What controls Turbo Core in Xeons?

Page 87 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

ajschot

Member
Feb 16, 2017
65
2
16
I don’t understand... with my E5-2696v3 i dis all and used some different efi drivers but my gain is almost nothing compared to without a efi driver.
Did some of you run geekbench?
I get in osx 3200 and 28000 and windows 3300 and 30000 is this a good value? I see a lot higher values in the geekbench scores with 1 cpu.
Am i doing something wrong?
 

sciff

Member
Mar 6, 2017
136
52
71
Sorry I am maybe stupid.... i try to edit my bios Asus ROG Strix X99 Gaming
steps 1-3 are done.... then i have a problem i don't understand step 4
Also... i need to remove MSR 0xE2 (for macOS) so i want to do a inbetweenstep between 4 and 5 (i think..) so Remove MSR 0xE2 with UEFIPatch.

Do you think this is the right order?

Also do i need to build my own EFI driver for E5-2696v3? Until now my E5-2683v3 on cheapest MSI board gives me almost same geekbench scores multicore then my E5-2696v3 on my ASUS RoG x99 Gaming (with v3x2_cup27.efi)

Also how do i get it back into a CAP file? I used BRenamerl but it just renames the file to .CAP but it is still a rom file not a AMI Aptio capsule
You will get your .bin file back to CAP if you open it in UBU, then type "0" for exit and choose 1 - Rename to ASUS USB BIOS Flashback

Regarding step 4, I described what needs to be done and how in one of my earlier posts:

mike999, you have to edit the original BIOS file in a hex editor first and only then pass it through UBU.

Use search function inside it and search for "2E06A0" hex value and then edit the file by adding 4 right below "2E" and subracting the same value three to the right. This way securecode feature of the BIOS file is disabled. But be careful, the searched part should be towards the end of the file and there can be two instances (in case of ASUS BIOS there usually are two), edit both. If you also find "2E06A0" somewhere in the beginning or in the middle of the BIOS file, ignore it, it's just a coincidence.


In this case, as a result I got 18 instead of 14 and 08 instead of 0C
P.S. don't forget to agree to rename the .CAP file for ASUS Flashback, when you get asked by UBU at the very end.
 

axlimaging

Junior Member
Dec 13, 2017
2
1
41
After hours of trying settings I have the modification working on an e5 4667 v3.


Here are some of the problems that I encountered and how I solved them. Maybe it's useful for someone else:

4. AVX workloads were still restricting the CPU to normal speed! I experimented with other microcode and EFI driver versions and found the following combination of settings to work best:

The exact one I'm using is v3_705020vcc1.8.EFI. This one magically solves the AVX problem... but not on it's own, because:

4.2 Based on sciff's idea that AVX speed may somehow depend on the BCLK I increased the BCLK slightly, to 102. A very small change such as 100.2 didn't work and 103 made my system unstable. Only after increasing the BCLK did the AVX multiplier go up.

4.3 Microcode: Version 27 supposedly gives the best speeds but my multiplier under AVX load actually went up from 28 to 29 when I updated the microcode to 39.
I'm now getting the maximum turbo frequency on all cores with prime95 and other programs that use AVX so there's nothing more that can be done do to improve the performance of this CPU.
I have the same CPU as you, on a Gigabyte X99P SLI using CANONKING's -50mv BIOS

I find that I can hit the x29 multi on all cores just fine, however any AVX load drops that one down to x28. Running 102 BCLK makes up ~50mhz so that's not too bad. So ~2950mhz becomes ~2850mhz which is damn near close enough to full time 2900mhz for me.

I've tried:

Microcode 3A & 39 - only drops 1 multi (29 to 28)*
Microcode 27 - cores all over the place, didn't stay anywhere near the max turbo at all, constantly fluctuating from 23 to 29 on all cores*
No microcode - dropped down from 29 to 23 when any AVX load (Photoshop/Camera RAW, linpack for example) was placed on the processor*

*I was up till midnight last night testing, I might be remembering wrong, pretty sure this is what was happening

I could only get the CPU to drop 1 multi after the -50mv bios was used. I see you have used a -70mv file so that's probably the next thing for me to try to get the full 29 multi for AVX loads.

