What controls Turbo Core in Xeons?

[n31l]

You should definately try 'v3x2_50_39_vcc.efi' with microcode 39 built in or 'v3x2_50_vcc.efi' without a microcode.

https://drive.google.com/open?id=0B4nDSxzsKIeOWk1NenhDSEFRbnM - I uploaded them but have no credit

You should delete microcode and 'mcupdate_GenuineIntel.dll' from system32 first.
If you delete the microcode but you have the 'mcupdate_GenuineIntel.dll' present in the system32 folder, it will automatically load ucode 36 as we see in one of your pictures. (hwinfo - at the top - uCU - 36) If you have no microcode loaded it will show 'uCU - '

how to delete an 'undeletable' file (mcupdate_GenuineIntel.dll): https://www.howtogeek.com/howto/windows-vista/how-to-delete-a-system-file-in-windows-vista/

Thanks for the suggestions.. I'll try them tonight...

I just tried loading the usual V3.efi and letting windows load it's usual mcupdate_GenuineIntel.dll (I took ownership and just added .old when testing) now I have 3199 but without the performance loss... but not much gain over default.. interesting..

 

[n31l]

Thanks for the suggestions.. I'll try them tonight...

I just tried loading the usual V3.efi and letting windows load it's usual mcupdate_GenuineIntel.dll (I took ownership and just added .old when testing) now I have 3199 but without the performance loss... but not much gain over default.. interesting..

Managed to try with v3x2_50_39_vcc.efi & v3x2_50_vcc.efi -- both results are roughly the same as V3.efi and using the 'mcupdate_GenuineIntell.dll' in 'system32'..

 

[Gellért]

Managed to try with v3x2_50_39_vcc.efi & v3x2_50_vcc.efi -- both results are roughly the same as V3.efi and using the 'mcupdate_GenuineIntell.dll' in 'system32'..

You should delete microcode first.
On the upper picture you can see you have microcode 27 (uCU - 27) present. If you load 'v3x2_50_vcc.efi' after you deleted the microcode you should have no microcode loaded ( uCU - )
Maybe this also interacts with the 'v3x2_50_39_vcc.efi' which loads uCode 39 (but with the ucode27 present I don't know if it's all good or not)

Thing is even after you do this I don't think you would achieve more.
As far as I know with v3x2 efis - the system doesn't have to be in performance mode all the time like with v3. Correct me if I'm wrong.

 

[cekim]

Can you disable some cores in the bios on the asus board like on the asrock one?
Then you should get higher frequencies even at stock. (without the 'hack')
I mean I get ~3,4ghz with 18 cores but with only 10 cores enabled I get ~3,9 Ghz on them.
That is what I'm interested in if you activate only 10 cores/cpu you would get the same result with 2 cpus or not.

I could, but I'm already seeing 3.8GHz on 9+9 cores (an 18 core compute job spread over 2 processors). So, I wouldn't gain anything by disabling cores other than lack of access to them for jobs that can use all 36 cores.

In fact, at 50% system load (18 cores 100% utilized), I am seeing all 72 threads at ~3.8GHz. Linux shows them bouncing around by small margins depending on second-to-second load and the performance governor.

So, short version is the system is behaving exactly as I'd want it to. If only using 18 of 36 cores for actual work, it is running at 3.8GHz (the 2696v3's max with bclk=100). If I need 36 cores for actual work, it will drop to 3.3-3.4, but that's expected and entirely tolerable.

 

[cekim]

You should delete microcode first.
As far as I know with v3x2 efis - the system doesn't have to be in performance mode all the time like with v3. Correct me if I'm wrong.

No it doesn't, I am able (in linux) to load the power-save governor and see the expected behavior (idling down to 1200MHz). Interesting tough, the system is in a higher performance state by default with (particularly) the 50_39_vcc EFI MSR writes...

In a stock system (without the ucode exploit), my system comes up by default in the power-saving mode and idles down to 1200MHz quickly. Changing nothing in the OS, but enabling the ucode exploit and EFI driver, it comes up in a performance mode characterized by a few cores at 3.0-3.8 even when the system is relatively quiet/idle.

