MP 1,1-5,1 testing TB3 AIC with MP 5,1

Discussion in 'Mac Pro' started by Flint Ironstag, Sep 25, 2018.

  1. kings79 macrumors regular

    kings79

    Joined:
    Sep 16, 2015
    #26
    Ok cool.

    Following along with eager interest.
     
  2. joevt macrumors 6502

    Joined:
    Jun 21, 2012
    #27
    I think the 3 in GPIO3 is counting a GPIO on one of the intel chips (PCH or SuperIO or something). Possible alternatives for the Force Power setting in PC BIOS are discussed at
    https://egpu.io/forums/thunderbolt-...dge-thunderbolt-3-devices/paged/3/#post-43827
    https://egpu.io/forums/thunderbolt-...dge-thunderbolt-3-devices/paged/3/#post-44262

    You can find schematics of motherboards with Thunderbolt headers on the internet and follow the pins to their sources using freely available software. Some of that software works on macOS, others may require Windows.


    ==============

    I did some tests with a GC-TITAN RIDGE in a Sonnet Echo Express III-D (Alpine Ridge based) connected to a MacBook Pro 2015 (Falcon Ridge) running Windows 10. Connected to the GC-TITAN RIDGE was a OWC Thunderbolt 3 Dock and a OWC Helios 3 with Samsung 960 Pro NVMe SSD.

    I had previously installed Intel Thunderbolt software, but Boot Camp stopped it from working. Driver Genius 18 reinstalled Intel Thunderbolt drivers (software package 17.4.77.400), then the OWC peripherals appeared, but not all the devices were working - only one of the FL1100 USB controllers was usable (the one controlling the dock's audio and two of the five USB ports). The Intel software only listed the GC-TITAN RIDGE controller. It did not list the Thunderbolt 2 controller of the MacBook Pro.

    I think this test proves you can have trees of Thunderbolt devices - not just two chains. The only problem is software being able to set it up correctly - hot plug, wake / sleep, bus numbers, interrupts, io memory, etc.

    ==============

    Ubuntu only showed the Thunderbolt 2 controller of the MacBook Pro in its UI. A problem with the kernel is that it doesn't show the Titan Ridge as a separate domain (/sys/bus/thunderbolt/devices). lspci shows no peripherals connected to the Titan Ridge (except the NHI and USB as usual). I tried the kernel parameters "pci=assign-busses,hpbussize=20" which does allocate extra bus numbers but it doesn't cause peripherals connected to Titan Ridge to be scanned. Also, there are devices marked as hotplug+ (shown using lspci -vvnn) which also get extra busses even though they don't need them. I think modifying the kernel so you can choose what devices get extra buses could be useful.

    I think instead of modifying the kernel, an EFI driver could be made to scan PCI devices and Thunderbolt devices, then shuffle bus numbers and other resources. For macOS, a kext could be made to handle hot plug or at least add a user controlled rescan option. You could make them as smart as you like and not depend on Apple's software.
     
  3. equals macrumors newbie

    equals

    Joined:
    Jun 24, 2018
    #28
    Thanks for testing and for sharing this useful information.
    Hope to be able to join your testing journey soon. Can't wait for my card to arrive :)
     
  4. joevt, Oct 7, 2018
    Last edited: Oct 7, 2018

    joevt macrumors 6502

    Joined:
    Jun 21, 2012
    #29
    I connected a GC-TITAN RIDGE to an Amfeltec x16 gen 3 (four M.2) carrier card using 45 cm NVMe to PCIe x4 cables (four of them to create four external slots). A Corsair SFX 450W power supply was used to power the four external PCIe x4 slots.

    I used a Samsung 960 Pro 1TB PCIe NVMe SSD in a Sonnet Echo Express SE I for the Thunderbolt 3 speed tests.

    The amfeltec was placed in slot 2 of the MacPro3,1 and the link was set to g2x16 using the fast.sh script (because the g3 amfeltec boots into g1 speed in this computer's g2 slot).

