Jump to content

Haswell-E 5820k 5930k installation guide with full speedstep working


358 posts in this topic

Recommended Posts

Hi all,

Us over in the 5960X thread are trying to get that CPU to boot in El Capitan.  To date, it seems that no one has successfully used this CPU.  To figure out what is going on we want to look at ACPI tables for the 5960X and the 5930K on the same motherboard.

 

Would anyone in this thread who has a 5930K on the Asus X99-E WS be kindly willing to upload their ACPI tables?

 

Thank you!

Stinga--I noticed this error in console and have been trying to address it, but it still remains.

X86PlatformShim::start - Failed to send stepper

Pike states that it is caused by a failure to find the board-id's frequency vectors. I know that the X79 Plugin kext is supposed to be injecting that information, but it does not seem to be working quite right. 

 

I also tried to modified the X86PlatforumPlugin manually to ensure that the MacPro Board-ID plist was in there (I stole the one from the X79 plugin), but it still generates this same error. 

 

I then tried a iMac14,2 SMBIOS, but no difference in power management.

 

 

 

 

I assume this is part of why we have incomplete power management with Hasewll-EP. For example, turbo states are not reached reliably, nor does a 100% load even result in non-turbo states being flat-lined in IPG. 

 

Running Cinebench gives dramatically different scores each time, with mine varying from as low as the 800s, to as high as 1710.

 

 

Are other people also having this problem? If not, please post your SSDT and config.plist.

I guess the question is, does anyone see?

X86PlatformShim::sendPStates - Success!
X86PlatformShim::sendStepper - Done!

If so, then they really have XCPM loaded. I only have it halfway installed. Not using AICPUPM, but not really XCPM.


@maleorderbride: same error on my side. Tried generating SSDT with several different flags (see https://github.com/Piker-Alpha/ssdtPRGen.sh/issues/65 ) but I always get the same message

Yea, I read that as well when googling last night. However Pike's latest ssdtPRGen doesn't even allow me to make new SSDTs as it no longer allows me to supply all values for CPUs that do not exist on his tables.

 

Too lazy to file a bug report as of yet...

Hey guys, after Vegas craziness I wanted to do some testing and see what I can come up with.

 

What's been tested:

 

Hardware: 5930k on X99-E WS

 

Different variations of SSDT, Clover and versions of Mac.

 

MacPro6,1 gives the best overall graphic performance on the GPU, but the CPU performance is not that great, plus you can actually see some noticeable lag in the UI.

iMac17,1 (all models) gives the best CPU performance, but there is a GPU lag and USB's acting up a bit strange (telling me that there is not enough juice to charge a mouse :) ).

 

So what I end up doing is removing SSDT,  and letting Clover do the P/C state injections (v.3338, oddly enough any version prior to that never managed to work on injecting the states.), replaced the the MacPro6,1 ( Mac-F60DEB81FF30ACF6 ) frequency vector with the one from iMac17,1 (i7 version, Mac-65CE76090165799A ), started clover with -xcmp and so far got the best compromise of speed and power. No FakeCPUID is being used.

 

One of the ways I've tested the UI lag is to preview Arabesque screensaver, almost on all of the combinations I've tried it was a 3-4fps slideshow except the last one.

 

My two cents; I know that this is still not a 100% flatline, but it gets close, and considering that it hovers in the overclocked turbo region is good enough for me and I think for most people unless you're doing some heavy CPU lifting e.g Scientific, video, audio I don't think you'd notice (doing compiles on xCode, android and some heavy VMware tasks myself I didn't see the any difference. )

 

Now since, Geekbench and Cinebench might not be the best to test, I end up testing it with Handbrake in real life situation e.g. converting Full HD mov into Full HD h264 MP4 file, and well... let me pictures speak for themselves.

 

Geekbench:

post-6786-0-41644800-1453777365_thumb.jpg

 

 

Cinebench:

post-6786-0-90645500-1453777352_thumb.jpg

 

Handbrake:

post-6786-0-22474000-1453777440_thumb.jpg

 

And looking to at sysctl machdep.xcpm and looking at the system log I think XCMP is loaded correctly.

post-6786-0-42187600-1453777455_thumb.jpg

 

Hope this helps someone.....