I would expect the same from windows. With the exploit + EFI drivers, you will find a system using higher idle power unless you intervene and set a power-saving profile.

 

[Gellért]

No it doesn't, I am able (in linux) to load the power-save governor and see the expected behavior (idling down to 1200MHz). Interesting tough, the system is in a higher performance state by default with (particularly) the 50_39_vcc EFI MSR writes...

In a stock system (without the ucode exploit), my system comes up by default in the power-saving mode and idles down to 1200MHz quickly. Changing nothing in the OS, but enabling the ucode exploit and EFI driver, it comes up in a performance mode characterized by a few cores at 3.0-3.8 even when the system is relatively quiet/idle.

I would expect the same from windows. With the exploit + EFI drivers, you will find a system using higher idle power unless you intervene and set a power-saving profile.

I see. I have the cores at ~1,2ghz at idle (normal mode - not performance nor power saving) (windows - taichi - 2696 v3 - 50_vcc.efi or 50_39_vcc.efi)
Withouth microcode though I have to wait a bit after logging in to windows to have the cores at idle because they stay at ~2.3 ghz for a while.

 

[cekim]

I see. I have the cores at ~1,2ghz at idle (normal mode - not performance nor power saving) (windows - taichi - 2696 v3 - 50_vcc.efi or 50_39_vcc.efi)
Withouth microcode though I have to wait a bit after logging in to windows to have the cores at idle because they stay at ~2.3 ghz for a while.

Windows is noisy... runs various things in the background that aren't always, shall we say, full optimized...

So, that's pretty normal. I believe they claim to have made some decent gains here recently with the latest big release (creator's edition or whatever they called it). They've been busy trying to do a better job setting such things to lower priorities to avoid spinning an idle machine and wasting battery/power.

I also see different behavior between Taichi single boards and Z10PE dual boards even with the same EFI/ucode. Can't explain why yet. The dual board remains "twitchy" (quick to ramp up to higher frequencies) by default where the taichi boards run more "normally" and get back to idle.

The twitchy behavior is actually better for performance, but does mean this system consumes more power at idle than it could otherwise.

 

[n31l]

You should delete microcode first.
On the upper picture you can see you have microcode 27 (uCU - 27) present. If you load 'v3x2_50_vcc.efi' after you deleted the microcode you should have no microcode loaded ( uCU - )
Maybe this also interacts with the 'v3x2_50_39_vcc.efi' which loads uCode 39 (but with the ucode27 present I don't know if it's all good or not)

Thing is even after you do this I don't think you would achieve more.
As far as I know with v3x2 efis - the system doesn't have to be in performance mode all the time like with v3. Correct me if I'm wrong.

I've tried all combinations, even going back down a bios revision (after removing microcode) however after resetting to defaults after flash and booting without inserting the efi drivers HWinfo still reports '27', if I unrename the windows 'dll' to allow Windows to load it then it shows 36, if I load 'v3x2_50_39_vcc.efi' it shows 39.. not sure how '27' keeps being persistent but it doesn't appear to effect performance, same as '36' and '39' don't appear too either..

Had some random crashes, mainly when I've been away from my PC for a while so I might add a little to the vCore - with Turbo unlocked and all cores stressed the max CPU core temp is 50c, so stability is not cooling related.

 

[cekim]

I have moved on from msi raider to attack ep2c612 ws have 2 e5-2683 v3 this thing a beast

What method of bios mod did you use if you don't mind me asking?

I'm collecting parts to try this board out as well. I have a ch341a writer, so I was going to manually edit the bios since this board has no flash-back, but if UBU.bat works and the bios will load it without complaint, then that's a little easier.

 

[James Kim]

Hi there,

With many helps here, I was able to unlock turbo on all cores.

I have Supermicro X10DRL-i with two E5-2658 v3 QS processors. The CPUs temp. in idle around 45C with air cooled. And I am using v3x2_50_39_vcc.efi driver. It's been stable for three days now with a couples of hyper-v guest OSes.

For my bios, I removed the v3 micro codes using the latest UBU 1.68 with mmtool 5.0.0.7. It was very simple process without hiccup. Then, flashed it. Set C3/C6 disabled. And then, put efi driver into flash drive and load it at the boot time. And, load OS. Since there is hyper-v host server that stay many weeks without reboot, I don't mind to load the driver manually.

