114 replies to this topic

[kdtt]

I got AppleTyMCEDriver.kext to load when I changed from triple channel mode (all 12 memory slots full) to dual channel mode (8 memory slots full, with slots 5 and 6 empty in each bank).
So I'm down from 24GB memory to 16GB, and if I want more memory I'll have to buy 4GB DIMMs.

Do you know whether Real Macpros use triple channel mode? if Macpros doen't support triple channel mode,then the problem is i guess evga boards doesn't let mac os decide channel mode and insists triple channel mode. is there anything related to that in dsdt or other acpi tables that you can fix

[d00d]

Do you know whether Real Macpros use triple channel mode? if Macpros doen't support triple channel mode,then the problem is i guess evga boards doesn't let mac os decide channel mode and insists triple channel mode. is there anything related to that in dsdt or other acpi tables that you can fix

The Nehalem Apple hardware MacPro[4|5],1 and Xserve3,1 can be configured with memory in dual or triple channel mode.
A dual CPU MacPro drops from triple to dual when going from 6 DIMMs to 8, presumably because triple is either 6, 12 or 18.
My single CPU Gigabyte MacPro4,1 was able to load the kext in triple channel mode, so I'm assuming it's something related to the EVGA BIOS.
This may be able to be fixed via DSDT or other method.

[kdtt]

As you might know already, Probably all evga x58 boards have problems in native power management since bios 44. when i installed lion previews i found that i don't have kernel panic when i boot without power management.kext and then install it. When you restart, you have kernel cache without power management kext.But during boot process you see cpupowermanagement initialization complete and after boot ,you'll see all the kexts are loaded.But you have to change the kernel cache before restart again with the kernel cache that is made without powermanagement.kext. if we could change the order of kextloadin that is client.kext before management.kext ,then maybe we can have native speedstep. i don't have any skill for that,but do you think it's possible?

[d00d]

As you might know already, Probably all evga x58 boards have problems in native power management since bios 44. when i installed lion previews i found that i don't have kernel panic when i boot without power management.kext and then install it. When you restart, you have kernel cache without power management kext.But during boot process you see cpupowermanagement initialization complete and after boot ,you'll see all the kexts are loaded.But you have to change the kernel cache before restart again with the kernel cache that is made without powermanagement.kext. if we could change the order of kextloadin that is client.kext before management.kext ,then maybe we can have native speedstep. i don't have any skill for that,but do you think it's possible?

Yes, I've read threads about the BIOS change with some EVGA MBs.
The SR-2 EVGA MB boots with NPM, but as you say, the client loads after.
AppleIntelCPUPowerManagementClient's IOResourceMatch is IOBSD, so it waits for IOBSD to become available.
AppleIntelCPUPowerManagement waits for IOKit, so which one loads first is dependent on it's respective IOResourceMatch.

EVGA SR-2 MB;
Apr 4 07:34:29 localhost kernel[0]: AppleIntelCPUPowerManagement: Turbo Ratios 222233
Apr 4 07:34:29 localhost kernel[0]: AppleIntelCPUPowerManagement: initialization complete
Apr 4 07:34:29 localhost kernel[0]: AppleIntelCPUPowerManagementClient: ready

Gigabyte EX58 MB (log from 2009);
Sep 21 12:36:34 localhost kernel[0]: AppleIntelCPUPowerManagementClient: ready
Sep 21 12:36:38 localhost kernel[0]: AppleIntelCPUPowerManagement: Turbo Ratios 1112
Sep 21 12:36:38 localhost kernel[0]: AppleIntelCPUPowerManagement: initialization complete

[kdtt]

Yes, I've read threads about the BIOS change with some EVGA MBs.
The SR-2 EVGA MB boots with NPM, but as you say, the client loads after.
AppleIntelCPUPowerManagementClient's IOResourceMatch is IOBSD, so it waits for IOBSD to become available.
AppleIntelCPUPowerManagement waits for IOKit, so which one loads first is dependent on it's respective IOResourceMatch.