Mind if I ask what you get in Cinebench R15 with your CPU? I'm ~2200 points, which seems to be about what an un-modded 2686 is pulling.
 
  • Like
Reactions: traderjay

ajschot

Member
Feb 16, 2017
65
2
16
You will get your .bin file back to CAP if you open it in UBU, then type "0" for exit and choose 1 - Rename to ASUS USB BIOS Flashback

Regarding step 4, I described what needs to be done and how in one of my earlier posts:
Thanks...
Did that but when a CAP file is extracted into ROM you can not go back even not with ubu.
I tred like at least 100 times. So i leave the ME disabling for what it is. rest is no problem.
 

ajschot

Member
Feb 16, 2017
65
2
16
You are a hero... i don't understand what i am doing wrong.... i think i need to check my bios... last time i tested in windows was almost a year ago. And in macOS was just last week when i had to do a new install and found out that without a EFI drive i got same result then with...
Maybe a year ago i had soemthing wrong in my bios i will check. Otherwise i will proberbly sell my Xeon E5 2696v3

EDIT: today test
https://browser.geekbench.com/v4/cpu/6292818
 
Last edited:

rottwag

Member
Apr 2, 2017
77
11
41
Hi,

1)
Is there a risk, that the Computer does not come up at all with a -70mV Bios Version when it is too less Voltage what then causes, that i will not be able to flash any other BIOS? I have an Asus X99 (no II and no 3.1)

2)
I currently use the on older "non microcode" BIOS where nothing else is injected"... therefore the .EFI drive is loaded separately from a GPT Partition.
Does it make sense to Keep it like this? I am now able to Change the EFI drive without the need to reflash my BIOS again and again with New versions. what do you think?

3)
Is there a Link somewhere with BIOS (no microcode but injected EFI for -50, -70mV etc) somewhere for my Asus x99 Board where I can download the .CAP files?

PS: 2 weeks left until my 2695 V3 arrives from China... .-)


Best regards
 
Last edited:

kniaugaudiskis

Junior Member
Jun 4, 2009
18
3
81
Another more delicate question! Can we get to go with E5 2696 to x35 all core ever?
Yes I can get my E5-2696V3 to that speed, the only problem is that it becomes quite unstable. I get 3.5GHz on all cores if I use -90mv efi driver and push my BLCK to 103MHz. If only it was somehow possible to make the CPU run at that speed without undervolting it...
 

Dinkydau

Member
Apr 1, 2012
50
5
71
Yes I can get my E5-2696V3 to that speed, the only problem is that it becomes quite unstable. I get 3.5GHz on all cores if I use -90mv efi driver and push my BLCK to 103MHz. If only it was somehow possible to make the CPU run at that speed without undervolting it...
Yes, why is this even required? Does the CPU detect a maximum current or something? If that could be manipulated it would be pretty good, but I would also be afraid it could damage the CPU. There may actually be good reasons to set such limits. Unfortunately intel confuses us with unnecessary limits.

And I was informed incorrectly...
https://security-center.intel.com/advisory.aspx?intelid=INTEL-SA-00088&languageid=en-fr

Of course, here's what *not* being shared by Intel:
SIMPLY DISABLING MANAGEMENT ENGINE (ME) IS ALL THAT IS NEEDED TO DEFEAT THIS THREAT
https://github.com/corna/me_cleaner
  • The network stack (partition NFTP) is removed
  • The PAVP (partition MDMV, module JCOM) is removed

To summarize: Intel would rather force a performance loss on you, the duped consumer, than they disable "Management Engine" in their fix and give up their back-door to your system.
Which threat are you talking about? There is a performance loss when using the fix for meltdown but that has nothing to do with the ME as far as I know.
 

