All We Know About Maximizing CPU Related Performance

Discussion in 'Mac Pro' started by Tutor, Mar 3, 2012.

  1. Tutor, Jun 22, 2015
    Last edited: Jun 22, 2015

    Tutor thread starter macrumors 65816

    Tutor

    Joined:
    Jun 25, 2009
    Location:
    Home of the Birmingham Civil Rights Institute
    #1276
    1) MB-X9DAXF&c = E5 V1 (Sandy Bridge) & E5 V2 (Ivy Bridge {and Sandy Bridge but w/o 1866 MHz memory speed})
    2) MB-X10DAX = E5 V3 (Haswell {but no currently good dual Xeon OSX support, since there're no Haswell MPs and from the looks of things, it currently looks like Apple's 2011 MP abandonment redux}). But hope springs eternal that no current MP update or upgrade doesn't mean that the child has been left on someone else's doorstep.
     
  2. Tutor thread starter macrumors 65816

    Tutor

    Joined:
    Jun 25, 2009
    Location:
    Home of the Birmingham Civil Rights Institute
    #1277
  3. Tutor thread starter macrumors 65816

    Tutor

    Joined:
    Jun 25, 2009
    Location:
    Home of the Birmingham Civil Rights Institute
    #1278
    Congratulations on your build. That's an excellent achievement.

    When I get some time to get native power management working, my Geekbench score should improve greatly to the high/mid 50k to low 60k range, using 32 real cores and no hyper cores.
     
  4. PunkNugget macrumors regular

    PunkNugget

    Joined:
    Apr 6, 2010
    #1279
    Thanks Tutor for your continued input. As I've been in communication with RD, he mentioned using this mobo as well, the ASUS Z10PA-D8C. But for some reason I don't see it available anywhere online for sale. I do however see that the ASUS Z10PA-D8 is available for sale, but doesn't allow you to have 3 GPUs, only 2 GPUs to use on this Mobo. The Z10PZ-D8 will hold more GPUs than the other one. Also, it's a more recent release for the E5-2600 v3's. So if ever an update were to happen to have the Mac OS X go over the 32core "cap" (if El Capitan can break that "Cap") then I'd like to look into this as well as a possible option. But only if RD says it's possible. His main concern is MEM MAP issues on startup. As he stated before:

    "Need to remember about the Mem Map issues so less options it has the less drivers it loads and the more stable it becomes."

    I need to remember that when I see a Mobo that I think could be used. What do you think Tutor? I only ask because of of the two Supermicro Mobos that you suggested and the level of detail you gave about the possibility of using them for Mac OS X. I will pass this off to RD and get his thoughts. Again, thanks… :)
     
  5. PunkNugget macrumors regular

    PunkNugget

    Joined:
    Apr 6, 2010
    #1280
    I was going to ask Tutor, what system setup will you be using to make that happen? Will you be able to do that on your existing hackintosh build?
     
  6. Tutor, Jun 22, 2015
    Last edited: Jun 22, 2015

    Tutor thread starter macrumors 65816

    Tutor

    Joined:
    Jun 25, 2009
    Location:
    Home of the Birmingham Civil Rights Institute
    #1281

    Here's my thoughts:

    How you’ll use GPUs determines how much the “x”s and “v”s really matter.

    Food for thought:

    (1) PCIe “x”s and “v”s don’t matter as much for compute GPUs, but those “x”s and “v”s usually are very important for display GPUs. For GPUs used only for compute functions, the PCIe “x"s and “v"s affect only the time it takes to get the data to be computed into the ram of the GPU. Once the data to be computed is in a GPU being used for compute-only purposes, it matters not much what is the speed of the PCIe connection because data once within the compute GPU, calculates at the speed of the GPU processor. To be sure, there may be seconds of transmission lag times, but unless you’re sitting their counting seconds, you probably won’t notice it. This is particularly so - the heavier the rendering/compute task. That’s where your GPU specs count and prove their worth. So x1, x4, x8 x16 and v1, v2 and v3 PCIe matters most mainly only for GPU(s) assigned display functions by the user. But, I’d personally prefer a system faster throughout the whole compute process. But $$ may be a limiter.

    (2) The display GPUs are best seated in x16 v2 or, preferably, v3 slots, or at least connected by an x16 to x16 powered riser cable to such a PCIe slot.

