All We Know About Maximizing CPU Related Performance

Tutor said:

Even as you encounter new problems, it does not diminsh what you've learned. Clearly, I don't know everything. But does that mean that I cannot help others by imparting what I do know. View each new obstacle as an opportunity to increase your knowledge base so that you can help others with those kinds of problems. And never forget that the more you learn, the better will be your guesses.

Click to expand...

Not too far off from what my father told me when I was a kid: "...the more you know the less you know..." Thanks for for your efforts...

Tutor said:

Have you tried unplugging it from every outside source and removing the battery for 1 min?

Click to expand...

As mentioned before I even went the distance to disconnect everything and let the system just sit for 2+ hours instead of the 5 to 30 min. minimum, (I found that out on another post somewhere). Then I reconnected everything back together, put the battery back on the mobo, and reset the CMOS and the system started right up.

So I went beyond the 1 min. minimum. I did a 2+ hour maximum... But I'm still trying...

Tutor said:

But have you tried unplugging it from every outside source and removing the battery for 1 min with the EVGA 480 in the PCIe slot?

Click to expand...

Yes, I tried that before and I did it one more time just now, and still FF... I'm wondering if it's the mobo or maybe the CPUs. Maybe I should take the X5680's out and put back the X5650's and try it out again...

Tutor said:

If you have another bios stored on your board you might give it a try.

Click to expand...

I tried that too. I'm going to have to basically take everything apart and re-assemble it together piece by piece to figure out what the problem is. This is not what I was expecting; especially after all the work that went into this and all the time you have contributed as well. I'm just puzzled as to what it could be and thought it was something simple?...

I hope it's just a wire or something and nothing else major. Thanks for the follow up...

Hi Tutor, well I put the X5650s back in and it booted and then put back the X5680s and they work fine (for now). I was curious about the temps on the mobo as they are reading around 30C to 31C. Is that normal for Underclocking? Thanks again for everything...

Tutor said:

Those temps are excellent; they're similar to mine. Did you try to underclock those X5650's and measure their Geekbench 2 performance when you had them in most recently, given what you now know? Just make sure to set CPU multi ratio to 13, QPI to ~ 4800, ram to 1333 (Auto) and BCLK to 175 - 180 for starters. All CPU Vcores can be set to Auto or one step below 1.35. Everything else remains the same as for the X5680s.

Click to expand...

Since I really just wanted to find out if was the mobo or the cpus, I just the 50's on there to see if the system would start up, and it did, but I got some kernel panic. Figuring that it at least got that far, I just decided to get the 80's back on there to see if they would work. If the the mobo would have FF'd on me, then I would have gone back to the 50's again to see if everything would work. Had that been the case then I would have known then that the 80's were problematic and I would have returned them to my source who was more than happy to extend my warranty to a full 30 days.

1) At this point I'm going to put them through the riggers by TT'ing and go from there. So far my system is set up like this:

CPU 0 Vcore boot - 1.35
CPU 0 Vcore eventual - 1.375
CPU 1 Vcore boot - 1.35
CPU 1 Vcore eventual - 1.375

CPU VTT both boot - 1.40
CPU VTT both eventual - 1.40

CPU 0 DIMM Vcore - 1.65
CPU 1 DIMM Vcore - 1.66
IOH 1.40

the CPU Freq is set to 165 and 13 for the other one.

With the RAM (for now) I'm keeping it at 1066 as it keeps not booting when I put it on 1333. So I don't know where to go from there but I will input 7-7-7-20 CR 1 when I boot it next time as that is where you and I left off from when you last gave me your input.

2) I did want to find out what TT method do you use (if any)? As I've mentioned before I've been using Terminal and using the command:

yes > /dev/null

But do you use another method or app. I noticed that you gave me this link last week:

http://www.mersenne.org/freesoft/

to test the memory, but would you use Prime95 for the CPU? I've heard conflicting thoughts on that as some programmers/builders don't use that app as it doesn't stress the CPU properly. Let me know your thoughts on this...

3) Lastly, I'm trying to figure out what this means:

- input: PCIRootUID=1 in org.chameleon.boot.plist

as that is one of the things that I'm supposed to input (if it's not in the org.chameleon.boot.plist file). I say this because I'm trying to get my GTX 580 card to get full res options and it's only seeing one res option. So I'm installing the GTX 5XX Enabler 10.7.2 app. Now I'm not expecting you to understand what I'm talking about with the app install stuff, but if I'm supposed to put that line in the org.chameleon.boot.plist file, is it supposed to look like this:

<string>PCIRootUID</string>
<key>1</key>

If it is, then I did it correctly. If not, then please let me know how it should look. Again, thanks for the help... Later...

