xeons.

I am seriously inspired now into writing my own drivers/scripts/EFI boot code - and hopefully I can contibute more to this thread soon (tm)

Yes, Thanks for the info!! I do understand v4 does not have this bug but I am still not giving up hope yet. I will hunt for more info on this errata BDF67.

But its really fun and something totally new to me. Its totally different from the usual desktop CPU overclocking. I am reading and learning right now. Hopefully one day I too might be able to contribute something.

 

[escksu]

I also need a bit of info. Whats the difference between 2696v4 abd 2699v4. Google around and people say 2696v4 is OEM version. But I could never find any info on it. Its not even listed on Intel website or in their documents.