    Superbiiz has the Supermicro X10DAX-O Dual LGA2011/ Intel C612/ DDR4/ SATA3&USB3.0/ A&2GbE/ EATX Server Motherboard for sale for $414.99. That “DAX” in the name means that it’s overclockable. This version (that supports v3 Xeons) is overclockable by a factor of x1.05 [ http://www.superbiiz.com/detail.php?name=MB-X10DAX ] . It has three x16 v3 double wide PCIe slots. For your described use, you could just insert up to three double wide GPUs in the three x16 double width space slots.

    But even more importantly, at least to me, is that it has three x8 sized slots (two v.3 and one v.2)*/. From my study of GPU centric builds, particularly those based on Supermicros, to me that means that this motherboard is likely capable of supporting at least six double wide GPUs or even more. To harvest this potential will require use of (A) at least one Amfeltec GPU-oriented Splitter [ http://amfeltec.com/splitters-gpu-oriented/?view=list ] or (B) a combination of three x16 to x16 powered riser cables [ http://www.ebay.com/itm/161730359407?_trksid=p2057872.m2749.l2649&ssPageName=STRK:MEBIDX:IT ] and three PCIExpress Riser Cards x8 to x16 Left Slot Adapter For 1U Servers [ http://www.ebay.com/itm/NEW-HOT-PCI...454?pt=LH_DefaultDomain_0&hash=item58a9682f7e ] to install up to three more GPUs in the three x8 sized slots. Moreover, using a combination of both (A) and (B), one may be able to successfully install even more than six GPUs in the system!

    So, to drive two displays I'd get one GPU with that has that capability and put it in the last/bottom PCIe slot of the X10DAX (depending on its orientation).

    */ The DAX version that supports V2 Xeons costs from $530 to $790 [ http://www.superbiiz.com/query.php?s=supermicro+dax ]. It has two PCIe x16 v3 slots (a single width's space apart) and four PCIe x8 slots (one v2 and two v3). To max out the X9DAX GPU-wise requires four PCIe Express Riser Card x8 to x16 Left Slot Adapter For 1U Servers and four x16 to x16 powered riser cables, plus one x16 to x16 powered riser cable to mount a second double wide PCIe card in the x16 PCie slot that's a single width's space from the last PCIe slot. One could use an Amfeltec GPU-oriented Splitter to harvest the IO potential indicative of having the four PCIe V3 x8 slots. But, note that the Amfeltec GPU-oriented Splitters operate at x4 transmission speeds - so one could use those GPUs for compute purposes with no significant disadvantage - in sum, that use-case is best used for compute-only GPUs. The X9 DAXs are overclockable by a factor of 1.0755 with well-binned Sandy and Ivy Bridge Xeons. Given the technological differences between E5 v2s and v3s Xeons, that 1.0755 (v1 & v2 E5 Xeons) vs. the 1.05 (v3 E5 Xeon) overclock factor may likely be a wash.
     
  7. PunkNugget macrumors regular

    PunkNugget

    Joined:
    Apr 6, 2010
    #1282
  8. PunkNugget macrumors regular

    PunkNugget

    Joined:
    Apr 6, 2010
    #1283
    Well, here's what I'm sending RD to start the installation tests for my next Hackintosh build:

    1 x Asus Z10PE-D16
    http://www.newegg.com/Product/Product.aspx?Item=N82E16813132257

    2 x E5-2640 v3's
    Engineering Samples that have been used and are warranted for 45 days. Plus I got them done from eBay so I'm protected regardless as we've had "paper trail" conversations about the details of these CPUs. In any case I only paid $500.00 each for them (as new ones retail for $967 each). Hopefully they work fine. If not then I'll return them and get new ones…

    2 x EVGA 980 Ti GPUs
    These will meet more than enough of my needs for now... until I buy a third one - LOL ! (just kidding the board can only handle 2 anyway).

    2 x Crucial DDR4-2133 64GB (4x16GB) RAM kits (128GB Total)
    http://www.superbiiz.com/detail.php?name=CTD421RX64

    I'm going to be sending RD my LEPA G1600 PSU for the test (as his doesn't have enough power to handle all this). In the end I'm going to be using my EVGA 1600 P2 PSU for the final build. Right now I'm flushing and testing for leaks with the three Coolgate Rads I just purchased. I'm really looking forward to this build and hope that RD is successful in getting all this to work with Yosemite 10.10.+, but we'll see. ;)
     
  9. PunkNugget, Jul 7, 2015
    Last edited: Jul 7, 2015

    PunkNugget macrumors regular

    PunkNugget

    Joined:
    Apr 6, 2010
    #1284
    Just to keep you guys updated with what's going on. RD has received most everything. Now all their is left for him to receive is the LEPA G1600 PSU and the Crucial 128GB of RAM. I'm excited and hopeful… :)