I usually don't mess with BIOS because I had bad experience before. And, I am not expert on this. But, there are so much useful information here that I could learn. It was too good to be true if I could unlock the turbo.

Before this tweak, I had 2800 scores on Cinebench R15 benchmark. Now, I have 3300.

 

[Cata40]

I'm a computer enthusiast, but I'm a beginner. All I've done is because of this thread and some of the participants here. I got the ownership at mcupdate_1GenuineIntel.dll, I also made photo in task and idle, but looking in system 32, mcu along with mcupdate_1GenuineIntel.dll, I found other packages with mcupdate. Do those packages, do not even have that microcode 27, let go to full turbo x35 or 36? I ask myself the question.....

 

[cekim]

What method of bios mod did you use if you don't mind me asking?

I'm collecting parts to try this board out as well. I have a ch341a writer, so I was going to manually edit the bios since this board has no flash-back, but if UBU.bat works and the bios will load it without complaint, then that's a little easier.

Answered my own question:
EP2C612-WS + 2.1bios ran through UBU without issue and is now up and running. updated from 1.3 to 2.1+ubu bios via ezflash without issue.

Interestingly, same CPUs (2696), ucode (0x39), EFI (50_vcc) as the Z10PED16-WS, but all-core turbo is 3.2 rather than the 50/50 mix of 3.3/3.4 that the Z10 gets...

_something_ about the asrock board is behaving the same as it did with taichi and different than the ASUS board.

 

[XDrego]

Answered my own question:
EP2C612-WS + 2.1bios ran through UBU without issue and is now up and running. updated from 1.3 to 2.1+ubu bios via ezflash without issue.

Interestingly, same CPUs (2696), ucode (0x39), EFI (50_vcc) as the Z10PED16-WS, but all-core turbo is 3.2 rather than the 50/50 mix of 3.3/3.4 that the Z10 gets...

_something_ about the asrock board is behaving the same as it did with taichi and different than the ASUS board.

well I was waiting for 2.13b bios chip asrock is sending me but if you want to help mod it ill send you the bios I have no ch341 if intant flash working that's great!

 

[XDrego]

well I was waiting for 2.13b bios chip asrock is sending me but if you want to help mod it ill send you the bios I have no ch341 if intant flash working that's great!

you disable c6 c3 states

 

[cekim]

well I was waiting for 2.13b bios chip asrock is sending me but if you want to help mod it ill send you the bios I have no ch341 if intant flash working that's great!

I just used UBU.bat on 2.1. No ch341a required, though I did confirm I could use it on the taichi.

Right this moment I'm having issues with CSM. This board (like many) is somewhat/strangely random as to whether it boots in UEFI of legacy mode mode, so naturally I disabled CSM... Asmedia VGA did NOT like that at all. It hangs and beeps 3 times at code d6 (video init failure).

Odd because ASUS uses the same chip and it has no issue with non-CSM boot there.

 

[cekim]

you disable c6 c3 states

I did and I needed to. Hangs in the usual spot if not.

 

[XDrego]

I did and I needed to. Hangs in the usual spot if not.

have an email send you bios

 

[XDrego]

I enable above 4g decoding

 

[cekim]

I enable above 4g decoding

hmmm... usually don't need it, but I suppose if they put the vga chip above 4G? That would be an odd choice, but possible... I'll try that and I need to move the file to the right place to send it to you, but I'll do that shortly. Thanks.

 

[XDrego]

hmmm... usually don't need it, but I suppose if they put the vga chip above 4G? That would be an odd choice, but possible... I'll try that and I need to move the file to the right place to send it to you, but I'll do that shortly. Thanks.

thanks good luck with 2.13b enables cluster on die

 

[cekim]

Fixed my CSM issues and the on-board video (4G didn't seem to matter, but I fiddled around with Legacy vs UEFI (set UEFI only and storage = uefi and vga = legacy) I must admit on the surface the settings didn't make sense (to have UEFI only, but then vga = legacy), but hey... it works!