Tutor said:

1) I Prime it.
2) That's how I would interpret it, but I'm not sure.

Click to expand...

Okay. I figured that I would try Prime95 one more time, but why is it when I use it that it seems to be stressing on one section of the RAM on CPU0 (VERY HOT to the touch) and not the other section of the RAM on CPU1 (just luke warm to the touch)? Does it alternate through it's testing from one to the other? Mind you, I'm using "Torture Test" - "Blend" version.

Also, it just quit on me after about 20 min. into the TT - Blend test. Let me know what you can. Thanks...

Stock Benchmarks...

Turbo mode up and running stock benchmarks at Geekbench 26,600. For what it's worth.
http://browse.geekbench.ca/geekbench2/602392

A cooler Mac Pro

Hi everyone,

I'm running a standard 2007 1,1. I have a 5770 in it with 4 gb of ram (need more but haven't gotten around to it).

I'm not much of a modder anymore. I used to be though but out of the loop.

My landlord put the kibosh on air conditioning unless I pony up 60 extra bux a month. So I'd like to put on cooler fans and heatsinks. Someone suggested to me these socket 771 fans and they look like they would do the job. Do you think I could replace what I got with these? Each link is the same product but different size.

Recommended #1

Recommended 2

I'd like to be able to only pay $120 once than 60 bux every month until september as I can live with fans and not AC.

Hi Tutor:

I've been shut down since April 5th, but I just picked up a pair 5690's (Retail version) and just did a GB score of 34,771 (found here):

http://browse.geekbench.ca/geekbench2/610857

So far so good. Thanks again for your belief in me and now my Mobo temps went from 30C to now 25C. So for that I'm very happy and the GTX 580 card works as well. Can't wait to get that water-cooled and everything fully put back together. Again, thanks...

PS - If the other guys need help, I'll try to guide you as best as I can...

braindeadfool said:

Turbo mode up and running stock benchmarks at Geekbench 26,600. For what it's worth.
http://browse.geekbench.ca/geekbench2/602392

Click to expand...

Good for you man !!! Keep going at it...

Tutor said:

Congratulations. Glad to hear that the new CPUs/installation were successful and have resolved the issues.

Click to expand...

I would not say that totally... It did lock up on my second round of geek bench scoring. So I'm going to have to figure out what the issue is later on and why it's still locking up. I haven't really tweaked the ram nor have I tweaked a whole bunch of other stuff from the last batch of input you gave me for my bios set up. I'll figure out what to do as I start looking into tweaking things here and there little by little. I'll probably have a question or two as time moves along...

Hi Tutor, I just PM'd you. When you have a moment I hope to hear from you... Later...

More on underclocking the EVGA Classified sr-2

A few things I've learned:

In an attempt to make the underclocking conversation a bit more comprehensible, I've taken a fairly systematic result in trying to understand the nuances of this process.

First off, you can only underclock if you have native cpu management working. Although the clients and server apps may load from kexts, you only have turbo boost enabled if your kernel.log has something like this in it:

Apr 20 16:41:53 localhost kernel[0]: AppleIntelCPUPowerManagement: Turbo Ratios EEEEFF

It will have only 4 numbers/letters if a 4 core processor.


The current versions of Chimera (1.9.1 ) and Chameleon v2.1 r1923 (ErmaC) work just fine to produce underclocking results. You do not have to use Chameleon 2 RC5 variants. This is nice for those of us with newer Nvidia cards.

There are currently two major development builds for FakeSMC (https://github.com/kozlek/HWSensors and http://www.assembla.com/code/fakesmc/subversion/nodes)
, as of April 19th 2012, either appear to work OK, but each has their individual problems when trying to evaluate monitoring. I'm using the current version from Slice (at asssembla.com) v4.2r604.Use hardware monitoring with caution. Some versions of the associated FAKESMC cpu monitors DO NOT WORK to underclock as they disable break turbo ratios. I install FakeSMC without any monitoring kexts first. If you want to play it safe up front, just use FakeSMC without any of the monitoring texts. I have not tested the current version of FakeSMC that comes with Multibeast 4.3

You do need to do all of the edits to DSDT and FakeSMC outlined by d00d in http://www.insanelymac.com/forum/index.php?showtopic=233891

Lion 10.7.3 works OK to produce underclocking results

system kext cache is dependent on the presence or absence of C-states when built; use KernelCache=No flag in org.chameleon.Boot.plist while mucking about. You can rebuild the system kext cache when you have settled on turbo ratios and whether or not to use C and P states.