EVGA SR-2 MB;
Apr 4 07:34:29 localhost kernel[0]: AppleIntelCPUPowerManagement: Turbo Ratios 222233
Apr 4 07:34:29 localhost kernel[0]: AppleIntelCPUPowerManagement: initialization complete
Apr 4 07:34:29 localhost kernel[0]: AppleIntelCPUPowerManagementClient: ready

Gigabyte EX58 MB (log from 2009);
Sep 21 12:36:34 localhost kernel[0]: AppleIntelCPUPowerManagementClient: ready
Sep 21 12:36:38 localhost kernel[0]: AppleIntelCPUPowerManagement: Turbo Ratios 1112
Sep 21 12:36:38 localhost kernel[0]: AppleIntelCPUPowerManagement: initialization complete

Thank you for your replies even though it's not related to SR-2 mobo. I read about nullcpu at this link,http://tgwbd.org/darwin/boot.html .I don't know about the tech stuff at all, but what if i put ioprobescore=100 in infoplist of client.kext.it's different .one is iobsd and the other is iokit.

[d00d]

Thank you for your replies even though it's not related to SR-2 mobo. I read about nullcpu at this link,http://tgwbd.org/darwin/boot.html .I don't know about the tech stuff at all, but what if i put ioprobescore=100 in infoplist of client.kext.it's different .one is iobsd and the other is iokit.

When I wrote NPM I meant Native Power Management, not NullCPUPowerManagement.
IOProbeScore is for two kexts with the same IOMatchCategory name, the one with the highest number loads instead of the one that's lower.
Are you saying that adding an IOProbeScore to AppleIntelCPUPowerManagementClient makes it load sooner?

[kdtt]

When I wrote NPM I meant Native Power Management, not NullCPUPowerManagement.
IOProbeScore is for two kexts with the same IOMatchCategory name, the one with the highest number loads instead of the one that's lower.
Are you saying that adding an IOProbeScore to AppleIntelCPUPowerManagementClient makes it load sooner?

I understood correctly what npm meant.but maybe i wrote something wrong.what i tried to say was, after reading the web page about null and ioprobescore, i thought maybe it'll work if i put ioprobescore in infoplist of client.kext .but i wasn't sure because management.kext and client.kext is two different kexts. so i was asking about that .

[kdtt]

I tried Appleintelcpupowermanagementclient.kext with modified info.plist a few times,but it seems not working at all.
new boot loaders for lion support kext folder,they say. if that function works properly,maybe i could boot without panic.

[maleorderbride]

Do you have sleep problems in dual channel mode?
Mine wouldn't wake from sleep, so I went back to triple channel mode and the AppleTyMCEDriver disabler.
Another problem is that the Parallels (5 and 6) and VMware Fusion (2 and 3 in both 32 and 64bit boot) applications use 100% CPU and slow way down after the EVGA wakes from sleep.
This happens after an image is started or resumed, and wasn't an issue with the Gigabyte.

Thanks for this, I have my system plugged into a UPS that reads out wattage, so I'll be testing if this will bring down the current ~255 W.

Well, I actually have sleep problems already. I disabled sleep and set the model to 3,1 to avoid having to update my IntelGraphicsPowerManagement and remove TyMCEDriver. Now that I can put my RAM in dual-channel I just set it back to 5,1 and still leave sleep disabled.

I might build a second one for a friend and see if perhaps it was just my overclock or some hardware that doesn't cause other problems. The computer is completely stable though and has been for months.

I dont use VMWare or Paralells--I just dual boot.

[Tutor1]

d00d,

As always less is more, making one of the world's slowest 12-core machines one of the fastest. Here are my some of my underclocking results for my 24-7 settings of:

1. No changes in Signal Tweaks yet.

2. No changes in Memory Configuration yet.

3. Frequency/Voltage Control

Both CPU's - Without VDroop

CPU 0 Vcore boot - 1.2750
CPU 0 Vcore eventual - 1.3000
CPU 1 Vcore boot - 1.2675
CPU 1 Vcore eventual - 1.2875

CPU VTT both boot - 1.325
CPU VTT both eventual - 1.350

CPU 0 DIMM Vcore - 1.65
CPU 1 DIMM Vcore - 1.66
IOH 1.35

Everything else at default or auto for now.


