[GUIDE] Full Power Management w/ Dual/Single Broadwell-EP/Haswell-EP CPUs on 10.12.

[GDS]

in my opinion (i need to optimize from my rig)

I will see the problem like that:
1/ stability and reliability (better to have a score lower but fully OK than unstable)
2/ i will choose the way that the best score is for the multi one (not single)
98% of the programs use multi (for 2 or 4 cores and much more) but it is very very rare to have a program stressing for one unique single core!
and much more in my case with 22 cores

so i will say (for me) that i will choose the one which is reliable and better on multi cores,

 

[ronalith]

in my opinion (i need to optimize from my rig)

I will see the problem like that:
1/ stability and reliability (better to have a score lower but fully OK than unstable)
2/ i will choose the way that the best score is for the multi one (not single)
98% of the programs use multi (for 2 or 4 cores and much more) but it is very very rare to have a program stressing for one unique single core!
and much more in my case with 22 cores

so i will say (for me) that i will choose the one which is reliable and better on multi cores,

Best performance multi-core (on benchmarks) so far for me is disabling xcpm.

 

[Jonny1980]

I run a 6850 and was about to start doing manual kernel patching per pikers example

all is can say is "high 5" - life saver - works PERFECT - for macpro6,1 that was running prior or any other specific situation - remove fakecpuid - it's not needed with the last listed kernel patch above

 

[insomniack]

This tutorial helped me install and run sierra. The only problem I have is the artifacts. No performance issues but i have artifacts after i install the nvidia web drivers. Can the issue be about bios settings ? (I have supermicro X10DAL and GTX 550 ti graphic adapter)

 

[liufei]

So, why is it that enabling xcpm results in lower benchmarks/performance? I'm getting 15k vs 20k geekbench 4 on an i7-6800k @ 4.2 GHz

mine is ok

 

[maleorderbride]

So, why is it that enabling xcpm results in lower benchmarks/performance? I'm getting 15k vs 20k geekbench 4 on an i7-6800k @ 4.2 GHz

I have lower multi-core performance with this method as well. It seems to be ~100-200MHz lower when all cores are in use, plus it fluctuates quite a bit.

 

[Ur-Quan]

Ok, this is my first shot at a guide, so it might suck. Hopefully it doesn't. But it might.

Hi, Metacollin.
I've got the same mobo (X10DAi) and did everything you're described. After installng 10.12 and generating SSDT for dual Xeon E5-2630v4 i got all 40 threads with sufficient performance in GeekBench and speedstep working. But after updating to 10.12.1 or 10.12.2 system crashes with message "SMC returned incorrect key". I've tried different versions and forks of FakeSMC but no result.
Can You share DSDT, clover config and may be BIOS settings? Thank You.
Sorry for bad English.

 

[mysticmerlin]

This tutorial helped me install and run sierra. The only problem I have is the artifacts. No performance issues but i have artifacts after i install the nvidia web drivers. Can the issue be about bios settings ? (I have supermicro X10DAL and GTX 550 ti graphic adapter)

I solved that by disabling hyperthreading in bios.

 

[GDS]

Hello all

i come back with some results and some screens
I am not so disappointed and a huge THX to the author of this thread, again!

i have to solve some stuff as nvidia webdriver is not ok but CUDA yes so .... no pb
and the bluetooth and some others points to optimize

but fully OK by following step by step the method
the pic of the rig is not "clean" (we can see some fixations)
i need to clean all the mess to have something more attractive


PS: funny to see that in bench, i am lower in t°C than in idle... 32°C in full vs 36°C in idle
also all the cores are at 2,8GHz, 24/24

 

[GDS]

I come back for one point:

To have the Nvidia web driver ACTIVE, i need to stop the hyper-threading in the bios

and strangely i have much better score in mono thread and also i loose tiny bit (10%) in multi (by going from 44 thread to 22)
but in openGL: i was before at 62 and now 88,45 meaning 40% more,
a huge step forward for renderings

next step: flashing the 2 Titan X bios



PS: for the fun i put the t°C during benchmark, in my home the t°C is 23°C
the CPU is at only 5°C more (case closed and fans at 900 RPM)
meaning easy for it