    Since the amfeltec is low profile, I was able to cut an opening in its PCIe bracket for the cables. Each cable needs two 90° folds. The cables for the two NVMe slots on the other side of the amfeltec require an additional 180° fold. Folding the cables was difficult and probably causes stability problems but it's necessary to make it fit into slot 2.

    The Mac would occasionally hang in Windows or macOS (especially if all four external slots were populated - I was trying two GC-TITAN RIDGE with two GC-ALPINE RIDGE). Some of the external slots did not function correctly. One was working at g3x1: 261 write, 720 read; another starts at g1x4: 685 write, 746 read, but the fast.sh script can be used to set it to g2: 820 write, 1485 read or g3: 884 write, 2667 read (MB/s).
    Code:
    export LC_ALL=en_CA.UTF-8
    export LANG=en_CA.UTF-8
    
    cd ~/amfeltec_tests
    
    #chmod 744 fast.sh
    #chmod 744 pcitree.sh
    
    alias pcitree.sh="sudo ~/amfeltec_tests/pcitree.sh"
    alias fast.sh="sudo ~/amfeltec_tests/fast.sh"
    
    fast.sh 2 00:01.0            # amfeltec in MacPro3,1 slot 2
    
    # Before: PCIe 1.0 x16
    #  After: PCIe 2.0 x16
    
    
    fast.sh 3 02:0a.0            # amfeltec slot 2
    # Before: PCIe 2.0 x4
    #  After: PCIe 3.0 x4
    
    # Before: PCIe 1.0 x4
    #  After: PCIe 2.0 x4
    
    fast.sh 3 02:09.0            # amfeltec slot 4
    # Before: PCIe 3.0 x1
    #  After: PCIe 3.0 x1
    
    # problems:
    # slot 2: starts at g1x4.       685 write, 746 read   g2x4:820 write, 1485 read        g3x4:884 write, 2667 read.
    # slot 4: runs only at g3x1.    261 write, 720 read
    
