joevt

You are mixing units - and doing it backwards. 22 Gbps = 2.75 GB/s = 2750 MB/s = 2.561 GiB/s. Binary units (such as GiB) should not be used for data transfer numbers anyway. I think the OWC number of 2800 MB/s is just 2750 MB/s rounded up. They don't show benchmarks so it's hard to say where they got that number from. The 22 Gbps limit of PCIe traffic over Thunderbolt 3 might not be a hard limit. Newer CPUs and controllers may allow slightly better performance. I saw a benchmark of 2978 MB/s for a Thunderbolt 4 controller in a YouTube video. Right. The TB3 number of 22 Gbps for PCIe data does not include any Thunderbolt and PCIe overhead that is required to transmit that amount of data. The USB gen 2x2 number of 20 Gbps, which is actually 19.39 Gbps after considering the 128b/132b wire encoding, still includes USB protocol overhead. It is interesting to note that the 40 Gbps number of Thunderbolt takes into account the 64/66b wire encoding of Thunderbolt (Thunderbolt transmits 41.25 Gbps on the wires). USB4 will use 40 Gbps on the wire but will use 41.25 Gbps when supporting Thunderbolt 3.

Did you mean 14:30 with the History Graph window? I don't recognize it either.

Right. You might get PCIe tunnelling to work, but probably all the Thunderbolt drivers will not load, so you can't see things like the cable connection (link rate and width), firmware versions, ports, etc. and Thunderbolt packets that are not PCIe or DisplayPort will not be usable (Thunderbolt IP (Mac/Windows/Linux), Thunderbolt Target Display Mode (Mac only), and Thunderbolt Target Disk Mode (Mac Only)). I don't think there's anything special about Thunderbolt displays like the LG UltraFine 5K or Apple Pro Display XDR - they only use DisplayPort and PCIe if they're being connected with Thunderbolt instead of USB-C with DisplayPort alt mode (but they have no buttons and may be finicky). osy86's work is a little more involved than just an SSDT. He was able to get the Thunderbolt info to show up in System Information.app like you show in your screen shots, so maybe Thunderbolt IP, Thunderbolt Target Display Mode, and Thunderbolt Target Disk Mode also work? I don't know if his changes require Thunderbolt stuff in EFI or a Thunderbolt header on the motherboard. If you can get this far, then maybe the macOS driver will enable dual HBR3 to get 6K from the XDR using the Radeon VIIs and a GC-TITAN RIDGE.