Now, under 18 core load (dual system 36 cores), I see a screen full of this:
Every 1.0s: grep -i mhz /proc/cpuinfo | grep ": 3" Wed Jun 7 00:57:58 2017

cpu MHz : 3767.777
cpu MHz : 3799.941
cpu MHz : 3799.941
cpu MHz : 3799.941
cpu MHz : 3799.941
cpu MHz : 3799.941
cpu MHz : 3799.941
cpu MHz : 3792.035
cpu MHz : 3799.941
cpu MHz : 3628.609
... and so on

however, performance doesn't match the clock rates (the matching Z10PE system is running an 18 core job in 20% less time. hmmm... more checking to do.

 

[cekim]

ok... now we're cooking with gas... 2.13b bios ran through UBU.bat without issue. Manually set memory to 2133 and now the EP2C612-WS is edging out the Z10PE slightly... < a few % in a real 18-core application.

The EPC2612 is still stopping hard at 3.2GHz all-core, but its performance on 18 core is (as above) as good as the Z10PE that is getting 3.3/3.4 (50/50 mix)

EDIT: oops! now that I'm running full-throttle, these 3U heat sinks aren't gonna do it... 76C! (That's why I went to 2xH80i v2 on the Z10PE... Same thing there.

 

[Gellért]

ok... now we're cooking with gas... 2.13b bios ran through UBU.bat without issue. Manually set memory to 2133 and now the EP2C612-WS is edging out the Z10PE slightly... < a few % in a real 18-core application.

The EPC2612 is still stopping hard at 3.2GHz all-core, but its performance on 18 core is (as above) as good as the Z10PE that is getting 3.3/3.4 (50/50 mix)

EDIT: oops! now that I'm running full-throttle, these 3U heat sinks aren't gonna do it... 76C! (That's why I went to 2xH80i v2 on the Z10PE... Same thing there.

I'm thinking about buying the z10pe d8 or 16 from asus.
Don't know which one. I only have 4 ram sticks 16gb/stick atm. The d16 has that 'ipmi' built in while with the d8 it's a separate purchashe.
Would buy an other 2696v3 for that.
So you are experiancing same performance with the asrock regardless of the lower clockspeed right? Interesting. What if even the taichi is the same like the x99 ws from asus regardless of the lower mhz
So much to consider

 

[cekim]

I'm thinking about buying the z10pe d8 or 16 from asus.
Don't know which one. I only have 4 ram sticks 16gb/stick atm. The d16 has that 'ipmi' built in while with the d8 it's a separate purchashe.
Would buy an other 2696v3 for that.
So you are experiancing same performance with the asrock regardless of the lower clockspeed right? Interesting. What if even the taichi is the same like the x99 ws from asus regardless of the lower mhz
So much to consider

d16 vs d8:
1. ram
d16 has more dimm slots, but unless you buy the much, much more expensive LRDIMMS, you won't be able to use them without losing performance. RDIMMS allow only 1 dimm per channel (4 dimms per socket) before DDR4 clock rate is lowered.

So, unless you already have or have easy access to LRDIMMS, you likely won't be able to make use of the extra dimm slots. You are better off getting 8x16G sticks for a total of 128G or 8x8G sticks for a total of 64G of memory.

2. PCIE slots - d8 has more, so if you ever want to do 4xGPU, the d8 will be better as the d16 can only do 3-way SLI owing to fewer physical PCIe slots.

3. ipmi - do you intend on using it?

Z10 vs EP2C (with 2.13b bios):
1. PCIe - compare to d8 - can do 4x GPUs where d16 can only do 3.
2. performance:
a. clocks - all-core
- The Z10PE-D16-WS is producing all-core turbo for me at 50% 3.3 and 50% 3.4GHz (36 threads show 3.3 36 show 3.4 under all-core load).
- The EP2C612WS is showing 3.2 for 72 threads under all-core load
b. clocks - heavy real load (not synthetic)
- both show ~3.8GHz for 18-20 core jobs - real work bounces all over the place - indistinguishable by eye-ball
c. here's some geek-bench runs for you to judge - easiest thing for me to run right now (no windows)