    
    pcitree.sh
    #=========================================================================================
    ┬[0000:00]
    ├─00.0               # g1x4           00:00.0 [8086:4003] [0600] (rev 20) Host bridge: Intel Corporation 5400 Chipset Memory Controller Hub
    ├┬01.0-[01-0e]       # g2x16          00:01.0 [8086:4021] [0604] (rev 20) PCI bridge: Intel Corporation 5400 Chipset PCI Express Port 1
    │└┬00.0-[02-0e]      # g3x16 > g2x16  01:00.0 [10b5:8732] [0604] (rev ca) PCI bridge: PLX Technology, Inc. PEX 8732 32-lane, 8-Port PCI Express Gen 3 (8.0 GT/s) Switch
    │ ├┬08.0-[03]        # g3x4 > g1x0    02:08.0 [10b5:8732] [0604] (rev ca) PCI bridge: PLX Technology, Inc. PEX 8732 32-lane, 8-Port PCI Express Gen 3 (8.0 GT/s) Switch
    │ ├┬09.0-[04]        # g3x4 > g1x0    02:09.0 [10b5:8732] [0604] (rev ca) PCI bridge: PLX Technology, Inc. PEX 8732 32-lane, 8-Port PCI Express Gen 3 (8.0 GT/s) Switch
    │ ├┬0a.0-[05-0d]     # g3x4           02:0a.0 [10b5:8732] [0604] (rev ca) PCI bridge: PLX Technology, Inc. PEX 8732 32-lane, 8-Port PCI Express Gen 3 (8.0 GT/s) Switch
    │ │└┬00.0-[06-0d]    # g3x4           05:00.0 [8086:15ea] [0604] (rev 06) PCI bridge: Intel Corporation JHL7540 Thunderbolt 3 Bridge [Titan Ridge 4C 2018]
    │ │ ├┬00.0-[07]      # g1x4           06:00.0 [8086:15ea] [0604] (rev 06) PCI bridge: Intel Corporation JHL7540 Thunderbolt 3 Bridge [Titan Ridge 4C 2018]
    │ │ │└─00.0          # g1x4           07:00.0 [8086:15eb] [0880] (rev 06) System peripheral: Intel Corporation JHL7540 Thunderbolt 3 NHI [Titan Ridge 4C 2018]
    │ │ ├┬01.0-[08-0b]   # g1x4           06:01.0 [8086:15ea] [0604] (rev 06) PCI bridge: Intel Corporation JHL7540 Thunderbolt 3 Bridge [Titan Ridge 4C 2018]
    │ │ │└┬00.0-[09-0b]  # g1x4           08:00.0 [8086:1578] [0604] PCI bridge: Intel Corporation DSL6540 Thunderbolt 3 Bridge [Alpine Ridge 4C 2015]
    │ │ │ ├┬01.0-[0a]    # g3x4           09:01.0 [8086:1578] [0604] PCI bridge: Intel Corporation DSL6540 Thunderbolt 3 Bridge [Alpine Ridge 4C 2015]
    │ │ │ │└─00.0        # g3x4           0a:00.0 [144d:a804] [0108] Non-Volatile memory controller: Samsung Electronics Co Ltd NVMe SSD Controller SM961/PM961
    │ │ │ └┬04.0-[0b]    # g1x4           09:04.0 [8086:1578] [0604] PCI bridge: Intel Corporation DSL6540 Thunderbolt 3 Bridge [Alpine Ridge 4C 2015]
    │ │ ├┬02.0-[0c]      # g1x4           06:02.0 [8086:15ea] [0604] (rev 06) PCI bridge: Intel Corporation JHL7540 Thunderbolt 3 Bridge [Titan Ridge 4C 2018]
    │ │ │└─00.0          # g1x4           0c:00.0 [8086:15ec] [0c03] (rev 06) USB controller: Intel Corporation JHL7540 Thunderbolt 3 USB Controller [Titan Ridge 4C 2018]
    │ │ └┬04.0-[0d]      # g1x4           06:04.0 [8086:15ea] [0604] (rev 06) PCI bridge: Intel Corporation JHL7540 Thunderbolt 3 Bridge [Titan Ridge 4C 2018]
    │ └┬0b.0-[0e]        # g3x4 > g1x0    02:0b.0 [10b5:8732] [0604] (rev ca) PCI bridge: PLX Technology, Inc. PEX 8732 32-lane, 8-Port PCI Express Gen 3 (8.0 GT/s) Switch
    ├┬05.0-[0f]          # g2x16 > g1x16  00:05.0 [8086:4025] [0604] (rev 20) PCI bridge: Intel Corporation 5400 Chipset PCI Express Port 5
    │├─00.0              # g2x16 > g1x16  0f:00.0 [10de:1180] [0300] (rev a1) VGA compatible controller: NVIDIA Corporation GK104 [GeForce GTX 680]
    │└─00.1              # g2x16 > g1x16  0f:00.1 [10de:0e0a] [0403] (rev a1) Audio device: NVIDIA Corporation GK104 HDMI Audio Controller
    ├┬09.0-[10-18]       # g2x4 > g1x4    00:09.0 [8086:4029] [0604] (rev 20) PCI bridge: Intel Corporation 5400 Chipset PCI Express Port 9
    │├┬00.0-[11-17]      # g1x8 > g1x4    10:00.0 [8086:3500] [0604] (rev 01) PCI bridge: Intel Corporation 6311ESB/6321ESB PCI Express Upstream Port
    ││├┬00.0-[12-15]     # g1x4           11:00.0 [8086:3510] [0604] (rev 01) PCI bridge: Intel Corporation 6311ESB/6321ESB PCI Express Downstream Port E1
    │││└┬00.0-[13-15]    # g2x4 > g1x4    12:00.0 [12d8:2308] [0604] PCI bridge: Pericom Semiconductor Device
    │││ ├┬01.0-[14]      # g2x2           13:01.0 [12d8:2308] [0604] PCI bridge: Pericom Semiconductor Device
    │││ │└─00.0          # g2x2           14:00.0 [1b21:1242] [0c03] USB controller: ASMedia Technology Inc. ASM1142 USB 3.1 Host Controller
    │││ └┬02.0-[15]      # g2x2           13:02.0 [12d8:2308] [0604] PCI bridge: Pericom Semiconductor Device
    │││  └─00.0          # g2x2           15:00.0 [1b21:0625] [0106] (rev 01) SATA controller: ASMedia Technology Inc. Device
    ││├┬01.0-[16]        # g1x4 > g1x0    11:01.0 [8086:3514] [0604] (rev 01) PCI bridge: Intel Corporation 6311ESB/6321ESB PCI Express Downstream Port E2
    ││└┬02.0-[17]        # g1x4           11:02.0 [8086:3518] [0604] (rev 01) PCI bridge: Intel Corporation 6311ESB/6321ESB PCI Express Downstream Port E3