rottwag

Member
Apr 2, 2017
77
11
41
hi all

one question: is air cooling sufficient for 14 cores@ 3,4Ghz or do i need a special cooling?

just asking because i get 70Celsius with my 8Core@ 3,4Ghz running prime with my "brocken 2" as cooling.

what do you use? bequiet dar rock pro 3? or noctua Dh14/15?
 

ajschot

Member
Feb 16, 2017
65
2
16
Really can't get to there i do not understand.... which efi do you run?
I now tried plain v3x4.efi
https://browser.geekbench.com/v4/cpu/6432258

Disabled C states and put most tweaker setting in bios on auto only disabled multicore enhanchement. Also put in a BLCK of 104, disabled EIST and VMX and that is pretty much it. Rest is mainly optimized settings. But results stay very low only single core is ok.
 
Last edited:

ajschot

Member
Feb 16, 2017
65
2
16
Yes I can get my E5-2696V3 to that speed, the only problem is that it becomes quite unstable. I get 3.5GHz on all cores if I use -90mv efi driver and push my BLCK to 103MHz. If only it was somehow possible to make the CPU run at that speed without undervolting it...
wow! i only get sometimes 3,1GHz but never higher. What are your bios settings?
 

timk1980

Member
May 11, 2017
25
1
41
wow! i only get sometimes 3,1GHz but never higher. What are your bios settings?
The big limiting factor is usually how much power your motherboard is able to delivery, along with the individual idiosyncrasies of the specific processor(s) you are using. That's why undervolting helps to get an extra 1-2x on the multiplier: the available current is enough to go one step higher.

Best I can do on my with all 18 cores enabled is the processors going to 31/32x. Because of how various multi threaded stuff works, at least in windows, I dialed back in the BIOS to only use 16 cores on each processor (64 total threads). With no u-code, I can get 32/33x that way, though with u-code 0x27 I get 31/32x.
 

kniaugaudiskis

Junior Member
Jun 4, 2009
18
3
81
wow! i only get sometimes 3,1GHz but never higher. What are your bios settings?
All BIOS settings regarding the CPU are all defaults, except for disabled CPU spread spectrum, C6 and C3 states. Oh and my motherboard is ASUS X99 Deluxe II, I run u-code 39 + power cut + -60mv on vcore and -50mv on cache. This configuration yields a 33x multi and combined with a 3MHz BLCK overclock I get a stable 3.4GHz OC on all 18 cores and 3.9GHz when few cores are loaded.
 

glitchman

Junior Member
Sep 23, 2017
11
0
11
All BIOS settings regarding the CPU are all defaults, except for disabled CPU spread spectrum, C6 and C3 states. Oh and my motherboard is ASUS X99 Deluxe II, I run u-code 39 + power cut + -60mv on vcore and -50mv on cache. This configuration yields a 33x multi and combined with a 3MHz BLCK overclock I get a stable 3.4GHz OC on all 18 cores and 3.9GHz when few cores are loaded.
Could you please check your WHEA count? Is it 0?
 

foureight84

Member
Nov 5, 2017
57
8
41
And I was informed incorrectly...
https://security-center.intel.com/advisory.aspx?intelid=INTEL-SA-00088&languageid=en-fr

Of course, here's what *not* being shared by Intel:
SIMPLY DISABLING MANAGEMENT ENGINE (ME) IS ALL THAT IS NEEDED TO DEFEAT THIS THREAT
https://github.com/corna/me_cleaner
  • The network stack (partition NFTP) is removed
  • The PAVP (partition MDMV, module JCOM) is removed

To summarize: Intel would rather force a performance loss on you, the duped consumer, than they disable "Management Engine" in their fix and give up their back-door to your system.

Take your system back and DISABLE Management Engine!

ME ENABLED:


All but critical ME services DISABLED (only 'SiEn' or Silicon Enable is ENABLED):
does the patch work with all bios? I have the same motherboard as you, asus z10pe-d8 ws and there's no options in the bios to do that without patching
 

ASK THE COMMUNITY