Z10PE-D16-WS
https://browser.geekbench.com/v4/cpu/3047882
single: 4274 multi: 74436

EP2C612WS
https://browser.geekbench.com/v4/cpu/3047980
single: 4335 multi: 73304

single 2690v4 (3.5GHz chip) for further reference:
https://browser.geekbench.com/v4/cpu/3047837
single: 3803 multi: 39893

4.4GHz 6950x for even more silly fun:
https://browser.geekbench.com/v4/cpu/3048053
single: 4527 multi: 37496

Frankly, I'd call those (z10 vs ep2c) dead even (well within noise...). That sort of difference could be anything in either one of these two systems such as background processes (network, nfs, etc...)

If I had to wager, I'd bet the Z10 would put up tiny little fractions more in something like cinebench on THESE TWO SPECIFIC systems, but I don't see anything here that conclusively says you'd see the same other than the slightly higher all-core turbo.

Lastly, the EP2C612 was slightly easier to mod - run UBU, flash bios vs Z10 which is hand-edit bios, run ubu, flash bios.

You'll do fine with any of them, the EP2C is cheaper, so there's that...

EDIT: the EP2C 2.13b bios gave me a performance bump, but I'm not sure at this point whether it was the bios or something else I did... The 2.13b bios is NOT available on their website yet... So. factor that into "easy to mod".
EDIT: Confirmed - you will need the 2.13b bios to get performance parity with the ASUS board.

 

[XDrego]

d16 vs d8:
1. ram
d16 has more dimm slots, but unless you buy the much, much more expensive LRDIMMS, you won't be able to use them without losing performance. RDIMMS allow only 1 dimm per channel (4 dimms per socket) before DDR4 clock rate is lowered.

So, unless you already have or have easy access to LRDIMMS, you likely won't be able to make use of the extra dimm slots. You are better off getting 8x16G sticks for a total of 128G or 8x8G sticks for a total of 64G of memory.

2. PCIE slots - d8 has more, so if you ever want to do 4xGPU, the d8 will be better as the d16 can only do 3-way SLI owing to fewer physical PCIe slots.

3. ipmi - do you intend on using it?

Z10 vs EP2C:
1. PCIe - compare to d8 - can do 4x GPUs where d16 can only do 3.
2. performance:
a. clocks - all-core
- The Z10PE-D16-WS is producing all-core turbo for me at 50% 3.3 and 50% 3.4GHz (36 threads show 3.3 36 show 3.4 under all-core load).
- The EP2C612WS is showing 3.2 for 72 threads under all-core load
b. clocks - heavy real load (not synthetic)
- both show ~3.8GHz for 18-20 core jobs - real work bounces all over the place - indistinguishable by eye-ball
c. here's some geek-bench runs for you to judge - easiest thing for me to run right now (no windows)

Z10PE-D16-WS
https://browser.geekbench.com/v4/cpu/3047882
single: 4274 multi: 74436

EP2C612WS
https://browser.geekbench.com/v4/cpu/3047980
single: 4335 multi: 73304

single 2690v4 (3.5GHz chip) for further reference:
https://browser.geekbench.com/v4/cpu/3047837
single: 3803 multi: 39893

4.4GHz 6950x for even more silly fun:
https://browser.geekbench.com/v4/cpu/3048053
single: 4527 multi: 37496

Frankly, I'd call those (z10 vs ep2c) dead even (well within noise...). That sort of difference could be anything in either one of these two systems such as background processes (network, nfs, etc...)

If I had to wager, I'd bet the Z10 would put up tiny little fractions more in something like cinebench on THESE TWO SPECIFIC systems, but I don't see anything here that conclusively says you'd see the same other than the slightly higher all-core turbo.

Lastly, the EP2C612 was slightly easier to mod - run UBU, flash bios vs Z10 which is hand-edit bios, run ubu, flash bios.

You'll do fine with any of them, the EP2C is cheaper, so there's that...

EDIT: the EP2C 2.13b bios gave me a performance bump, but I'm not sure at this point whether it was the bios or something else I did... The 2.13b bios is NOT available on their website yet... So. factor that into "easy to mod".

you used instant flash correct