They could connect the Apple Pro Display XDR and get 6K if they had a Navi card because Navi supports Display Stream Compression (DSC). The W5700 has a USB-C port so they can also get the USB functions of the display (USB 2.0 ports, brightness control, presets, rotate?). Even if you have a Navi GPU that doesn't have a USB-C port, you might be able to get the USB functions using a Sunix UPD2018, Huawei VR 2 Computer Connection Cable, or Wacom Link Plus since only HBR2 link rate is required with DSC for 6K (I have all the above but don't have a Navi GPU or an XDR display to test). GPUs like the Radeon VII that don't support DSC need a Thunderbolt connection to get 6K using a dual HBR3 connection. For this you would use a GC-TITAN RIDGE, but something would need to be changed to make it allow dual HBR3. macOS does this for Thunderbolt 3 Titan Ridge controllers of Macs, and Blackmagic eGPUs recently got a firmware update to allow this with its Titan Ridge Thunderbolt 3 controller (it's the only eGPU that wires the GPU into the Thunderbolt 3 controller). The problem is that dual HBR3 would normally exceed the 40 Gbps bandwidth of Thunderbolt 3 but Apple can allow dual HBR3 because 6K doesn't require the entire bandwidth of dual HBR3 and Thunderbolt does not transmit the stuffing symbols of the DisplayPort streams. The bandwidth allocation method for DisplayPort out adapters is described in the USB4 spec and I think it also applies to Thunderbolt. Normally, if you connect a DisplayPort 1.4 display, the Thunderbolt controller gives it a HBR3 connection even if you use a resolution that doesn't require HBR3. The next connected display to the same Thunderbolt port can only get a HBR connection. In macOS, the AppleThunderboltDPOutAdapter.kext has a string "Thunderbolt DP - bandwidth to be derived from local and remote caps instead of common cap" which might be applicable. To make a GC-ALPINE RIDGE or GC-TITAN RIDGE usable on a motherboard that does not support Thunderbolt (no Thunderbolt header?) they should have their Force Power pin tied to 3.3V (SATA power?) with a pull up resister (8.2kΩ) (I made a breadboard to connect two of each card). For the GC-ALPINE RIDGE, you'll also need to poke some stuff into the PCIe configuration space (0x54C) (this can be done using the EFI Shell). That should allow the USB 3.1 gen 2 controller of the Thunderbolt controller to be usable at least. I suppose some SSDT modifications for Thunderbolt (loaded by Clover) can do that and might allow PCIe tunnelling and even hot plug of Thunderbolt devices. Without SSDT modifications, you can get PCIe tunnelling with the GC-TITAN RIDGE just by connecting Thunderbolt devices in Windows and warm booting to macOS (works with older NVM firmware 23.00 but maybe not with the latest 43.00 firmware). PCIe tunnelling isn't required for DisplayPort to Thunderbolt conversion, but you still have the problem of enabling the ability to connect dual HBR3 for the XDR display. As for the Hack Pro design, I might have tried taking advantage of the fact that the Mac cheese grater tower has 4 cm of clearance underneath and placed the rear radiator at the bottom to leave room at the rear for more PCIe expansion. There are PCIe backplanes now that have PCIe 4.0 slots. One Stop Systems has 8x8 or 5x16 gen 4 backplanes although any gen 3 backplane with at least x4 slots would also be useful. The height of a PCIe backplate is 120mm - the same size as the fans on the radiator but many PCIe cards extend above the PCIe back plate so they might not fit. Are the motherboard standoffs taller than they need to be (I think 11 mm is enough for a PCIe card - I know they need room underneath for cabling - un-sleaved cables might take less space). The power supply and its cables will need adjusting - maybe a smaller reservoir (or no reservoir) would be needed. Things might get too tight. The backplane might be too large. Maybe it can be replaced by an M.2 card with PCIe switch (HighPoint or Amfeltec) to add 4 slots (or 6 with the Amfeltec) using some ADT-Link M.2 to PCIe cables. The slots could be sideways instead of vertical to give the cards more room. U.2 adapters/cables may be more flexible.

Linus neglected to mention XDR's support for Display Stream Compression (DSC). The XDR supports many connections. Thunderbolt (dual HBR3, no DSC) is only one of them. HBR2 (1440p60 12bpc, 4K60 10bpc, 5K 6bpc) HBR3 (this might not be allowed - I haven't seen evidence that single cable HBR3 works - there is a small amount of evidence that single cable is limited to HBR2) Dual HBR2 (5K60 10bpc) Thunderbolt only - Alpine Ridge, or Blackmagic eGPU before the firmware update. Dual HBR3 (6K60 10bpc - maybe 12bpc with EDID override using CRU?) Thunderbolt only. PC Thunderbolt 3 Titan Ridge driver doesn't support this because dual HBR3 exceeds 40Gbps, but 6K doesn't use full dual HBR3 and Thunderbolt doesn't transmit DisplayPort stuffing symbols so Apple tells the controller to ignore this artificial limit. HBR2 DSC (6K60 12bpc) There is no support for YCbCr and therefore no support for chroma sub sampling so DisplayPort 1.4 GPUs that don't support DSC cannot get 6K (no 6K on Nvidia Pascal or AMD Polaris for example). So the best GPUs to use for XDR in Windows are Nvidia RTX or AMD Navi. The RTX and W5700 have USB-C ports that can transmit HBR2 with DSC and USB 2.0. Since only HBR2 is required with DSC to get 6K you can probably combine USB 2.0 with DisplayPort using a Wacom Link Plus, Huawei VR 2 Computer Connection Cable, or Sunix UPD2018 (though I haven't seen anyone try that yet). As for the Boot Camp drivers - there's probably a way to get them installed on a PC with some minor work. If you don't care about having USB 2.0 (or the presets), then a bidirectional USB-C to DisplayPort cable (such as the one made by Moshi) is sufficient for non-Thunderbolt connections.

