What controls Turbo Core in Xeons?

[corinthos]

Dual CPU boards usually route PCIe lanes out from each cpu, so depending on the number of Wellsburg chips, could be 40 (CPU Integrated IO) + 8 (Wellsburg PCH) per CPU. Typically you only see one Wellsburg per CPU though - multi-PCH setups are very unusual even in the server space. A typical dual CPU board would have 40+40+8 lanes.

The PCIe/DMI port 0 can be run as a PCIe x4 port, but I've never seen a board that routes them that way. When you have so many PCIe lanes, routing a rather bandwidth limited x4 port out isn't worth it.

Thanks, so my dual cpu c612 board likely has 40 + 40 + 8 lanes then. Even better than 64

 

[Dufus]

hi, Dufus
If you may share your tools or HOW TO, we will do the test.
There are so many 2696v3 here.

As already stated, I don't have any tools just compile something rough and recompile if wanting to change values. UEFITool, now that's a tool and an awesome one at that. I don't have the patience or perhaps the skills to do something like UEFITool. Testing would need to be done personally as it may require many recompilations to the software and/or firmware. I did look at getting someone to test with stepping 1 and 0 over on OCN but in hindsight it was probably just as well no one was interested at the time.

Hopefully over the next few months prices will come down and maybe I'll be in a position to get something like the 2696.

 

[CANONKONG]

Would you please make another BIOS with FFS MOD for ASUS X99-E WS/USB 3.1 with vcore offset -60 vcach -20. with -70 i have blue screen. I tried to make one by myself but after flashing the BIOS my pc does not post.

You can try it now：http://pan.baidu.com/s/1i4FfsHn

 

[CANONKONG]

Can you help me with a bios ffs for a single CPU, Asrock X99 OCFormula that is like this: system agent offset voltage, vcore additional offset & cpu cahe voltage offset all to -70?
Or teach me how to create an ffs!

You can try it now：http://pan.baidu.com/s/1bOhat4

 

[CANONKONG]

Please make a mod bios with ffs mod for ASUS X99-E WS/USB 3.1 with offset -50 .

You can try it now：http://pan.baidu.com/s/1misUKti

 

[CANONKONG]

As I know, not modded 2683v3 in dual CPU can make nearly 3200pts in Cinebench, OCed Threadripper 1950x can do it with 3450pts. Single 2686v3 can make 1750pts without mod, and 2100pts with mod - what is 20% more. So if You mod a dual 2683v3 You can get ~3800pts in cinebench. You have 10% more than OCed Threadripper. Only one cone, is that dual CPU consumes more power. But with dual CPU, You can also buy a 2696v3 and there will be much more advantage comparing to Threadripper. EPYC is the only better way for me

CANONKONG where I can get newer microcodes? I only have the old ones. And the second question, will it work in nearly future with V4 Xeons? I'm very interested in it.

Newer microcode：http://pan.baidu.com/s/1nuEXiJZ

 

[CANONKONG]

There are some websites,and you can know how to make efi files and ffs file
FASM（.asm to efi）：http://flatassembler.net/
EFI files writing：http://wiki.osdev.org/Uefi.inc
EFI files to FFS files：https://github.com/pbatard/efifs/wiki/Adding-a-driver-to-a-UEFI-firmware

 

[Cata40]

You can try it now：http://pan.baidu.com/s/1bOhat4

Thanks very much man, i try now bios
I'm waiting for you to post, as you said about 2 days ago, steps to make a bios as you like

 

[junkim]

You can try it now：http://pan.baidu.com/s/1i4FfsHn

Thanks a lot. Its works perfect

 

[Cata40]

Newer microcode：http://pan.baidu.com/s/1nuEXiJZ

Thanks for the bios, its ok, till now, work good.
But I would have liked to be at -70, including system agent voltage offset (not +100) like the v3x2_50_vcc.efi model, which works best for me, but I did not integrate it into bios, it's for dual cpu and perhaps that one It gives me error, it stops at number 70 on the motherboard

 

[CANONKONG]

I wrote a Guide that how to modify X99 BIOS for some want to know ，but my English is very poor, I don't know you can read it or not.
http://pan.baidu.com/s/1o8Ue3yE

 

[Dingmel]

I wrote a Guide that how to modify X99 BIOS for some want to know ，but my English is very poor, I don't know you can read it or not.
http://pan.baidu.com/s/1o8Ue3yE

@CANONKONG

Thanks for your guide.


Just a few things need clarification.

Steps 1 and 2, they're not needed for non ASrock bioses right?

For ASUS bioses, UEFItool shows 2 entries that need to be extracted as raw files and modded. Do both entries need to be modded and rebuilt or only the first one?

Lastly, how would we know if the inject has been successful, ie would there be a change in the microcode reading within the bios?

Thanks for all the hard work !!!

 

[Cata40]

I wrote a Guide that how to modify X99 BIOS for some want to know ，but my English is very poor, I don't know you can read it or not.
http://pan.baidu.com/s/1o8Ue3yE

We try to understand, thanks for your kindness
If i dont know somethink, i ask

 

[CANONKONG]

@CANONKONG

Thanks for your guide.