4. Frequency/Voltage Control
No Dummy O.C.
Target CPU - 2.17
CPU Freq. - 167
PCIE Freq. - 103
CPU Multi - 13
QPI - 4.8
Memory Freq. - 1066

Everything else at default or auto for now.


5. CPU Configuration

Everything is Enabled and/or set to Auto, including the newly exposed C1E, except MaxCPUID which is disabled.

All else is as you have advised, except that I have moved the MB and all other internals to a easily modified Antec Twelve Hundred case. There are now only 17 internal fans in this puppy, including all the Antec 1200 standard fans and others on various PCIE cards, the CPU cooling systems and the ones I added over the PWM for CPU 0 and the side panel. It's blazing hot in the southern US this summer, so my System temps are in the mid to low 40's C. and cores are in the mid 30's when at idle and range from 45 to 68 when running benchmarks and rendering but quickly drop into the mid to high 30's throughout. But when booting with -v I see that my turbo boosting pay day is DDDDEE. In other words when there is no turbo boost its a 12 core 2.171 GHz cowardly lion; however when I need it to roar, it musters up its core to {Correction - for each CPU} (1) 4.344.4 GHz, 4.344.4 GHz, 4.344.4 GHz, and 4.344.4 GHz, or (2) 4.512 GHz and 4,512 GHz, depending on the environment and the prey. {Addition - So essentially some cores more than double their speed, yielding benching scores about 40% faster than the top-of-the-line mid-2010 machine, for comparably about 80% of the cost and at reasonable voltages.} Below are pics of underclockings' 24-7 yield on the second day of our hot summer.



 24_7_Settings_Sys.png   19.06KB   17 downloads
 24_7_Settings_GB2.png   78.3KB   30 downloads
 24_7_Settings_CB11.5.png   182.75KB   19 downloads

[d00d]

Thanks Tutor1.
I see you updated to 10.6.8 to get the following line in System Profiler;
Processor Name: 6-Core Intel Xeon
Do you also see only four places instead of six in kernel.log?
Here's what I get;
Jun 23 20:25:56 localhost kernel[0]: AppleIntelCPUPowerManagement: Turbo Ratios 0000

[d00d]

Adding the following to com.apple.Boot.plist apparently brings back CStates, as I again see lower CPU temperatures in Temperature Monitor, and lower wattage in my UPS display;
<key>GenerateCStates</key>
<string>yes</string>

CSTInfo has been renamed to AICPMVers in ioreg at IOService:/AppleACPIPlatformExpert/CPU0@0/AppleACPICPU/ACPI_SMC_PlatformPlugin as of 10.6.8.

Looks like we had a warning about the missing CStates last month;
http://www.insanelym...p;#entry1685041

As for the missing turbo multipliers seen in kernel.log, looking in ioreg at IOService:/AppleACPIPlatformExpert/CPU0@0/AppleACPICPU/ACPI_SMC_PlatformPlugin I still see 10 values in the PState's PerformanceStateArray, so the message may be only cosmetic and not functional.

10.7.0 Update:
GenerateCStates doesn't need to be set anymore to have CStates.

[Tutor1]

Thanks Tutor1.
I see you updated to 10.6.8 to get the following line in System Profiler;
Processor Name: 6-Core Intel Xeon
Do you also see only four places instead of six in kernel.log?
Here's what I get;
Jun 23 20:25:56 localhost kernel[0]: AppleIntelCPUPowerManagement: Turbo Ratios 0000

d00d,

I'm running 10.6.7.

