What controls Turbo Core in Xeons?

[kjboughton]

My problem is that i have QDE4 engineering sample and it is 4770K.

Post your CPUID or simply a CPU-Z screen shot if you are unsure what is being asked of you.

 

[kjboughton]

Hello i have a problem
then i try to Load v3.efi in to shell it says Wrong cpu but i have a E5 2650 v3 but is this because its an es model

Post your CPUID or simply a CPU-Z screen shot if you are unsure what is being asked of you.

 

[Zladimir]

Thanks for answer. I ran last night sandsifter for you. Here is the file, it is most efficient compress that I have ever done, 238 Mb > 2,3 Mb

sandsifter_instructions

Tell me if you need the log file with different settings.

Thank you! When I have time I look into it.

Balanced Performance perhaps allows C3 at the core level (which cannot be stopped with the in-BIOS Package limit) and BSOD can result.

For me it did not solve the bug. Anyway its not only the BSOD you will face problems with. See above mentioned bugs. Much earlier post in this thread already discussed this issue and suggestions kept disabling C3.
However as you already said, most of the trick does state C1. That is all what is required for IntelSpeedstep to work. There is no problem to lower in high performance mode the idle frequency to 1%. You ll have the power savings and maximum performance without any unresolved bugs.

Nearest I can tell EDP is a hardwired (fused) limit set at time of binning must like how multipliers are set.

It should be fused. However if you dig into the technical papers there are many points which may raise your eyebrows. Intel clearly defined ways to change how the EDP is determined, and even disabled. Further going from no microcode to latest, Intel changed twice the EDP calculation for bug resolving. So if its in the code, its code (logic, he?^^).
Similar I think about features like bclk strap and unlocked multiplier. First is an explicit Xeon Haswell feature, but the MSR is gone and the cpu gets into a lockdown, if you try to set it. Last one is clearly recognized on many engineering samples, but setting the multiplier just does nothing. This leads to (my) believe that a locked multiplier and a gone strap are last minute features, which may be done on a software side.

 

[korzychxp]

Post your CPUID or simply a CPU-Z screen shot if you are unsure what is being asked of you.

And i can't change multiplier.

 

[kjboughton]

And i can't change multiplier.

This is not Haswell-E

 

[korzychxp]

I know but maybe we can do the same with it like with Xeons

 

[er557]

MAYBE if grandma had wheels ,she could be a skate board.


ON TOPIC: @Zladimir
On my setup, 2x e5 2686 v3, asrock rack, I ENABLE both c3 and c6, only the setting c0/c1 is enabled instead of c6 retention, to avoid lock up on windows boot or otherwise bsod.
I load latest windows microcode, such as 3D or 3C, and never had bsod with c states enabled. I pass with flying colors linpack xtreme on Linux boot cd, the most challenging stability test.
I used to undervolt both cpus at -90,-60,-50, which used to bsod once a month or so, so I lowered the stakes and now fully stable at undervolt -80,-50,-50. Cinebench R15 result is 5000.

 

[xeon_fan]

Hi does anyone has already modified Asrock Taichi x99 bios in version 1.80 ? I ask cause I already was using the one in 1.40 but due adding new nvme drive I need new bios

Thanks in advance!

[I removed microcodes from v1.80 via two different ways -but even they don't show up in MMTool when I load the V3.EFI i get information:
V3 - Microcode present.

Update:

I just made weird thing, wonder if this will work but I took bios in version 1.80 and just updated all microcodes to newest version except ...
CPUID 306F2 which downgraded to 0x27 ... wonder if this will work

... it did not work neither ... this even shows in windows that I'm using uC 0x27 but sill not working and if I install V3.EFI it shows that microcodes are present ...

@kjboughton -

...

Could I ask you for the Taichi x99 bios which you gave to shorty1111 ?

Thank you

 

[foureight84]

Could I ask you for the Taichi x99 bios which you gave to shorty1111 ?

Thank you

here you go. made one for you. removed haswell microcode: http://www.mediafire.com/file/vrvi4ruah3uovv0/mod_X99TC_1.80/file

 

[xeon_fan]

here you go. made one for you. removed haswell microcode: http://www.mediafire.com/file/vrvi4ruah3uovv0/mod_X99TC_1.80/file

Thank you ! I'm going to deploy it now

Update:

Weird, I checked and the microcodes are not in the given bios yet ... it did not work:

https://ibb.co/iCqk3L

When I load v3.efi it says the Microcodes are present.

On second bios I have version 1.40 and the v3.efi does work there ... yet both checked bioses shows exactly the same status in MMtool for microcodes.

 

[ziollos]

Hello everyone,

What pair of xeons E5 v3(V4) do you recommend to get max performance after OCing (I will be using them for poker solving calculations which means it's 100% load)?