On the side note:
USBs are working perfectly without and inject or mods, the only thing needed is up-ing the limit of ports to get 2 more ASM's loaded.

post-6786-0-40246200-1453777464_thumb.jpg

 

 

 

 

 

Very nice work Balamut! I'll give that a shot somewhere in the nearish future.

 

For now, I have just disabled speestep, which allows me to stay continuously in turbo states and gives me absolute flat lines during renders in IGP. This gives me extremely solid and consistence performance, but obviously everything runs a bit warmer and I am always consuming ~50W+ witht he CPU. ;p

 

Your solution looks like a much better year round application. Thanks for posting that!

Hey guys, after Vegas craziness I wanted to do some testing and see what I can come up with.

 

What's been tested:

 

Hardware: 5930k on X99-E WS

 

Different variations of SSDT, Clover and versions of Mac.

 

MacPro6,1 gives the best overall graphic performance on the GPU, but the CPU performance is not that great, plus you can actually see some noticeable lag in the UI.

iMac17,1 (all models) gives the best CPU performance, but there is a GPU lag and USB's acting up a bit strange (telling me that there is not enough juice to charge a mouse :) ).

 

So what I end up doing is removing SSDT,  and letting Clover do the P/C state injections (v.3338, oddly enough any version prior to that never managed to work on injecting the states.), replaced the the MacPro6,1 ( Mac-F60DEB81FF30ACF6 ) frequency vector with the one from iMac17,1 (i7 version, Mac-65CE76090165799A ), started clover with -xcmp and so far got the best compromise of speed and power. No FakeCPUID is being used.

 

One of the ways I've tested the UI lag is to preview Arabesque screensaver, almost on all of the combinations I've tried it was a 3-4fps slideshow except the last one.

 

My two cents; I know that this is still not a 100% flatline, but it gets close, and considering that it hovers in the overclocked turbo region is good enough for me and I think for most people unless you're doing some heavy CPU lifting e.g Scientific, video, audio I don't think you'd notice (doing compiles on xCode, android and some heavy VMware tasks myself I didn't see the any difference. )

 

Now since, Geekbench and Cinebench might not be the best to test, I end up testing it with Handbrake in real life situation e.g. converting Full HD mov into Full HD h264 MP4 file, and well... let me pictures speak for themselves.

 

Geekbench:

attachicon.gifGeekbnechClover.jpg

 

 

Cinebench:

attachicon.gifCinebenchClover.jpg

 

Handbrake:

attachicon.gifHandBrakeClover.jpg

 

And looking to at sysctl machdep.xcpm and looking at the system log I think XCMP is loaded correctly.

attachicon.gifxcpmClover.jpg

 

Hope this helps someone.....

 

On the side note:

USBs are working perfectly without and inject or mods, the only thing needed is up-ing the limit of ports to get 2 more ASM's loaded.

attachicon.gifUSBX99.jpg

Its look nice but them 62.0 C maybe High.

  • 2 weeks later...

 

To get all USB3.0 working you need to use this binary patch and download my X99 injector. Put the injector into the /Library/Extensions theattachicon.gifX99 USB3.0.zip

Thank à lot

I stayed in Mac Pro 5.1

The SpeedsTep function flawlessly

I adjusted your kext for my X99-A

Usb 3 Operate

config.plist.zip

X99_Injector.kext.zip

post-1181448-0-83118500-1454589410_thumb.png

post-1181448-0-07101200-1454589431_thumb.png

post-1181448-0-61199000-1454589443_thumb.png

post-1181448-0-38626000-1455555281_thumb.png

DarwinDumper_2.9.9.2_AMI_X64_3346_Cap_Fljagd.zip

Hi all, hi Stinga.

sorry maybe for the double post :)

 

i have a 5820k on X99-A and my bench is low. 

4447 - 19937  @ 4.4ghz   :surprised:   it's like the OC doesn't works, however i can see the cpu go up and down between x12-x44. 

i used your FakeCPUID settings if not the cpu is stuck @3.3 (no speedstep).

 

As you can see the single core is not really bad but the multicore is low.

i read some 5820k users have around 25000 bench score. 

 

what do you think?

 

thanks for you help

 

Info: i'm on Yosemite 10.10.5

@samtimbaud: using the configuration suggested by Balamut above (i.e. no fakecpuid, frequency vector from iMac17,1) I get better scores (23k @46x overclock, xcmp). You can get much better scores using other fakeCPUIDs (+ NULLCPU), but the CPU is most of the times stuck at the highest multiplier, resulting in higher temperatures (I reached 29k @47x overclock).