The XDR can be driven by the RTX USB-C port. You won't need the GC-TITAN RIDGE + dual DisplayPort input because the RTX and XDR support Display Stream Compression. With DSC, you can get 6K 60Hz RGB 12bpc with a single HBR2 link rate connection. If you have a GPU that doesn't have a USB-C port (but still supports DSC), then you can use a USB-C to DisplayPort bidirectional cable to get 6K but you'll be missing the USB features of the display (USB ports, brightness control, presets, rotation). You might be able to get the USB features using a Wacom Link Plus, Sunix UPD2018, or Huawei VR 2 Computer Connection Cable (they support DisplayPort 1.2 but that might be ok since you just need the HBR2 link rate with DSC to get 6K). Maybe you can connect multiple XDR displays this way. For GPUs that don't support DSC, you need the GC-TITAN RIDGE and two HBR3 connections. The GC-TITAN RIDGE doesn't support dual HBR3 though because two full HBR3 connections would exceed the bandwidth of Thunderbolt 3. However 6K does not require the full bandwidth of HBR3 and Thunderbolt 3 doesn't transmit DisplayPort stuffing symbols, so Apple's Titan Ridge Thunderbolt 3 controllers are set to allow dual HBR3. I don't think the Windows Thunderbolt 3 drivers will enable that (or I haven't seen anyone succeed yet). You might want a GC-TITAN RIDGE anyway if you have a GPU that supports DSC so you can get USB 3.0 speed from the USB ports of the display (and also to connect other Thunderbolt devices to the GC-TITAN RIDGE's other Thunderbolt 3 port). Some AMD cards also have a USB-C port (W5700 for example). And Navi cards (5xxx) support DSC.

I've received AGDCDiagnose info from someone with a Blackmagic eGPU (contains Radeon Pro 580 which does not support the DSC mode of the XDR display). It shows dual HBR3 connection from the Radeon Pro 580 to the Apple Pro Display XDR (this can be done only through Thunderbolt 3 since the Blackmagic eGPU has no other method to connect to the display). The EDID vendor/product is APPAE2E. This is the first I've seen having the ae2e product ID - it is also the largest because it has both the tile info/timings (tiled 5K and 6K) and the 4K, 5K, and 6K non tiled timings (7 EDID blocks total).

The manual says you need Dual Link DVI for max resolution/refresh rate 1920x1080@144Hz. Since your RTX card has a USB-C port, you could use a USB-C to Dual Link DVI adapter (if you're not using the USB-C port for a USB-C display or hard drive or dock). I like Club 3D stuff because they have a user forum and have VESA certified cables for HBR3 speed. https://www.club-3d.com/en/detail/2471/usb_type_c_to_dvi_i_dual_link_active_adapter/ The adapter doesn't have a separate USB type A power connection because it can get USB power from the USB-C port. If you want to save your USB-C port for something else, then they have a DisplayPort version. Remember, any DisplayPort to Dual Link DVI adapter should have a separate USB type A connection for extra power because the conversion process requires more power than DisplayPort can provide by itself - otherwise the adapter would probably be single link.

I believe so. I'm waiting for someone to post results of the following commands to prove it. /System/Library/Extensions/AppleGraphicsControl.kext/Contents/MacOS/AGDCDiagnose -a > AGDCDiagnose_a.txt 2>&1 /System/Library/Extensions/AppleGraphicsControl.kext/Contents/MacOS/DisplayDiagnose -a > DisplayDiagnose_a.txt 2>&1 The Apple Pro Display XDR Tech Specs do not mention any reduced visual capability for Macs or MPX modules that do not support DSC (if they at least have AMD graphics and Titan Ridge Thunderbolt controllers). Google: xdr site:support.apple.com/en-us The only reduced capability is the bandwidth left remaining for USB. No-one has made any complaints about reduced capability (except when they do something wrong like try to connect two displays to the same Thunderbolt controller).