Just a few things need clarification.

Steps 1 and 2, they're not needed for non ASrock bioses right?

For ASUS bioses, UEFItool shows 2 entries that need to be extracted as raw files and modded. Do both entries need to be modded and rebuilt or only the first one?

Lastly, how would we know if the inject has been successful, ie would there be a change in the microcode reading within the bios?

Thanks for all the hard work !!!

Yes，Steps 1 and 2, they're not needed for non ASrock bioses.
UEFItool shows 2 entries that need to be extracted as raw files and modded. both entries need to be modded and rebuilt，because they arethe same，so use the same is OK.
you can use HxD to found the mc that the first line numbers,if you can not found it,you got it.

 

[Dingmel]

Yes，Steps 1 and 2, they're not needed for non ASrock bioses.
UEFItool shows 2 entries that need to be extracted as raw files and modded. both entries need to be modded and rebuilt，because they arethe same，so use the same is OK.
you can use HxD to found the mc that the first line numbers,if you can not found it,you got it.

Thanks for your reply. Is there anyway to confirm a successful inject besides stressing the system after flashing?

 

[CANONKONG]

Thanks for your reply. Is there anyway to confirm a successful inject besides stressing the system after flashing?

use Aida64 or hwinfo

 

[Dingmel]

use Aida64 or hwinfo

It just shows no microcode. is that accurate?

 

[n0taz]

I have, E5-2686v3 QS and E5-2696v3 QS. I am using it on Astock Taichi x99. Solo cpu socket ofcourse.

With 2686v3 and C-Payne v0.1 patch for ASROCK mainboards it worked with very high performance and stable as rock (with 27 microcode installed in system, bus 105mhz) without any changes at bios (c-states on, c1e on, spread spectrum on). For example cinebench 2700+ (and even once was 3k), corona1.3 = 1.04, geekbench4 = 32700+.

At HWInfo a saw next parameters:

IA Overclocking: Voltage offset = -70 mV
GT Overclocking: Voltage offset = 0mV
CLR (CBo/LLC/Ring Overclocking): Voltage offset = -51 mV
Uncore Overclocking: Voltage offset = +100 mV (?)

https://img-fotki.yandex.ru/get/225650/51780633.0/0_14eda8_becd03fb_orig.png

When changed 2686v3 to 2696v3, it seems that he worked no so fast as it need

For example, stable working with C_Payne v0.1 asrock patch only when bus no more than 102 Mhz and bios energy saving futures are off (c1e, spread spectrum, c-states). And results at tests very bad. With 102 Mhz stable, but 2500+ at cinebench. If i want more, for example 3000 at cinebench - i use 103.5 bus but some applications didnt working

Do u have any advices for me? I think at first i need try to use v3x2_50_39vcc for maxmum results with maaximum stability? ( i can try it only after a week, now i am in hospital).

p.s. sorry for my english.

 

[n0taz]

I have, E5-2686v3 QS and E5-2696v3 QS. I am using it on Astock Taichi x99. Solo cpu socket ofcourse.

and more information - with this mainboard (Taichi) - i can't working without microcode - system freezes after 1-2 minutes of working. It was with variuos processors - 2669, 2686, 2696. It think it's asrock's fiture

 

[timk1980]

I have, E5-2686v3 QS and E5-2696v3 QS. I am using it on Astock Taichi x99. Solo cpu socket ofcourse.

With 2686v3 and C-Payne v0.1 patch for ASROCK mainboards it worked with very high performance and stable as rock (with 27 microcode installed in system, bus 105mhz) without any changes at bios (c-states on, c1e on, spread spectrum on). For example cinebench 2700+ (and even once was 3k), corona1.3 = 1.04, geekbench4 = 32700+.

At HWInfo a saw next parameters:

IA Overclocking: Voltage offset = -70 mV
GT Overclocking: Voltage offset = 0mV
CLR (CBo/LLC/Ring Overclocking): Voltage offset = -51 mV
Uncore Overclocking: Voltage offset = +100 mV (?)

https://img-fotki.yandex.ru/get/225650/51780633.0/0_14eda8_becd03fb_orig.png

When changed 2686v3 to 2696v3, it seems that he worked no so fast as it need

For example, stable working with C_Payne v0.1 asrock patch only when bus no more than 102 Mhz and bios energy saving futures are off (c1e, spread spectrum, c-states). And results at tests very bad. With 102 Mhz stable, but 2500+ at cinebench. If i want more, for example 3000 at cinebench - i use 103.5 bus but some applications didnt working

Do u have any advices for me? I think at first i need try to use v3x2_50_39vcc for maxmum results with maaximum stability? ( i can try it only after a week, now i am in hospital).

p.s. sorry for my english.

I'm not using the same board as you, however, did have this hack in place for my system on a different processor before obtaining 2x E5-2696v3. There's probably a few things going on, many of them related to power consumption, some unfortunately just inherent in the chip itself.