    On a side note, I had thoroughly cleaned out my 3 Rads and boy were they dirty. I don't care what any company out there claims, you MUST thoroughly clean your Rads. The ones that I have (Coolgate 2 x CG360-G2 & 1 x CG240-G2) about a dozen thorough washings, by using my home water softening system along with a 5 filter pass system of water. I then boiled that water poured into the rad, then vigorously shaked them on one end then turning over the rad and vigorously shaking them on the other end, then dumping out the water in a white bowl to see the sediment. One of the rads literally took 12 times before I felt comfortable enough that is was clean where I couldn't see anything. For the finishing touch (in the cleaning and prepping of my Rads), I used this 1800LPH pump with distilled water to flush out any other contaminants that might of been left behind. The 1800LPH pump that I purchased was this one:

    http://www.ebay.com/itm/291122646878?_trksid=p2047675.m570.l4467&_trkparms=gh1g=I291122646878.N19.S2.M-9504

    and let me tell you, the head pressure on this thing is amazing. It pumps out water like a paired up set of Swiftech's MCP355's:

    http://www.swiftech.com/mcp35512vdcpump.aspx

    and if you're into water cooling, then you know what I'm talking about. The pump was that strong. I tried to find a cheaper pump than that, and the next lowest priced one was around $98.00, then the next one after that was $119.00. Again, great pump and it was the finishing touch to throughly cleaning out anything else out of those Rads and I made sure to use distilled water. I'll be loading up a video soon to show you what I was able to do in my cleaning process. Let me tell you it wasn't easy and took about 7 hours of dedicated time to clean all three Rads. It's worth it though as I won't be having any issues later with them getting dirty or gummed up with crud, as that can happen over time.

    You're supposed to flush out your system at least every 12 to 18 months, but I don't buy that theory, as I have used the current Hackinbeast (over 3 years old) with the very same coolant (Koolance UV Green Coolant), and all I do every 6 to 8 months is just add coolant to the existing system and it's just as clean as it was the day I put in the fluid. The advice is not to use Red or Purple dye (as purple dye has red in it). Something about red dye causes the inside of the pump reservoir and rads to get gummed up. I had that problem with my initial Hackintosh (the Black Sheep) and the purple dye after about 18 months later had this murky-goopy nasty look to it. It look like flowing mucus - LOL! Just gross and nasty. The green dye though, was just fine, no issues at all.
     
  10. PunkNugget macrumors regular

    PunkNugget

    Joined:
    Apr 6, 2010
    #1286
    I hope so. RD's install process (and he just emailed this about 2 min ago) is going to be this:

    "We are starting with 10.9.5 for testing with a GTX 780 and then see if we can install normally with HT disabled. Then if that fails install using another system to the test HDD and if that fails apply the 32 Core Patch to the kernel and then go from there and move to Mac OS X 10.10 and 10.11. That's the plan."

    Again, I hope this works. We'll find out in a couple of weeks… ;)
     
  11. lowendlinux Contributor

    lowendlinux

    Joined:
    Sep 24, 2014
    Location:
    North Country (way upstate NY)
    #1287
    I’ll stay on 10.9.5 and then go to 10.11 Yosi doesn’t do much for me, mine was tough your’s will be tougher.
     
  12. PunkNugget macrumors regular

    PunkNugget

    Joined:
    Apr 6, 2010
    #1288
    What is your current system setup?
     
  13. PunkNugget macrumors regular

    PunkNugget

    Joined:
    Apr 6, 2010
    #1290
    Ok, what about the CPUs, GPUs and RAM? Also, what is your GB score?
     
  14. lowendlinux Contributor

    lowendlinux

    Joined:
    Sep 24, 2014
    Location:
    North Country (way upstate NY)
    #1291
    It’s got a pair of 5675’s, 32GB, and I just sold the pair of GTX 560’s that were in it so in the morning it will get a I’ll get the GTX 780 non-Ti on it’s way. Though as I think about it I might just go maxwell even if it means going 10.10 that way I cut some back on some of the power consumption but I’ll think about it. I’ve got nothing going on right now and the FX1800 will drive the screen.
     
  15. PunkNugget macrumors regular

    PunkNugget

    Joined:
    Apr 6, 2010
    #1292
    Cool. So your system is as "dated" as mine (3 to 4+ years old). Well at least it's still working for you. I like my current setup, but feel it's time to update...
     