And although you would be lead to believe that all energy states are being implemented by the proponents of underclocking, this is not so. You can not underclock this system and have C and P state generation enabled (actually this may only be true for P states, havent' tested thoroughly).

To get underclocking to successfully work (as of this writing) using current chameleon and FakeSMC versions, you must have the following two flags set in org.chameleon.Boot.plist: GeneratePStates=No and GenerateCStates=No

So what does this all mean. Here are some early results (no efforts to optimize anything). Turbo mode enabled for all, BIOS settings as per d00d and Tutor:
Geek bench scores average of 3
Stock is multiplier (26) times BLCK (133). Turbo is 111122
Power consumption includes my monitors (about 80 watts total)

Stock Intel Xeon x5690 x 2 at 3.46GHz, +C + P states: Geekbench 27,076 325Watts Idle, 600Watts peak

Stock -C -P states: Geekbench 27,170
360 Watts idle, 600 watts peak

Underclocked is multiplier (13) times BLCK 160 (2.08 GHz). Turbo is EEEEFF (4.48 highest turbo cores).

Underclocked +C +P states: Geekbench 17,800, 295 watts, peak 450 watts
Underclocked -C -P states: Geekbench 33,852 385 watts idle, peak 700 watts

Overclocked is 160x 26 multiplier, (4.48 Ghz highest turbo, 4.16 Base)
Overclocked +C +P states: Geekbench 32,150 350 watts idle, peak 685 Watts

So clearly underclocking offers an advantage in performance versus traditional overclocking. Depending on processor utilization, it is not so clear that it saves energy, since the C state generation is allowing less energy use.

braindeadfool said:

A few things I've learned:

In an attempt to make the underclocking conversation a bit more comprehensible, I've taken a fairly systematic result in trying to understand the nuances of this process.

First off, you can only underclock if you have native cpu management working. Although the clients and server apps may load from kexts, you only have turbo boost enabled if your kernel.log has something like this in it:

Apr 20 16:41:53 localhost kernel[0]: AppleIntelCPUPowerManagement: Turbo Ratios EEEEFF

It will have only 4 numbers/letters if a 4 core processor.


The current versions of Chimera (1.9.1 ) and Chameleon v2.1 r1923 (ErmaC) work just fine to produce underclocking results. You do not have to use Chameleon 2 RC5 variants. This is nice for those of us with newer Nvidia cards.

There are currently two major development builds for FakeSMC (https://github.com/kozlek/HWSensors and http://www.assembla.com/code/fakesmc/subversion/nodes)
, as of April 19th 2012, either appear to work OK, but each has their individual problems when trying to evaluate monitoring. I'm using the current version from Slice (at asssembla.com) v4.2r604.Use hardware monitoring with caution. Some versions of the associated FAKESMC cpu monitors DO NOT WORK to underclock as they disable break turbo ratios. I install FakeSMC without any monitoring kexts first. If you want to play it safe up front, just use FakeSMC without any of the monitoring texts. I have not tested the current version of FakeSMC that comes with Multibeast 4.3

You do need to do all of the edits to DSDT and FakeSMC outlined by d00d in http://www.insanelymac.com/forum/index.php?showtopic=233891

Lion 10.7.3 works OK to produce underclocking results

system kext cache is dependent on the presence or absence of C-states when built; use KernelCache=No flag in org.chameleon.Boot.plist while mucking about. You can rebuild the system kext cache when you have settled on turbo ratios and whether or not to use C and P states.

And although you would be lead to believe that all energy states are being implemented by the proponents of underclocking, this is not so. You can not underclock this system and have C and P state generation enabled (actually this may only be true for P states, havent' tested thoroughly).