    ││ ├─00.0            # g1x4           17:00.0 [8086:1096] [0200] (rev 01) Ethernet controller: Intel Corporation 80003ES2LAN Gigabit Ethernet Controller (Copper)
    ││ └─00.1            # g1x4           17:00.1 [8086:1096] [0200] (rev 01) Ethernet controller: Intel Corporation 80003ES2LAN Gigabit Ethernet Controller (Copper)
    │└┬00.3-[18]         # g1x8 > g1x4    10:00.3 [8086:350c] [0604] (rev 01) PCI bridge: Intel Corporation 6311ESB/6321ESB PCI Express to PCI-X Bridge
    ├─0f.0               # g0x0           00:0f.0 [8086:402f] [0880] (rev 20) System peripheral: Intel Corporation 5400 Chipset QuickData Technology Device
    ├─10.0               #                00:10.0 [8086:4030] [0600] (rev 20) Host bridge: Intel Corporation 5400 Chipset FSB Registers
    ├─10.1               #                00:10.1 [8086:4030] [0600] (rev 20) Host bridge: Intel Corporation 5400 Chipset FSB Registers
    ├─10.2               #                00:10.2 [8086:4030] [0600] (rev 20) Host bridge: Intel Corporation 5400 Chipset FSB Registers
    ├─10.3               #                00:10.3 [8086:4030] [0600] (rev 20) Host bridge: Intel Corporation 5400 Chipset FSB Registers
    ├─10.4               #                00:10.4 [8086:4030] [0600] (rev 20) Host bridge: Intel Corporation 5400 Chipset FSB Registers
    ├─11.0               #                00:11.0 [8086:4031] [0600] (rev 20) Host bridge: Intel Corporation 5400 Chipset CE/SF Registers
    ├─15.0               #                00:15.0 [8086:4035] [0600] (rev 20) Host bridge: Intel Corporation 5400 Chipset FBD Registers
    ├─15.1               #                00:15.1 [8086:4035] [0600] (rev 20) Host bridge: Intel Corporation 5400 Chipset FBD Registers
    ├─16.0               #                00:16.0 [8086:4036] [0600] (rev 20) Host bridge: Intel Corporation 5400 Chipset FBD Registers
    ├─16.1               #                00:16.1 [8086:4036] [0600] (rev 20) Host bridge: Intel Corporation 5400 Chipset FBD Registers
    ├─1b.0               # g0x0           00:1b.0 [8086:269a] [0403] (rev 09) Audio device: Intel Corporation 631xESB/632xESB High Definition Audio Controller
    ├┬1c.0-[19]          # g1x1 > g1x0    00:1c.0 [8086:2690] [0604] (rev 09) PCI bridge: Intel Corporation 631xESB/632xESB/3100 Chipset PCI Express Root Port 1
    ├┬1c.1-[1a]          # g1x1 > g1x0    00:1c.1 [8086:2692] [0604] (rev 09) PCI bridge: Intel Corporation 631xESB/632xESB/3100 Chipset PCI Express Root Port 2
    ├┬1c.2-[1b-1c]       # g1x1           00:1c.2 [8086:2694] [0604] (rev 09) PCI bridge: Intel Corporation 631xESB/632xESB/3100 Chipset PCI Express Root Port 3
    │└┬00.0-[1c]         # g1x1           1b:00.0 [104c:823e] [0604] PCI bridge: Texas Instruments XIO2213A/B/XIO2221 PCI Express to PCI Bridge [Cheetah Express]
    │ └─00.0             #                1c:00.0 [104c:823f] [0c00] FireWire (IEEE 1394): Texas Instruments XIO2213A/B/XIO2221 IEEE-1394b OHCI Controller [Cheetah Express]
    ├┬1c.3-[1d]          # g1x1           00:1c.3 [8086:2696] [0604] (rev 09) PCI bridge: Intel Corporation 631xESB/632xESB/3100 Chipset PCI Express Root Port 4
    │└─00.0              # g1x1           1d:00.0 [14e4:4328] [0280] (rev 03) Network controller: Broadcom Inc. and subsidiaries BCM4321 802.11a/b/g/n
    ├─1d.0               #                00:1d.0 [8086:2688] [0c03] (rev 09) USB controller: Intel Corporation 631xESB/632xESB/3100 Chipset UHCI USB Controller #1
    ├─1d.1               #                00:1d.1 [8086:2689] [0c03] (rev 09) USB controller: Intel Corporation 631xESB/632xESB/3100 Chipset UHCI USB Controller #2
    ├─1d.2               #                00:1d.2 [8086:268a] [0c03] (rev 09) USB controller: Intel Corporation 631xESB/632xESB/3100 Chipset UHCI USB Controller #3
    ├─1d.3               #                00:1d.3 [8086:268b] [0c03] (rev 09) USB controller: Intel Corporation 631xESB/632xESB/3100 Chipset UHCI USB Controller #4
    ├─1d.7               #                00:1d.7 [8086:268c] [0c03] (rev 09) USB controller: Intel Corporation 631xESB/632xESB/3100 Chipset EHCI USB2 Controller
    ├┬1e.0-[1e]          #                00:1e.0 [8086:244e] [0604] (rev d9) PCI bridge: Intel Corporation 82801 PCI Bridge
    ├─1f.0               #                00:1f.0 [8086:2670] [0601] (rev 09) ISA bridge: Intel Corporation 631xESB/632xESB/3100 Chipset LPC Interface Controller
    ├─1f.1               #                00:1f.1 [8086:269e] [0101] (rev 09) IDE interface: Intel Corporation 631xESB/632xESB IDE Controller
    ├─1f.2               #                00:1f.2 [8086:2681] [0106] (rev 09) SATA controller: Intel Corporation 631xESB/632xESB SATA AHCI Controller
    └─1f.3               #                00:1f.3 [8086:269b] [0c05] (rev 09) SMBus: Intel Corporation 631xESB/632xESB/3100 Chipset SMBus Controller
    