It's a DIY thing. Put it in a box. Attach a power supply. And you got an adapter! Then attach DisplayPort cables and LG UltraFine 5K. It's kind of expensive though, like the HDMI 2.0 to DisplayPort adapters. I think the Wacom Link Plus is at least worth looking at (the HDMI input did not work well for me but DisplayPort works). It's limited to DisplayPort 1.2 (4 lanes HBR2) and USB 2.0 (the input is USB Micro B). There's a similar adapter from BizLink which supports DisplayPort 1.4 and USB 3.1 gen 2 and VirtualLink but I don't think it's a real product which is sad. Does anything use VirtualLink? Even if VirtualLink alt mode is useless, the adapter also supports DisplayPort alt mode.

I forgot one item missing from the original post: For DisplayPort or USB-C to VGA adapters, there do exist adapters that can do up to 340 MHz. I have a Plugable usbc-vga adapter which can convert DisplayPort 1.2 (2 lanes of HBR2) to VGA (up to 330 MHz). I don't know why they only advertise 1920×1200@60Hz. For some reason, macOS limits the adapter to 160 MHz. Maybe Apple meant to limit some specific adapter but managed to limit all of them (I made a patch to remove that limit for macOS Intel graphics drivers). Many of the adapters are discussed at https://hardforum.com/threads/24-widescreen-crt-fw900-from-ebay-arrived-comments.952788/page-426

The simplest method is to use the DVI input of your monitor: Use a passive HDMI to DVI cable or adapter. DVI uses the same signals as HDMI. If you want to use the DisplayPort input of your monitor then: There exist HDMI 1.4 to DisplayPort active adapters (1440p60, 4K30). Your monitor is only 1080p so HDMI 1.4 is good enough. This info is in the original post. There also exists really expensive HDMI 2.0 to DisplayPort active adapters (4K60). This info is missing from the "HDMI Source to DisplayPort Display" section if the original post (adapters from SIIG and gofanco https://insights.club-3d.com/thread/hdmi-2-0-to-displayport-1-2-2/6/ ) . DisplayPort to HDMI adapters also exist. This is mentioned in the original post. I don't know why you say "displayport adapter in first page one way its possible and another way its not" since the original post doe mention that adapters exist for both directions. The rest of this post is for other things (unrelated to your problem) that are missing from the original post: If you have a dual link DVI monitor (your monitor is only 1080p so it is only single link): You can chain a DisplayPort to Dual Link DVI adapter to a HDMI 1.4 to DisplayPort adapter to convert single link HDMI (340 MHz) to dual link DVI (330 MHz). This adapter chaining method may be unreliable - some OSs (Ubuntu) may handle it better than others (macOS) because of some plug detect issue? I think the "HDMI Source to DVI Display" section of the original post should be updated with this info. If you have a dual link output only (old graphics card) and a DisplayPort monitor: There exists the Gefen Dual Link DVI to Mini DP Converter. This is missing from the original post. For DisplayPort to DisplayPort: An MST Hub can convert different DisplayPort lane counts (1,2,4) and link rates (RBR, HBR, HBR2, HBR3) from source to sink (much like a PCIe switch can convert PCIe 3.0 x4 to PCIe 2.0 or 1.0 x16). There now exists DisplayPort 1.4 MST hubs (Club 3D, Delock, and soon StarTech) that can accept Display Stream Compression input and transmit that to output or decompress it for non DSC supporting displays. https://insights.club-3d.com/thread/displayport-1-4-mst-hub/ MST hubs can also support multiple displays (but macOS can only use one display from a hub - the rest are mirrored). The original post doesn't mention MST. Other things not mentioned: The DisplayPort 1.2 standard can be found on the internet (but isn't supposed to be?) The DSC spec is freely available from the VESA free standard website. It contains code that you can use to see how visually lossless the algorithm is. Other things to update in the original post: "Thunderbolt 3 Source to DisplayPort Display": Missing mention of DisplayPort 1.3, DisplayPort 1.4 support since Titan Ridge. "DisplayPort Source to Thunderbolt 3 Display": A Thunderbolt 3 add-in (such as GC-TITAN RIDGE or GC-ALPINE RIDGE) card can be used to convert DisplayPort to Thunderbolt 3 if you want to connect a Thunderbolt display. If the add-in card is not connected by PCIe to your computer, then you won't be able to use the extra functionality of the display that requires PCIe tunnelling (brightness control, camera, USB audio, FireWire, Ethernet, etc.) For some unknown reason, The Apple Thunderbolt Display (Thunderbolt 1) doesn't work from a GC-TITAN RIDGE (but does work from any Titan Ridge of a Mac). "DisplayPort Source to USB Type-C DisplayPort Alternate Mode Display": this can be done with the above Thunderbolt add-in card method. There also exists PCIe cards from Delock and SUNIX. And there are external adapters such as the Wacom Link Plus (but not the Wacom Link). https://hardforum.com/threads/use-usb-c-monitor-without-usb-c.1911817/page-2 Component Video ? Maybe this goes under VGA since they are both analog. Or it doesn't matter since it's old. Also in the VGA category could be the old Apple 15 pin connector which is mostly just VGA but in a wider two row D-SUB connector. Before EDID, Apple used a 3 pin sense line scheme to differentiate different connected displays. The pins were connected together (or not) with wires or diodes and grounded or not for 512 different possible circuits producing 45 different results though maybe only 17 were ever used (one of them being "no-connect"). There were old Mac to VGA display adapters with dip switches to select one of those results to make the old Mac think the VGA display was, for example, an Apple 2 Page Mono display, or RGB 19 inch display, etc. Many of these displays only supported a single refresh rate (they were not multi-sync).