Thanks Macandrea for your Answer.

i'im thinking to change the motherboard by one i already know, asus deluxe 99. 

i read on a forum that the X99A was very bad for OC and hard to past 20k with overclocking. 

look, you have to up the cpu until 4.7hz for only 23000. it's crazy!

I let you know this afternoon..

 

Best,

Sam

How do you change frequency vectors?  Thru SMBios in config.plist?


@samtimbaud: using the configuration suggested by Balamut above (i.e. no fakecpuid, frequency vector from iMac17,1) I get better scores (23k @46x overclock, xcmp). You can get much better scores using other fakeCPUIDs (+ NULLCPU), but the CPU is most of the times stuck at the highest multiplier, resulting in higher temperatures (I reached 29k @47x overclock).

If I overclock using a 100 BCLK, then use NULLCPU El Cap still only recognizes the 3.5Ghz clock.  Speedstep aside, how do I get El Cap to run at the highest multiplier?  In other words (and I'm not asking for a guide, just basic settings) how to I achieve the last result (29k @4.7Ghz) you mentioned?  I've only just started to experiment with El Cap and things like FakeCPUID are all new to me.  I really am not concerned with power draw or a bump in temp.

@onmybikedrunk

 

here the results I get with two different configuration files, using a 46x max multiplier with BCLK set to 100:

 

config-1.plist: no FakeCPUID, using xcmp

post-238487-0-95396600-1454966201_thumb.png

 

config-2.plist: FakeCPUID set to 0x0306e2. NULLCPU

post-238487-0-65955700-1454966209_thumb.png

 

I had some stability issues @47x, so I got lower scores than expected (had to boot @46x) - config.plist files attached.

Please tell me if you also need to know my BIOS settings. I am using latest 10.11.4 Beta (15E39c).

 

 

Config_files.zip

@onmybikedrunk

 

here the results I get with two different configuration files, using a 46x max multiplier with BCLK set to 100:

 

config-1.plist: no FakeCPUID, using xcmp

attachicon.gifScreen Shot 2016-02-08 at 22.15.37.png

 

config-2.plist: FakeCPUID set to 0x0306e2. NULLCPU

attachicon.gifScreen Shot 2016-02-08 at 22.15.53.png

 

I had some stability issues @47x, so I got lower scores than expected (had to boot @46x) - config.plist files attached.

Please tell me if you also need to know my BIOS settings. I am using latest 10.11.4 Beta (15E39c).

 

Awesome!  This is exactly what I was looking for.  Your help is appreciated.  As far as BIOS settings for NullCPU I assume you have EIST disabled?

 

EDIT: Actually, yes -- your BIOS settings would be extremely helpful as well.

Edited by onmybikedrunk

@onmybikedrunk

 

Cpu core ratio: sync all cores

Asus multicore enhancement: disabled

Ai overclock tuner: auto

Internal pll overvoltage: enabled

Epu power saving mode: disabled

EIST: disabled

Turbo: enabled

Long duration package power limit: 140

Ratio limit: 46

Cpu c-states: enabled

EnhAnced c1 state: enable

Cpu c3 report:disabled

Cpu c6 report:disabled

 

Everything else is set to auto or to the standard Factory value

 

I've used the same settings for both configs. Hope it helps.

 

Andrea

@onmybikedrunk

 

Cpu core ratio: sync all cores

Asus multicore enhancement: disabled

Ai overclock tuner: auto

Internal pll overvoltage: enabled

Epu power saving mode: disabled

EIST: disabled

Turbo: enabled

Long duration package power limit: 140

Ratio limit: 46

Cpu c-states: enabled

EnhAnced c1 state: enable

Cpu c3 report:disabled

Cpu c6 report:disabled

 

Everything else is set to auto or to the standard Factory value

 

I've used the same settings for both configs. Hope it helps.

 

Andrea

 

Awesome.  Thanks a ton.  I haven't really paid much attention to El Capitan because previously I had a 5960X which I swapped out for a 5930K to have an updatable system.  I'm sure this will help others too, not just me.  So thank you!

×
×
  • Create New...