    GC-Titan Ridge connected to Amfeltec.JPG
     
  5. handheldgames macrumors 68000

    handheldgames

    Joined:
    Apr 4, 2009
    Location:
    Pacific NW, USA
    #30
    Thanks again for the details. Have you had any luck installing the TB3 kext from the macbook pro that was mentioned earlier in the thread? It would be great to see this without having to initially boot from windows.

    It also looks like you are getting fast Reads with the Amfeltec pcie 3.0 adapter - congrats on having your cMP make the jump with the x4 PCIe 3.0 -> m.2 adapter. Although your writes are still somewhat capped. Any thoughts on why?
     
  6. MIKX macrumors 6502a

    MIKX

    Joined:
    Dec 16, 2004
    Location:
    Japan
    #31
    WOW ! This is amazing year for the cMP .. . :D .. . first we got NVMe booting then 5.0 GT/s with the 138 bootrom .. now perhaps Thunderbolt.
     
  7. joevt macrumors 6502

    Joined:
    Jun 21, 2012
    #32
    I don't know what kext you are referring to. I am using Mojave, which presumably has all the kexts that a MacBook Pro would be using.

    Maybe you are referring to EFI driver? I won't be trying to extract EFI modules.

    Maybe adding EFI strings (device properties) for Thunderbolt devices matching those used in a Titan Ridge (or any Thunderbolt) Mac may help. Device paths use the EFI DEVICE PATH PROTOCOL. There are protocols to convert device paths to and from text. Device paths are strings of device path nodes. Source code exists for all the device path node types. GfxUtil (various versions) is a utility to create device properties and implements some of the device path protocol.

