What controls Turbo Core in Xeons?

Page 15 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Welsper

Member
Mar 5, 2017
25
18
81
Pretty sure I saw them disabled by default in my various systems. Any reason in particularly you are trying to get them to work? They show up frequently in changelogs of BIOS releases trying to fix bugs, straight up through Skylake.
I've about 100W from 2x CPUs with only C1 state on idle.
http://s018.radikal.ru/i519/1703/cd/255cdc3b6a31.png
With C6 state it's about 40W on idle.

I have the same with c3/c6 state. No chance with enabled.
After I load the v3x2.efi with success , my BIOS boot again and the efi isn't loaded.
"load" - just check that all ok. For loading at boot: copy from usb flash to local hdd and "bcfg driver add..." from local.
 
Last edited:

Ace123

Junior Member
Jan 23, 2008
21
2
66
Ok after modifying the instructions a bit to fit my specific case I was able to make this work on the epc612d4i itx board. I have an e5 2683v3 and whIle not under hard loads the cpu is at 3ghz. During cinebench r15 runs the cpu jumps around from 2.7ghz to 2.8ghz. I assume TDP limit? Chip stays real cool as its watercooled.
I'm using dufus v3 and 29 microcode. Would uninstalling all microcode in windows allow for stronger turbo clocks under heavy load? Heard a couple people mention this.

Also freakin stellar work coming up with this guys.
 

Ace123

Junior Member
Jan 23, 2008
21
2
66
I really wish I could but this little board has about the most stripped down bios being a little server board lol. Both microcodes act almost identically. Uninstalling and running no microcodes it looks like causes more cores at all times to load and jump around to 2.8ghz even when just sitting and under full load they all stay around 2.8ghz but actually score in benchmarks just shy of running either of the two microcodes which both seem to jump around between 2.7 and 2.8ghz under load
 

Vuk Djordjevic

Junior Member
Mar 13, 2017
4
1
16
Hi Cekim did you try some render benchmark for the 2696 v3 on 3.1ghz all core turbo? Would you be kind to try the Corona benchmark if you haven't with the 2696 v3? > https://corona-renderer.com/benchmark/?cpu-type=2686+v3&submit=Search

I am interested in buying this cpu but would like to know if the 3.1ghz turbo can be sustainted or the AVX will actually lower it down to 2.8 default in the render benchmark. Jann told me that his 2686 even though clocked at 2.8ghz all core turbo actually goes down to 2.4ghz in the corona benchmark.
 

cekim

Member
Mar 6, 2017
87
19
41
Hi Cekim did you try some render benchmark for the 2696 v3 on 3.1ghz all core turbo? Would you be kind to try the Corona benchmark if you haven't with the 2696 v3? > https://corona-renderer.com/benchmark/?cpu-type=2686+v3&submit=Search

I am interested in buying this cpu but would like to know if the 3.1ghz turbo can be sustainted or the AVX will actually lower it down to 2.8 default in the render benchmark. Jann told me that his 2686 even though clocked at 2.8ghz all core turbo actually goes down to 2.4ghz in the corona benchmark.
I bricked my windows drive playing around this weekend and that box is busy crunching numbers in linux during the day, but I can re-image windows later and try this evening.

I can already tell you that you should definitely expect AVX to down-clock, but it was doing to do that even more so before the mod, so the real question is to what you are comparing this chip?

With some more use I am finding this BIOS mod essentially brings the v3 up to near v4 performance and with faster memory timings (only possible on x99 not C6xx/dual systems) can exceed v4 performance in some instances. So, avx down-clocking is all relative to what the alternative would do.

BTW: that's for "high-core-count" loads. The 2696's 3.8GHz means it out-performs the 2690v4 in low-core-count/single threaded loads:
ex: sqlite select avg(some_col) from some_table - walk an unindexed table with 40,804,545 rows (in an even bigger DB):
2690v4:
Run Time: real 45.000 user 42.030330 sys 4.638871
2696v3 (mod w/standard CAS15 memory):
Run Time: real 40.000 user 36.964254 sys 3.522384

Both of those runs were on the same NFS mount over 10GbE to a mount that can deliver sustained 800-900MB/s throughput (and a warm nfsd cache in both cases).
 