 

[foureight84]

Hello everyone,

What pair of xeons E5 v3(V4) do you recommend to get max performance after OCing (I will be using them for poker solving calculations which means it's 100% load)?

V4 won't work with this all core turbo mod. As for the V3 I believe the 2696v3/2699v3 will get you the highest overclock speed of about 3.9ghz on all cores. I believe I saw that that chip has 38 multipliers so you would only need to push your bclk to 102.6mhz to get 3.9ghz.

I am currently running 2667v3 which has 36 multipliers and I am able to get 3.8ghz overclocked with a bclk of 105.6. Which is pushing my motherboard's stability -- for this use @kjboughton's base driver without voltage offset. Else you will have stability issues. He and I are both running on the Asus z10pe-d8 ws. It's a solid motherboard.

I am currently using this for work and gaming and I do find that with 1080ti SLI, you will still have CPU bottleneck with 3440x1440 (https://github.com/freecableguy/v3x4/releases/tag/v3x4-0.10b-i306f2-rc9). I don't know if the 3.9ghz or maybe even 4ghz (if that's doable) on the 2699/2696 would help SLI for gaming. If you're looking for a gaming rig, it's probably better to get a second hand i9-9740x or 9760x. Much higher oc capability.

It really does suck that the multiplier on these xeons are locked. I wish there was some hack to unlock them. I am pretty sure they can handle the oc.

 

[foureight84]

Thank you ! I'm going to deploy it now

Update:

Weird, I checked and the microcodes are not in the given bios yet ... it did not work:

https://ibb.co/iCqk3L

When I load v3.efi it says the Microcodes are present.

On second bios I have version 1.40 and the v3.efi does work there ... yet both checked bioses shows exactly the same status in MMtool for microcodes.

That's strange. Not sure what's going on there. I could try again tomorrow and give you the file. Which version do you need? Link me

 

[raun0]

V4 won't work with this all core turbo mod. As for the V3 I believe the 2696v3/2699v3 will get you the highest overclock speed of about 3.9ghz on all cores. I believe I saw that that chip has 38 multipliers so you would only need to push your bclk to 102.6mhz to get 3.9ghz.

But if you have a m.2 SSD drive you could stuck under 101 Mhz BCLK, like me.

i7-5930k is bang for the buck to gaming with 2011-3 socket. It is 4,7 Ghz OC capable if you are lucky. Clock-to-clock perfermance is awsome compared to newer skylake and etc.

 

[foureight84]

But if you have a m.2 SSD drive you could stuck under 101 Mhz BCLK, like me.

i7-5930k is bang for the buck to gaming with 2011-3 socket. It is 4,7 Ghz OC capable if you are lucky. Clock-to-clock perfermance is awsome compared to newer skylake and etc.

I actually have 5 m.2 nvme drives. All Samsung 970 evo

 

[ziollos]

V4 won't work with this all core turbo mod. As for the V3 I believe the 2696v3/2699v3 will get you the highest overclock speed of about 3.9ghz on all cores. I believe I saw that that chip has 38 multipliers so you would only need to push your bclk to 102.6mhz to get 3.9ghz.

I am currently running 2667v3 which has 36 multipliers and I am able to get 3.8ghz overclocked with a bclk of 105.6. Which is pushing my motherboard's stability -- for this use @kjboughton's base driver without voltage offset. Else you will have stability issues. He and I are both running on the Asus z10pe-d8 ws. It's a solid motherboard.

I am currently using this for work and gaming and I do find that with 1080ti SLI, you will still have CPU bottleneck with 3440x1440 (https://github.com/freecableguy/v3x4/releases/tag/v3x4-0.10b-i306f2-rc9). I don't know if the 3.9ghz or maybe even 4ghz (if that's doable) on the 2699/2696 would help SLI for gaming. If you're looking for a gaming rig, it's probably better to get a second hand i9-9740x or 9760x. Much higher oc capability.

It really does suck that the multiplier on these xeons are locked. I wish there was some hack to unlock them. I am pretty sure they can handle the oc.

I won't be using it for gaming at all. Only usecase will be poker solver applications which basically means 100% load all the time. I have seen on previous pages that 18 core xeons can sustain only 3.1GHz turbo on all cores during 100% load, is that still correct or something changed? What for mobo I already own ASUS Z10PE-D16 WS.