  16. Silicman macrumors newbie

    Joined:
    Oct 11, 2013
    #1293

    Hi ccr4700,

    I've got Z9PA-D8C board like you, and 2 E5-2670 v1 Sandy, 1 EVGA GTX680 4G, 1 GA 760 2G. I made it up and run just fine with Mavericks 10.9.5. But I can't get Yosemite or El Capitan on it, always stuck at PCI configuration begin. Is there any luck with your board? I need Yosemite or El Capitan to work with FCPX.

    Here some result from my system.

    Main: Z9PA-D8C
    CPU: 2 x E5-2670 (v1)
    RAM: 8 x 8G bus 1333
    VGA: 1 x GTX 680 4G, 1 x GTX 760 2G
    SSD: Samsung 850 evo
    OSx: 10.9.5


    ** CPU Power management:

    Feb 6 21:05:29 localhost kernel[0]: AICPUPMI: CPU P-States [ 26 ]
    Feb 6 21:05:29 localhost kernel[0]: AICPUPMI: CPU P-States [ 12 26 ]
    Feb 6 21:05:31 dungpc1 kernel[0]: AICPUPMI: CPU P-States [ 12 26 30 ]
    Feb 6 21:05:33 dungpc1 kernel[0]: AICPUPMI: CPU P-States [ 12 26 30 32 ]
    Feb 6 21:07:54 dungpc1 kernel[0]: AICPUPMI: CPU P-States [ 12 26 30 32 33 ]
    Feb 6 21:12:06 dungpc1 kernel[0]: AICPUPMI: CPU P-States [ 12 26 30 31 32 33 ]

    ** Geekbench: http://browser.primatelabs.com/geekbench3/5170280

    ** VGA:
    Screen Shot 2016-02-06 at 10.39.23 PM.png Screen Shot 2016-02-06 at 9.55.08 PM.png
     
  17. ClassicMartini macrumors newbie

    ClassicMartini

    Joined:
    Oct 28, 2015
    Location:
    Kessel Run
    #1294
    There was a non obvious follow up about 5 months later.

     
  18. maccam70 macrumors newbie

    Joined:
    Apr 10, 2016
    #1295
    Hi, I'm trying to install mavericks in this mobo (Z9PA-D8 +2x E5-2670 v1 + quadro K2000) with no avail, I can get to the installer only with -x npci=0x3000 nv_disable=1 otherwise black screen.
    How do you manage to boot with two graphic cards? do yo use a patched dsdt or ssdt?
     
  19. ccr4700 macrumors newbie

    Joined:
    Jun 21, 2015
    Location:
    Japan
    #1296
    Sadly I couldn't resolve the issue of Z9PA-D8...
    I gave up using Z9PA-D8, and I bought Z9PE-D8 WS. I struggled with this mobo, but finally I made El Capitan working.

    I have no solution to work GPU connected to CPU1, so I had to use Z9PE-D8 WS (It has slot5-7 connected to CPU2).
     
  20. maccam70 macrumors newbie

    Joined:
    Apr 10, 2016
    #1297
    Hi ccr4700, sorry to hear that.
    Anyway I was asking to Silicman as he seems to have Mavericks running with TWO graphics cards plugged in the Z9PA-D8!!!
    I've managed to install both Mavericks and El Capitan, but I can't get graphics acceleration in any of them…
    Mavericks seems to work better as it recognizes all my memory right away (with el Capitan only half is shown) and I think is a bit smoother & responsive.
    The main use of my workstation is 3DS MAX so OSX is not really needed although I'll keep trying just for the fun of it.
     
  21. LBattis macrumors newbie

    LBattis

    Joined:
    Aug 10, 2014
    Location:
    Berkeley, CA
    #1298
    Hi Tutor,

    Just a note to make sure you're OK. It's been a while since I've seen a post from you.

    Laurens Battis
     
  22. ClassicMartini, Dec 1, 2016
    Last edited: Dec 1, 2016

    ClassicMartini macrumors newbie

    ClassicMartini

    Joined:
    Oct 28, 2015
    Location:
    Kessel Run
    #1299
    Tutor is active in the "Octane" forums.

    Octane (OTOY) is a CUDA gpu renderer. The more high end CUDA (nVidia) cards you have the faster the render output.

    Something no modern Mac can do anymore.

    We miss Tutor's wisdom, insights, tenacity and generosity here too.
     
  23. LBattis macrumors newbie

    LBattis

    Joined:
    Aug 10, 2014
    Location:
    Berkeley, CA
    #1300
    Glad to hear it, Thanks for the update.
     

Share This Page