    The reads are correctly scaling with the speed of the link g1, g2, and g3. The speeds for g1 and g2 links match my earlier tests without the amfeltec so I don't think the problem lies with the amfeltec.

    I did Titan Ridge USB 3.1 gen 2 tests at #2095. Read was capped at 400 MB/s in a g1 slot but was not capped in a g2 slot even when the g2 slot was running at g1 speed. Maybe the initial speed of the slot has something to do with it.
    --- Post Merged, Oct 8, 2018 ---
    Device properties can also be defined in ACPI. Use MaciASL to look at the DSDT for DSM methods that call DTGP.

    The efi strings are setup by efi modules that call the EFI_DEVICE_PATH_PROPERTY_DATABASE_PROTOCOL. I think GfxUtil can view those.

    efi strings can also be defined in /Library/Preferences/SystemConfiguration/com.apple.Boot.plist.

    kexts can also add device properties.
     
  8. edgerider macrumors regular

    edgerider

    Joined:
    Apr 30, 2018
    #33
    Hi there all!

    If you guys manage to make thunderbolt3 to works on a cMP, It is just plenty awsome, missing native usbc and thunderbolt connectivity on a cMP is the only thinh that I am missing on my Maxed out cMP.....
    I think if you manage to do so, Apple is in trouble big time, because powerwise a double X5690 with a good gpu and 128gb of ram is plenty enough for most of us...
     
  9. Flint Ironstag thread starter macrumors 6502a

    Flint Ironstag

    Joined:
    Dec 1, 2013
    Location:
    Houston, TX USA
    #34
    Potentially breathes some new life into Xserve, as well.
     
  10. Manzanito macrumors 6502

    Manzanito

    Joined:
    Apr 9, 2010
    #35
    Would a thunderbolt display work with a cMP using this card?
     
  11. joevt macrumors 6502

    Joined:
    Jun 21, 2012
    #36
    Probably as well as any Thunderbolt peripheral can. The DisplayPort conversion should always work, even with an Alpine Ridge card. The functions of the display that require PCIe communication over Thunderbolt may or may not work (USB, audio, camera, brightness control). To get the peripherals working, I have to boot into Windows first. I have not tried a Thunderbolt display. The closest thing to a Thunderbolt display that I have is a dock with DisplayPort output (which is basically what a Thunderbolt display is). Hot plug, sleep, wake, may all have problems or not work.
     
  12. LightBulbFun macrumors 68020

    LightBulbFun

    Joined:
    Nov 17, 2013
    Location:
    London UK
    #37
    this is horribly off-topic but are you the same Joevt from the Xpostfacto forum days? :)
     
  13. joevt macrumors 6502

    Joined:
    Jun 21, 2012
    #38
    Yes. I have the source code for my PowerMac 8600 Mac OS X video input driver somewhere on this computer. My PowerMac 8600 and 9500 are packed away though.
     
  14. LightBulbFun macrumors 68020

    LightBulbFun

    Joined:
    Nov 17, 2013
    Location:
    London UK
    #39
    hah, very cool :) its a shame the old forums are no longer online, so much valuable info there and the web archive only has a small part of it archived sadly

    have you checked out the PowerPC Section of MR?

    theres some projects on the go there that you may enjoy

    including a couple of my own

    https://forums.macrumors.com/thread...os-x-10-5-8-power-macintosh-9600-fun.2144305/

    https://forums.macrumors.com/threads/success-with-the-power-macintosh-4400.2077767/

    anywho to keep this on topic

    I remember coming across this a while back https://www.journaldulapin.com/2018/01/27/thunderbolt-displayport/

    even if we cant use the TB3 part of it, in theory we could use it to drive a 5K LG UltraFine in a MacPro5,1 :)
     
  15. MIKX macrumors 6502a

    MIKX

    Joined:
    Dec 16, 2004
    Location:
    Japan
    #40
    joevt

    I really admire the sheer tenacity and doggedness of people like you & Gilles, dosdude1, foxfoobar, tsialex, h9826790, handheld games etc.

    Since 2017 it has been an amazing time for the cMP. Just a few years ago we learned how to flash GPUs for EFI & Link speeds of 5.0 GT/s. We also started being able to use USB 3.0, now we can also boot from M.2 NVMe. We have a plethora of new, powerful GPU's.