To get underclocking to successfully work (as of this writing) using current chameleon and FakeSMC versions, you must have the following two flags set in org.chameleon.Boot.plist: GeneratePStates=No and GenerateCStates=No

So what does this all mean. Here are some early results (no efforts to optimize anything). Turbo mode enabled for all, BIOS settings as per d00d and Tutor:
Geek bench scores average of 3
Stock is multiplier (26) times BLCK (133). Turbo is 111122
Power consumption includes my monitors (about 80 watts total)

Stock Intel Xeon x5690 x 2 at 3.46GHz, +C + P states: Geekbench 27,076 325Watts Idle, 600Watts peak

Stock -C -P states: Geekbench 27,170
360 Watts idle, 600 watts peak

Underclocked is multiplier (13) times BLCK 160 (2.08 GHz). Turbo is EEEEFF (4.48 highest turbo cores).

Underclocked +C +P states: Geekbench 17,800, 295 watts, peak 450 watts
Underclocked -C -P states: Geekbench 33,852 385 watts idle, peak 700 watts

Overclocked is 160x 26 multiplier, (4.48 Ghz highest turbo, 4.16 Base)
Overclocked +C +P states: Geekbench 32,150 350 watts idle, peak 685 Watts

So clearly underclocking offers an advantage in performance versus traditional overclocking. Depending on processor utilization, it is not so clear that it saves energy, since the C state generation is allowing less energy use.

Click to expand...

In this area of understanding, you are a far better student than I am, so congrats. I definitely would love to hear what Tutor has to say on this. BTW, what kind of RAM are you running with your system?...

SR2Mac said:

In this area of understanding, you are a far better student than I am, so congrats. I definitely would love to hear what Tutor has to say on this. BTW, what kind of RAM are you running with your system?...

Click to expand...

Corsair Dominator DDR3 2000 6 x 4 GB at the moment; I originally planned to use ECC memory for my purposes, but since we don't use AppleTMCYEdriver.kext with triple channel memory on this board in Lion (crashes, Disable.kext used to bypass), it added another layer of complexity and expense to the build at the cost of fewer overclocking range. In my experience, RAM errors are very rare on the Mac Pro; the role of the AppleTMCYEdriver is in part to log memory errors when they do occur. At some point I will putz around and see what impact double channel vs. triple channel memory has on performance. If it's not substantial I may switch to double channel ECC memory, which let my use AppleTMCYEdriver

I haven't tried tweaking underclocking at all at this point, I really just got it working last night. I had all the pieces in various states of play, but it seemed from all the threads that C and P states should be generated since they are enabled in BIOS. When Tutor sent me his org.chameleon.Boot.plist I realized that disabling C and P state generation was the key; then it took a few more iterations of trouble shooting to realize that I needed to rebuild kext caches without C and P states (or boot each time loading the kexts).

I suspect that the problem is really in the way that Mac/hacks handle P states when they are generated; tutor has shown that underclocking works well in Windows, and presumably in that case (where the P and C state generation is turned on in BIOS) P states are being generated.

braindeadfool said:

Corsair Dominator DDR3 2000 6 x 4 GB at the moment; I originally planned to use ECC memory for my purposes, but since we don't use AppleTMCYEdriver.kext with triple channel memory on this board in Lion (crashes, Disable.kext used to bypass), it added another layer of complexity and expense to the build at the cost of fewer overclocking range. In my experience, RAM errors are very rare on the Mac Pro; the role of the AppleTMCYEdriver is in part to log memory errors when they do occur. At some point I will putz around and see what impact double channel vs. triple channel memory has on performance. If it's not substantial I may switch to double channel ECC memory, which let my use AppleTMCYEdriver

I haven't tried tweaking underclocking at all at this point, I really just got it working last night. I had all the pieces in various states of play, but it seemed from all the threads that C and P states should be generated since they are enabled in BIOS. When Tutor sent me his org.chameleon.Boot.plist I realized that disabling C and P state generation was the key; then it took a few more iterations of trouble shooting to realize that I needed to rebuild kext caches without C and P states (or boot each time loading the kexts).

I suspect that the problem is really in the way that Mac/hacks handle P states when they are generated; tutor has shown that underclocking works well in Windows, and presumably in that case (where the P and C state generation is turned on in BIOS) P states are being generated.

Click to expand...

This is the RAM that Tutor is using and that I purchased.

http://www.buy.com/pr/product.aspx?sku=217438084

Is this the proper RAM to use or is what you're describing revealing issues with the current RAM that I have? Let me know thank you...

Also, could you please give me the specific RAM model kit that you're using. Thanks...

Lastly, I just PM'd you... Later...