 

[xeon_fan]

That's strange. Not sure what's going on there. I could try again tomorrow and give you the file. Which version do you need? Link me

Hi foureight84, I need following bios (new NVME drivers - with version 1.40 my PC randomly freezes now with WD Black NVME drive):
ftp://asrock.cn/BIOS/2011/X99%20Taichi(1.80)ROM.zip

In my first attempt (previous post) I have removed all microcodes and re-used V3.EFI (from page 55) as I was doing with version 1.40 but it did not work

But based on following @raun0(thanks!) instruction:
https://www.win-raid.com/t3874f16-GUIDE-Haswell-E-EP-Xeon-CPUID-F-Turbo-Unlock.html

I realized I can replace using of EFI load with FFS load from bios directly [the rest like setting/disabling uC in OS stays)
Is this correct? Please correct me if I'm wrong

What I just did:

1) I have removed all microcodes and added some old [copied from yours - I realized that in my version 1.40 there were also some Broadwell microcodes - 406F1] - it was MMTool based modification

2)Then created ffs based on freecableguy (thanks!) release v3x4-0.10b-i306f2-rc9_50x3.efi from:
https://github.com/freecableguy/v3x4/releases
[generated with this instruction - https://github.com/pbatard/efifs/wiki/Adding-a-driver-to-a-UEFI-firmware]

3) then I included ffs at last position as per @raun0 instruction

Results of my changes are below - bios + screenshot [google drive contains screenshot + bios file with ffs included and the uC used in your version of 1.80 [ I don't know if that was done on purpose or not but as I'm lame I re-used them - that's are some old Broadwell uC
https://drive.google.com/drive/folders/1-vdldM28EFX0LfIij5-xBcNrgUd3gT4z?usp=sharing

Could I ask for verification ?

I will upload it on friday when I will be back at home [If I get green light that it's ok].

Thank you !

PS. I took this ffs with 3x-50 cause I don't overclock any more as I use nvme drive. My QS version of 2697 v3 was working stable on 14c - 104.5 x 36

 

[raun0]

I actually have 5 m.2 nvme drives. All Samsung 970 evo

I have EVO 960 256 GB on my own. Maybe 970 can hand more BLCK.

I must test the CPU limit with sata drive.

 

[m1919]

Got the hack applied on my Z10PE-D8 WS with dual E5-2697 v3s.

Sadly, I couldn't get to 3.9 GHz, but it's close.

Running the 3407 bios with microcode removed and the v3x2_80-50-50_39_vcc1.8.efi driver MOF posted a while back.

I left the latest microcode update from Intel in place. It seems to boost to 3.8 in most situations.

 

[xeon_fan]

Got the hack applied on my Z10PE-D8 WS with dual E5-2697 v3s.

Sadly, I couldn't get to 3.9 GHz, but it's close.

Running the 3407 bios with microcode removed and the v3x2_80-50-50_39_vcc1.8.efi driver MOF posted a while back.

I left the latest microcode update from Intel in place. It seems to boost to 3.8 in most situations.

Hey I am missing something with the ratio ? is not the e5 2697 v3 x36 max ?

 

[m1919]

Hey I am missing something with the ratio ? is not the e5 2697 v3 x36 max ?

I believe the driver is maxing out turbo and 37x is apparently the highest available multiplier for this CPU.

Before I put the hack in place, CPU-Z and AIDA64 were reporting 36x for two-core turbo. Not sure what the deal is there, but I'm not complaining.

 

[ziollos]

Got the hack applied on my Z10PE-D8 WS with dual E5-2697 v3s.

Sadly, I couldn't get to 3.9 GHz, but it's close.

Running the 3407 bios with microcode removed and the v3x2_80-50-50_39_vcc1.8.efi driver MOF posted a while back.

I left the latest microcode update from Intel in place. It seems to boost to 3.8 in most situations.

This result is amazing. I was considering buying these two chips: https://www.ebay.es/itm/Intel-Xeon-...=item2604adc264:g:FzIAAOSwDYxbsOFO:rk:14:pf:0 and possibly OC them to 3.6GHz. Are these QS chips a good idea? I was looking for 18 core xeon, but non-ES chips are still quite expensive.

 

[xeon_fan]

Hi one additional question ...
what this mean ? where did you left the microcode update ? in Windows ?

I left the latest microcode update from Intel in place. It seems to boost to 3.8 in most situations.

Thanks !

Would be great to run e5 2697 v3 [QS] at x37

@ziollos - I have QS version of CPU and it does work flawlessly !

 

[ziollos]

What is the max turbo for all cores that someone managed to achieve using E5-2698 v3 or E5-2697 v3?

 

[xeon_fan]

What is the max turbo for all cores that someone managed to achieve using E5-2698 v3 or E5-2697 v3?

I had for e5 2697 v3[QS] :

• 1-11 cores - turbo 3.6 ghz (x36)
• 12-13 cores - turbo 3.5 ghz (x35)
• 14 cores - turbo 3.4 ghz (x34)

But now I see the x37 can be even achieved ... need to check that