Jump to content

El Capitan XCPM for X99 - One patch solution


SammlerG
 Share

9 posts in this topic

Recommended Posts

Hi,
 
for El Capitan u can use a one patch solution to get xcpm working correct.
 
Since El Capitan release all users with a x99 system can´t reach max. CPU frequency. You remember? 
So we all used a FakeCpuID + NullCpuPM kext to get pm to work.
 
The reason for this bad performance since 10.11 is the handling and programming of the MSRs with xcpm, which was enabled by default in El Capitan with OEM CpuID.
This MSR is not compatible with the most X99 boards, it enables Intel Speed Step by default!
 
 
With Clover and kernel&kext patching we can solve this problems.
 
Instead of using FakeCpuID and NullCpuPM kext, we can change the MSR 0x01A0 to correct settings without enabling EIST. 
 
Things to do:
 
1. with El Capitan xcpm is always enabled when you use the correct 5920k/5930k/5960x oem CpuID (0306F2)
2. don´t use a FakeCpuID
3. don´t use NullCpuPowerManagement kext
4. go to Kernel and Kext patches / KernelToPatch and enter the following patch

find: A0010000DE17000000000000000000000000000040000000010005000000000000000000000000000000000000000000
replace: A0010000DE17000000000000000000000000000040000000010004000000000000000000000000000000000000000000
Comment: OS 10.11.6 EIST Patch - fix MSR 0x01A0 by Pike R. Alpha

 

That´s all. Tested with 10.11.6 latest updates. 

 

 

Credits goes to Pike R. Alpha and his blog about xcpm for unsupported processors! 

  • Like 6
Link to comment
Share on other sites

OC Setting Frequency:X99 Gaming G1

 

XMP: profile 1

Uncore Ratio:43x

C1E: Disable

C6/C7: Disable

CPU Thermal Monitor: Disable

C3: Enable

EIST: Disable

 

Hi guys  please share your system OC settings

Thank you.

Link to comment
Share on other sites

  • 2 weeks later...

I finally tested this on a 5960X and it worked quite well. I had steady maximum performance and P-States reached, but still would drop to 1.2Ghz and low power usage.

 

BIOS Settings:

EIST: Disabled

CPU Core Ratio should be set to a multiplier, or Sync All Core

C-States: Enabled

C3: Enabled 

 

 

This does not seem to work for Broadwell, so perhaps the title should make clear that this is Haswell-E/P only instead of X99? Or does someone have this working in Broadwell with an additional patch or two?

  • Like 1
Link to comment
Share on other sites

BIOS Settings:

EIST/C1E/C3/C6/C7 - Enable

 

The correct would be 22000~23000 in Geekbench 4, see: https://browser.primatelabs.com/v4/cpu/605038

Not use: FakeCPUID, NullCPU ...

 

Cinebench score: 910, correct would be 1100~1150, 4.2ghz OC  :(

 

Intel Power Gadget dont work, only Sandybrigde CPUs  :(

 

 

 

1. Install IPG using Pacifist.

 

2. Your BIOS settings are incorrect, use the ones that I specify.

 

3. If you still do not get a normal score, then add your IGP screenshot while Cinebench is running and your upload your config.plist & Clover kexts folder.

Link to comment
Share on other sites

1. Install IPG using Pacifist.

 

2. Your BIOS settings are incorrect, use the ones that I specify.

 

3. If you still do not get a normal score, then add your IGP screenshot while Cinebench is running and your upload your config.plist & Clover kexts folder.

 

which BIOS setting to use?

Link to comment
Share on other sites

  • 2 weeks later...

what's the one patch solution for Sierra 10.12.0

 

Using Gigabyte X99P-SLI and Broadwell-E 6800K

 

And you must always use FakeCPUID for Broadwell-E, i haven't seen one solution work without it

 

TIA

 

I "ported" (no actual work done ;p) this patch to Sierra, but it does not have any observable effect. Does anyone more knowledgeable than I see a problem with it in the kernel?

 

However, this is only for Haswell-E/P and the thread is just awaiting a re-name.

 

If anyone is interested, you can test it yourself:

40000000010005000000000000000000000000000000000000000000 to 40000000010004000000000000000000000000000000000000000000

Oddly, the original El Capitan patch works with some of my Haswell Xeons, but not others. I have yet to determine why that is. For example, my extremely old engineering sample E5-2693 V3 2.8Ghz 12-core runs at the highest possible turbo ratios under this patch, but never drops out of turbo.

 

While, my E5-2673 V3 2.4GHz 12-core (a very late stage ES) runs perfectly, just like my retail i7-5960X. It consistently flat-lines during benchmarks, then fluctuates between 1.2-1.5Ghz in a way that is similar to Windows.

 

So perhaps my problem is that I have this janky old ES CPU in one of my builds? Does anyone else have earlier stage ES Haswell-EP Xeons and can test this patch and see how it goes?

  • Like 1
Link to comment
Share on other sites

  • 3 weeks later...
  • 4 weeks later...
 Share

×
×
  • Create New...