I get six places in kernel.log: Apr 16 00:33:48 localhost kernel[0]: AppleIntelCPUPowerManagement: Turbo Ratios DDDDEE

[Tutor1]

What follows is the only way I know that one can achieve the benefits that I've expounded on from underclocking: I began each of my Gigabyte systems by installing Digital Dreamer's distros, following his every instruction for the version of Snow leopard that I had at that time. See, for example, http://www.insanelymac.com/forum/index. ... pic=185097. Then, I learned about d00d and what he had achieved in improving his systems' performance. So I then made all, not some, of the non-DSDT changes that his guide suggests and downloaded his DSDT.zip that was appropriate for my CPU. After unzipping it, I took the DSDT.dsl file and placed it within the Debug folder within the DSDT_Patcher folder within the ~extra folder within the HackInstaller folder within the Run2MakeHackintoshMacHD folder from Digital Dreamer. Then I ran DD's RUN_HACKINSTALLER.command terminal script, and when prompted entered my system password, selected the drive for installation and ran (by entering the no. 7) Run DSDT Patcher. I then exited the script by entering the no. 20 and returned to the Debug folder and copied my newly minted DSDT.aml file and dropped it in my Extras folder then rebooted. The rest is as I've stated earlier, which is partly based on what I read on posts by Sam Oslo (who had achieved a Geekbench 2 CPU score of 27+ on the EVGA SR-2 mobo with some unusually behaving chips) and Movieman (both on yet another forum - Xstreme systems) about BCLK sweet spots). I followed this same path for my EVGA SR 2, except that I followed all, not some, of d00d's guide for that mobo, even changing my version of Chameleon to the exact version d00d recommends. I used the same process to convert my DSDT.dsl file to my working DSDT.aml file. Any deviation from this path may not yield all/any of the benefits that I have described.

[Tutor1]

Thanks Tutor1.
I see you updated to 10.6.8 to get the following line in System Profiler;
Processor Name: 6-Core Intel Xeon
Do you also see only four places instead of six in kernel.log?
Here's what I get;
Jun 23 20:25:56 localhost kernel[0]: AppleIntelCPUPowerManagement: Turbo Ratios 0000

I tried 10.6.8 and BEYOND, but both screwed up my underclocking advantages; so I reverted to 10.6.7 on another HD. For me, "AppleIntelCPUPowerManagement: Turbo Ratios 0000" had negative results. This is the edge I got back (1) by downgrading to 10.6.7 on another hard drive so that underclocking ability is fully restored and (2) by increasing the eventual Vcore (to 3.50 for CPU0 and to 3.25 for CPU1) and the BCLK to 182, yielding a maximum turbo boost of 4.914 GHz.

 Cinebench11.5_24.00n45.88.png   144.05KB   23 downloads
 GBench2_39535.png   71.9KB   23 downloads

[maleorderbride]

Anyone using Lion? I was thinking of trying it out this week on my SR-2. Maybe get sleep working perfectly for once.

[d00d]

Anyone using Lion? I was thinking of trying it out this week on my SR-2. Maybe get sleep working perfectly for once.

Yes.
Wake from sleep works with triple channel memory (12 slots), but not double (8 slots), just like SL.

[maleorderbride]

Yes.
Wake from sleep works with triple channel memory (12 slots), but not double (8 slots), just like SL.

Can you upload your Extensions folder for Lion? Thanks!

[d00d]

Can you upload your Extensions folder for Lion? Thanks!

Post 1:Configuration shows what I'm using.

[maleorderbride]

Post 1:Configuration shows what I'm using.

Are you having any problems with the VoodooHDA that you linked in the first post with Lion? I am having intermittent KPs on bootup that normally cite IOPCIFamily, Audio Kexts, or VoodooHDA. I tried deleting AppleHDA, but it wasn't loading anyway. No dice.

I dialed the overclock back down to stock just in case, but no help there either.


edit:
The most recent RC5 Chameleon (1183 or something) causes a KP when combined with the current version of VoodooHDA. Rolled back to 1086 and I am fine.

eidt2:
A newer version of Voodoo works much much better in Lion. I don't get the same KPs when switching bootloader revisions.

http://www.projectos...p?showtopic=355

Oh, and the newer version also allows me to see HDMI outputs for my GTX 570. I don't have an HDMI monitor, so no idea if they are actually working.