BLCK ability is partly something that simply depends on the CPU in question... some behave better than others... if 102 is the best you can get, that might be where it has to stay.

You might have an issue where the undervolt is simply too aggressive for your chip. Trying some of the drivers that don't undervolt, or only do so by maybe -20, might give you some different results.

Now, in my experience, when using all 18 cores, the frequency boosts were very hit and miss, and individual cores would downclock then boost again pretty haphazardly. As you might imagine, this wasn't good for most of my benchmarks. With 18 cores being operated beyond their originally labelled frequency range, the CPU is needing more power than it's probably getting. Using one of the _vcc drivers to effectively neuter the power monitoring in the CPU will probably help this a bit. However, do be VERY SURE you have good cooling on the chip, as it can potentially pull 300+Watts. (Using my dual-CPU setup, if I do some heavy benching in Prime95, I can make my UPS alarm go off for my system pulling more than 900W total... insae)

You can also try increasing Load Line Calibration (LLC) in the bios, assuming it offers that option (most x99 should)... may help a bit.

Something else to try, if only for experimentation, is in the BIOS setup, reduce the number of cores in use. By default it's set on 0 (all cores), but if you manually set it to use less cores, you may actually find better--or at least more consistent and less troubling-- performance. I run my CPUs with only 16 cores. Both Cinebench and Corona are *marginally* lower scoring (42s instead of 39s in corona, and 4600 instead of 4900 in cinebench), but consistent reliable system performance is greatly improved. Though it's not great as a benchmarking tool, Intel Burn Test works nicely as a quick stability check, and actually highlights the frequency stuttering issue I see pretty easily. Currently, I get all-core (16) boost ratios of 32x and 33x when running non-avx workloads (IBT for example), and ratios fluctuating between 30-32x in something like Corona.

Let us know how it goes.

 

[timk1980]

and more information - with this mainboard (Taichi) - i can't working without microcode - system freezes after 1-2 minutes of working. It was with variuos processors - 2669, 2686, 2696. It think it's asrock's fiture

The system freezing that you describe sounds (to me) like a smoking gun for the undervolt being something to look at first. With my current CPUs, I can't even finish an EFI driver load (with more than -20mv) without freezing my system and hard-rebooting.

 

[n0taz]

The system freezing that you describe sounds (to me) like a smoking gun for the undervolt being something to look at first. With my current CPUs, I can't even finish an EFI driver load (with more than -20mv) without freezing my system and hard-rebooting.

This is was even 6 mounths early, when we didn't know anything about undervolting. Almost half a year i used v1.efi. Without microcode system freezes.

Other advices i will check after a week, when i will back to home

 

[knopflerbruce]

I flashed a BIOS provided by our chinese(?) friend, and judging from the behavior of the system it's running around 2.9-3GHz on all threads. According to HWInfo, the power limit is exceeded, which might be the reason it does this, I guess. Temperatures are fine. There seem to be four settings related to CPU current in this BIOS, Long Duration Power Limit, Long Duration Maintained, Short Duration Power Limit and Primary Plane Current Limit. I maxed out all. C-states are also disabled.

This is an ASRock X99 WS-E/10G, in case anyone wants to know...

Not sure if I'm asking too much, but the non-USB 3.1 version of the ASUS X99-E WS has no BIOS uploaded to baidu: https://www.asus.com/Commercial-Servers-Workstations/X99E_WS/HelpDesk_Download/ - it just so happens that I have one of those boards, too... with another 18 core in it. Then there's this one: https://www.asus.com/Commercial-Servers-Workstations/X99-M-WSSE/HelpDesk_Download/ - which I also own Also no shared BIOS, only the non-SE one has been uploaded.

 

[CANONKONG]

I flashed a BIOS provided by our chinese(?) friend, and judging from the behavior of the system it's running around 2.9-3GHz on all threads. According to HWInfo, the power limit is exceeded, which might be the reason it does this, I guess. Temperatures are fine. There seem to be four settings related to CPU current in this BIOS, Long Duration Power Limit, Long Duration Maintained, Short Duration Power Limit and Primary Plane Current Limit. I maxed out all. C-states are also disabled.

This is an ASRock X99 WS-E/10G, in case anyone wants to know...

Not sure if I'm asking too much, but the non-USB 3.1 version of the ASUS X99-E WS has no BIOS uploaded to baidu: https://www.asus.com/Commercial-Servers-Workstations/X99E_WS/HelpDesk_Download/ - it just so happens that I have one of those boards, too... with another 18 core in it. Then there's this one: https://www.asus.com/Commercial-Servers-Workstations/X99-M-WSSE/HelpDesk_Download/ - which I also own Also no shared BIOS, only the non-SE one has been uploaded.

Thank you ,I had reuploaded, add more 9 models for ASUS.

 

[Dingmel]

Thank you ,I had reuploaded, add more 9 models for ASUS.

Thanks CANONKONG. already figured out injecting and checking.
Thank you for your contribution to those less technical.