Us mere mortals don't get access to the real Thunderbolt documentation. The USB4 spec is publicly available and may describe most of how Thunderbolt works. Section 10 is all about DisplayPort tunnelling. Section 10.4.2.1 is about Path Configuration: Setup. The next page (pg 362) describes how the Connection Manager calculates bandwidth available to limit the maximal link rate and lane count of the DP IN adapter's DP_REMOTE_CAP register. I guess Apple changed this calculation to allow two HBR3 four lane connections. I need an AGDCDiagnose output to prove that (but I believe there's no other way to get 6K without DSC). I wonder if two HBR3 connections is allowed only for the XDR display, or if it will also work with my two displays (It's actually one Acer XV273K with two DisplayPort 1.4 inputs). If it does work with the Acer to allow two HBR3 connections, I wonder what happens if I increase the refresh rates so that the bandwidth exceeds the 40 Gbps limit (does it gracefully disconnect, stop, or show garbage). I don't have a Titan Ridge Mac with AMD dGPU to test. A correction to what I said previously about connecting two 4K displays to Thunderbolt 2 (method: connect HBR display first, then first 4K, then replace HBR display with second 4K). The AGDCDiagnose outputs show that they both connect with DisplayPort 1.1 HBR four lanes. To get one to be connected with DisplayPort 1.2 HBR2 four lanes, they both need to be disconnected first. This behavior matches what is described by the USB4 spec (I'm not sure why the DisplayPort version needs to change - just the link width and lane count should be sufficient). The limit calculation described by USB4 has a problem. If I have 4 lanes of HBR available, will it allow 2 lanes of HBR2? I did this test with a two lane DisplayPort 1.2 adapter after setting up a single HBR x4 connection to Thunderbolt 2. The answer is that it connects with 2 lanes of HBR instead of HBR2. This is not optimal. Apple is right to allow max link rate and lanes (exceeding 40 Gbps) instead of that silly calculation. The problem (that Apple bypasses for the XDR?) is that a limit is placed on both lanes and link rate, but bandwidth is the product of both, where HBR x4 is the same as HBR2 x2. Another but: for Thunderbolt, bandwidth is about pixels per second and not related to DisplayPort lanes and link rate (stuffing symbols are removed - see Figure 10-22 in the USB4 spec). 6K doesn't require full dual HBR3 but does require more than dual HBR2 so Thunderbolt 3 Titan Ridge should be able to handle 6K as Apple has made it do. Ideally, a user should be able to lower the bandwidth used by a display (decrease refresh rate or resolution or pixel depth) to increase the bandwidth allowed for another display connected to the same Thunderbolt 3 port without disconnecting/reconnecting the displays. Another situation where the limit calculation has a problem: If I have a Thunderbolt 3 dock with an internal DisplayPort 1.4 MST hub (like the HP Thunderbolt Dock G2) connected to a Thunderbolt 2 port, then any display (even if it is HBR only) connected to the hub will use the full DisplayPort 1.2 HBR2 four lane connection of the hub so a second display cannot be connected to the Thunderbolt port of the HP Thunderbolt Dock G2. Of course, an MST hub allows multiple displays to be connected, but that doesn't work in macOS which doesn't support MST for multiple displays (but does support MST for old 4K displays that use a separate stream for each side of the display).