    Thunderbolt in the cMP is a whole other kettle of fish. I wish you good luck and inspiration.

    PS : Apple in having included NVMe booting in 140.0.0.0.0 bootrom and 'turning on '5.0 GT/s proves to me that they DO watch what we are doing here, have listened to us & realize that there are still hundreds of thousands of viable cPM owners out there who would make nice Mac Pro 7.1 customers.

    Of course Apple will price the MP 7,1 out of reach of non-professionals.
     
  16. AlexMaximus macrumors 6502a

    AlexMaximus

    Joined:
    Aug 15, 2006
    Location:
    A400M Base
    #41
    Thunderbolt 3 is mostly about PCIe card expansion. What do you think about the PCIe expansion based on PCI ?
    Is this solution a lot slower on the bus bandwith, then TB3 ?

     
  17. joevt macrumors 6502

    Joined:
    Jun 21, 2012
    #42
    Transmission of PCIe data over a Thunderbolt 3 cable is capped at 22 Gbps (2750 MB/s) which is less than the max allowed by the Thunderbolt controller's PCIe 3.0 x4 connection.

    PCIe expansion is more expensive (the cables) but isn't capped like Thunderbolt. If you use a PCIe 3.0 x4 upstream cable, then you get PCIe 3.0x4 speed. You can get cables up to PCIe 3.0x16 (126 Gbps). There is no bandwidth used for DisplayPort.

    My solution with the amfeltec, if I could get it to actually work without errors, would be most similar to a PCIe expansion box with PCIe 3.0x16 upstream and four PCIe 3.0x4 slots. The difference is where the cable lengths are. In my case the upstream cable is 0 m (direct connect to the Mac Pro's PCIe slot), and the downstream port cables are 0.5 m (from the M.2 adapters). With a PCIe expansion box, the upstream cable is 1.5m and the slots are 0 m. The cables of the PCIe expansion box are easier to use - I had to do some horrible things with the M.2 adapter cables from the amfeltec - not to mention the hole I put in the amfeltec's back plate. I attached a picture of the parts:
    amfeltec pieces.JPG

    I think I would have had better luck using a PCIe x16 riser cable, connect the amfeltec to that, then used non-cabled M.2 to PCIe x4 adapters. The PCIe 2.0 signals to the amfeltec can probably survive better over a cable than the PCIe 3.0 signals from the M.2 slots.
     
  18. edgerider macrumors regular

    edgerider

    Joined:
    Apr 30, 2018
    #43
    hi there!

    after a lot of goofing around, I went the Pcie expension way !
    first with gen2x8 on a netstor 16 bay rack, then on a gen2x16 cyclone microsytem 5 slot ATX expender because I hated loosing 8 lane on my second 16x slot of the macpro. This one was placed in a SC848 chassis with 24 bay and a redundent 1200w psu.
    the cool part was that i could had my anfeltec in slot 1, the expender host card in slot 2, and in the expender my 3 P4000, my raid card, and a free slot, a usb3 card in slot 3, and a 10Gbe card in slot 4.
    the only problem was the sc846 chassis is loud as ****... no kidding I mean really loud!!! And the heat generated by the 3 P4000 the raid card and 24 sas drive is just stupid.
    So because I’m an idiot I bought two 3.1 xserve for 200€, have slaped them in a 12U rolling enclosure and my data will be hosted on one xserve with a 10gbe card, running the sc846 the raid.
    For the second xserve, i have found a cheap nvidia tesla 4 GPU enclosure with the host card and 2 cable, it will have 4 P4000 gpu and will only be a dedicated video converter/exporter for premiere.
    the 12U chassis will stay ine the basement.
    Now I am gutting out an old Powermac G5 to have my new 10 slot cyclone expender in a silent and « MacPro look »
    enclosure siting next to my 4.1.
    this will have 4 P4000, and as much nvme drive on regular Kryo M2 adapter,
    to build a fast scratch drive.

    over the 10Gbe network I get consistent 700/900 Mb/s read write to the xserve.
    and as my preview will be on the nvme array, i think i will be ok.

    the good thing is those cyclone micro systems expender are cheap to buy if you know witch one to look for.

    but you have to know what you are buying because some board are only 4x some are 8x and a very few are 16x.
    some are Gen1....

    any how, if you have the right host/cable/backplane, they are basically a straight plug an play.