Last edited:

cekim

Member
Mar 6, 2017
87
19
41
So AVX is definitely down-clocking significantly. Corona shows all cores (HT enabled) @2.7GHz

This is uCode 0x27 which does perform a little better than 0x36 in win10

CAS 15 BCLK 100
HT disabled
1. Cinebench
2112 3.1GHz all-core
2. Corona
1:50 2.7GHz all-core

HT Enabled
1. Cinebench
2664 3.1GHz all-core
2. Corona
1:19 2.7GHz all-core

The question I can't easily answer now is what is the clock without the bios mod? 0x27 and a stripped bios produce a very unstable system WITHOUT V3.EFI. So, I'd have to un-strip my bios to test that. Anyone have a stock 2696v3 they can check? Might be able to check that later this week.

1:19 is roughly in-line with what appears on Corona's website as a good score for a normal (BCLK 100, un-modded) 2696v3.
 

Vuk Djordjevic

Junior Member
Mar 13, 2017
4
1
16
Hi Cekim thnx for the test, much appreciated! That I know a stock 2696 v3 should have 2.8 Ghz all core turbo from what I have seen on ebay cpu-z screenshots that sellers post. But anyway that resault is good and in line with the rest of the results on corona benchmark :). Thanks again for doing the benchmark.
 

Alex S

Member
Mar 14, 2017
32
9
41
3dnews.ru
Hello. Can i ask for your help, guys?
We are from russian resource 3DNews.ru, and we are planning to experiment with one of these multi-core turbo-boosted Xeons. But we want to use dual socket board (C612 chipset). Is there any luck on these and which model can you recommend? We simply plan to use 2 unlocked processors a bit later and compare this multi-threaded monster with ordinary systems, so we need dual LGA 2011-3 board. That would be interesting - test 28C/56T OCed monster (2x14 core Xeon, for example) against standard setups.
 
  • Like
Reactions: custom90gt

Welsper

Member
Mar 5, 2017
25
18
81
V3x2_mc39.efi:
https://ufile.io/1b08f

It does this:
1) update boost multipliers on each found CPU (as V3x2).
2) update CPU(s) microcode with revision 39. (which integraded into efi file, so it's ~40KB now)

Profit: OS independed microcode update.
You don't have to remove mcupdate_GenuineIntel.dll and install CPU microcode with VMWare tool for example (Win10).

dual socket board
I've success with "ASRockRack EP2C612 WS"
 
  • Like
Reactions: Mercado

ondoy

Member
Oct 1, 2012
50
0
66
V3x2_mc39.efi:
https://ufile.io/1b08f

It does this:
1) update boost multipliers on each found CPU (as V3x2).
2) update CPU(s) microcode with revision 39. (which integraded into efi file, so it's ~40KB now)

Profit: OS independed microcode update.
You don't have to remove mcupdate_GenuineIntel.dll and install CPU microcode with VMWare tool for example (Win10).


I've success with "ASRockRack EP2C612 WS"

does this work on supermicro motherboards ?
 

Alex S

Member
Mar 14, 2017
32
9
41
3dnews.ru
I have no any way to check.
Will ASRock Rack EP2C612D16C-4L work with this? This model is a way cheaper (less PCIe slots).
If anyone has luck with Supermicro X10DAL or DRL - let us know, please. These are cheap, compact, and ebay selling a lots of them. ASUS seems nice, but these are too pricey, we must spend no more than 1100-1300 EUR on first phase (because we need memory and two CPUs, or we can test MB with one CPU, then later with both).

The idea of project is to create stable and fully usable workstation 28C/56T class (depends on CPUs we'll use.) And then we'll test it and publish performance results. (Maybe games too, but only on Fury X in 1440p - currently i have no other card).
 

custom90gt

Member
Feb 9, 2017
38
7
41
I updated in Ubu, my latest BIOS, ASRock X99 oc formula, and use of user skip all the steps for the Haswell 0 for microcode update.
I copied the new USB BIOS stick, fat 32 format default, along with others in the shell folder.
I restart and gave the stick to boot, but when you give the command load fs0: \ V3.EFi is not found or does not recognize any command or v3.efi
What is the cause?
Can you help me?

Your drive mapping is likely different.

Use:

map -b

that will show you all of the drives.
Say you want to see if fs1 is the right one then you would do the following:

fs1:
ls -b

If your V3.efi is there then you are golden, otherwise you'll have to look through the other file systems...
 
  • Like
Reactions: lucien_br

custom90gt

Member
Feb 9, 2017
38
7
41
Im a newbie. Cam you be more specific? I taste load fs1:\V3.EFI?

Not sure of the best way to explain this...

Basically the FS# is like a different drive letter. You know when you go to computer in windows you have c: d: e: etc drives?

The command "map -b" will display all of the available "drive letters". Which one contains your V3.efi and which one has your windows boot information in it is up to you to find.

To go into a "drive letter" you type in "fs#:" and that will take you into that drive

From there you can type "ls -b" to see what contents are in each drive. For instance if you only had v3.efi in drive fs3: after typing ls -b you will see "V3.efi" listed there. That will tell you what fs# your v3.efi will be in. Use that # in the command that the instructions provide.
 

custom90gt

Member
Feb 9, 2017
38
7
41
i understand, was fs7.
i Update microcode in windows, i delete mcumpaupdate, genuine windows and replace with x39 microcode
but still the cores are to 2583.
I miss somethings?

When you load v3.efi does it say all core enable or whatever? Your bios isn't loading microcode for the cpu right?
 

planeswalker

Junior Member
Mar 15, 2017
2
2
81
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 :)
 

ondoy

Member
Oct 1, 2012
50
0
66
I've made my own V3x2.efi
https://ufile.io/541182
( Unlock boost for E5-26xx-v3 multiprocessor systems, based on original V3.efi ).

PS: It is my first experience with uefi programming... it works, on my PC at least... but why I have got x28 on the first and x30 on the second CPU... I'll check tomorrow.

have you managed to look into this ?
i'm also having this issue 1st cpu is @ 2.8ghz 2nd cpu is @ 2.9ghz
 

custom90gt

Member
Feb 9, 2017
38
7
41
it say image nr ..... loaded with succes.
a da capo to repeat the process with bios of all
now I know what to do
back with details
every time, i must boot like this?

Unless you complete the rest of the steps and copy it to your hard drive and have it load as a device...
 

cekim

Member
Mar 6, 2017
87
19
41
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)
Help that 6950x out with an OC. No reason to leave it languishing at 3.5GHz... It handles 4.0 without breaking a sweat and should be able to go to 4.2-4.3 with a little more effort. 4.4/4.5 are harder but depending on your chip and configuration may be possible. I have mine setup to do 4.5 under 4 or less core load and 4.4 for 5-10 core load. 2696 @ 3.8GHz does well for single/few thread loads - quite a bit better than my 2690v4, but the 6950x beats them both for such things.

I had been planning on a dual socket system for 2x2696, but for compute needs, I'm finding 2 modded 2696 systems are more useful than 1 2x2696. I lose cool benchmark results, but get more work done.
 
  • Like
Reactions: planeswalker

cekim

Member
Mar 6, 2017
87
19
41
after I boot into Windows, I wiped genuine intel system mcumupdate 32,
I shoved and microcode with mcumupdate 27microcode, but show me the speed hwinfo 2591 x25
I just like XEON E5 2686 OC V3 INSTRUCTIONS but something goes wrong
can the two microcode are not good, because I have E5-2683 v3-xeon, wich is OEM
excuse me for my English worse
If you are seeing 0x25 uCode after your efforts there are 2 likely reasons:
1. your bios mod didn't really remove the bios uCode. Have you confirmed that you see uCode 0 in the bios after the bios mod step?
OR
2. your windows uCode install via the vmware tool didn't work. (or you used the wrong .dat file?) Not sure where you would have gotten 0x25 without some hunting other than #1 or a stale version of windows.

Deleting the mcupdate file in windows 10 is tricky, it defaults to a high level of security that required me to change ownership and permissions before I could even delete it. "As Administrator" was not sufficient to rename this file prior to running the vmware tool to install the new uCode. Also, the vmware tool needs to be run as administrator as well I believe (I did not try it without that).

Speaking of which, which version of windows are you using?
 

Andi64

Junior Member
Sep 27, 2007
4
0
61
Hi,

I've flashed the modified Asrock X99 Xtreme4 UEFI and it's booting fine. Problem is, when I boot to UEFI and execute V3.EFI the output says:

V3 - Wrong CPU
Image 'FS5:\V3.EFI' loaded at 5A99D000 - Success

I have a Xeon V3 12C/24T Engineering Sample. Any ideas?

Thanks id advance.