    I am still experimenting with the tesla enclosure, because it needs every slot populated to turn on, but other than that it is just a 4x16 epender in a 1u chassis with it’s onwn psu. the obsolete tesla card can be replaced by newer one, but you have to find card with a low TDP such as a P4000 because the psu is only 900w max so 4x200w max peak.
     
  19. rawweb macrumors 6502a

    rawweb

    Joined:
    Aug 7, 2015
    #44
    I would be very tempted to pick up this card and try it with my 2012 5,1. I've got 2 LG UltraFine 5k's sitting here collecting dust. Frankly, even if it could drive 4k I would be pretty happy. I just don't think this will work.
     
  20. LightBulbFun macrumors 68020

    LightBulbFun

    Joined:
    Nov 17, 2013
    Location:
    London UK
    #45

    iv not personally tested it, but AFAIK it seems like as long as the card is powered on somehow and you feed it a proper DP signal then you should be able to plug a thunderbolt display and have it display a picture

    I know this has been done with an Apple thunderbolt display, I dont know of anyone doing it with an ultraFine 5K monitor however in theory using a TB3 card and feeding it a DP1.2 signal it should work :)
     
  21. Pressure macrumors 68040

    Pressure

    Joined:
    May 30, 2006
    Location:
    Denmark
    #46
    The GC-TITAN RIDGE supports DP1.4, so it should be able to drive the 5K displays if it initializes ... otherwise I would gladly take one of those 5K displays off your hands if the price is right :p
     
  22. joevt macrumors 6502

    Joined:
    Jun 21, 2012
    #47
    The GC-TITAN RIDGE supports DP1.4, but the LG 5K is a dual link DP1.2 display - it requires two DisplayPort signals to support 5K. So you need two DisplayPort cables from the graphics card to the GC-TITAN RIDGE (or any Thunderbolt 3 add-in card with two DisplayPort inputs). Each LG 5K display requires a separate add-in card because each add-in card only has two DisplayPort inputs. The add-in card needs to be connected to your computer to be able to use the functions of the display that require USB. With previous add-in cards this did not work - the PCIe functions were not enabled. With the GC-TITAN RIDGE, you can enable the PCIe functions by booting into Windows 10 first, then warm boot into macOS. The PCIe functions allow communication to the LG 5K's USB controller, which controls the audio, USB ports, camera, and brightness control.

    I believe if the add-in card only has one DisplayPort connection, then the LG 5K will operate at 4K. I don't know if you can get two LG 5K displays to work at 4K with one add-in card. Maybe connect one DisplayPort only, then a 5K display, then the other DisplayPort, then the other 5K. If that doesn't work, make the first display a different display (4K or less), continue following the same steps, then disconnect the 4K, and connect the 5K.
     
  23. Pressure macrumors 68040

    Pressure

    Joined:
    May 30, 2006
    Location:
    Denmark
    #48
    The LG Ultrafine 5K is TB3. You must be referring to the old one.

    I was talking about DP1.4 output from the graphic card into the GC-TITAN RIDGE and TB3 to the monitor.

    Otherwise you would just plug the 5K monitor straight into the graphic card.
     
  24. Squuiid macrumors 65816

    Squuiid

    Joined:
    Oct 31, 2006
    #49
    No. @joevt is correct, the LG Ultrafine 5K does indeed require two DP1.2 signals. Sure, these signals currently pass through one Thunderbolt 3 cable but they are in fact two separate DP1.2 signals.

    See here for more details on the new TB3 controller which may change things in future:
    https://www.macrumors.com/2018/07/20/2018-macbook-pro-titan-ridge-displayport/
     
  25. Pressure macrumors 68040

    Pressure

    Joined:
    May 30, 2006
    Location:
    Denmark
    #50

Share This Page