I don't think there's anything to fix with Titan Ridge. Somehow Apple makes it work to support 6K with GPU's that don't support DSC (read my post linked below). The capability required a firmware update for the Blackmagic eGPU's Thunderbolt controller (but I need an AGDCDiagnose output to verify that dual HBR3 is working from the BlackMagic eGPU and Macs - I only have info for dual HBR2 and HBR2 DSC modes). My test with two HBR3 displays with a GC-TITAN RIDGE shows one display can connect at HBR3 speed and the other can connect at HBR speed, which is the same bandwidth as dual HBR2 (34.56 Gbit/s). With Thunderbolt 2, I explained how to connect two HBR2 connections. Thunderbolt does not transmit DisplayPort stuffing symbols which reduces the actual amount of DisplayPort data transmitted. This allows the remaining bandwidth to be used by PCIe packets. Stuffing symbols are recreated when a Thunderbolt controller converts the Thunderbolt DisplayPort packets back to DisplayPort. I think with USB4 you're going to see more complicated chains (and trees!) with more than two DisplayPort endpoints where this bandwidth issue will need to be addressed or it will just get worse. This wasn't a problem with MST, because the OS (not macOS) knows the MST topology and can know exactly how much bandwidth is available taking into consideration the bandwidth currently being used by connected displays based on their timing and pixel format. Maybe the OS needs to treat Thunderbolt more like MST if that's possible. Otherwise, the only way to fix this is for the OS to allow the user to select how much to reserve for the next connected display (if the OS is able to convey this info to the Thunderbolt controller with the free DisplayPort input).

Can you link the Snazzy Labs video that shows the display working at 6K on a Surface Laptop 3? DisplayLink is a USB to DisplayPort converter. It doesn't use the GPU. Really, DisplayLink has nothing to do with the XDR or the GPU. Does DisplayLink support 5K dual cable displays like the Dell UP2715K? Maybe. Does DisplayLink support HBR3 or Display Stream Compression (DSC)? I don't think so, therefore it can't support 6K. Read more about RTX, VirtualLink, USB-C, XDR at:

The video should be updated with more info/testing with different PCs/Macs using different connection modes. The XDR has a Titan Ridge Thunderbolt 3 controller which can accept Thunderbolt or USB-C with DisplayPort alt mode input (Thunderbolt is also an alt mode of USB-C). The XDR supports the following input modes: 1440p HBR (good for booting) 4K 10bpc HBR2 5K 8bpc HBR3 (not sure about this one - needs testing with a DisplayPort 1.4 GPU that doesn't support DSC) 5K 10bpc HBR2x2 Thunderbolt 3 (from Alpine Ridge) 6K 12bpc HBR2 DSC 6K 12bpc? HBR3x2 Thunderbolt 3 (from Titan Ridge) The XDR display is 10bpc but allows 12bpc input. The 4K, 5K, and 6K modes have timings for the following refresh rates (Hz): 47.95, 48.00, 50.00, 59.94, 60.00. I don't know if custom timings are supported. The 1440p mode is only 60Hz. The timings for the lower refresh rates are strange because only the vertical blanking is changed. The pixels are drawn at the same rate for each refresh rate (so the same bandwidth is required for all refresh rates). This means each frame is drawn in the same amount of time but lower refresh rates will show the frame for longer. Bandwidth is reduced for lower frame sizes though (so 5K and 4K don't use the same bandwidth as 6K). For USB functionality, (brightness control, USB ports, presets): A USB-C with DisplayPort alt mode connection includes USB 2.0 (4 lanes of DisplayPort leaves no lines available for USB 3.x). A Thunderbolt connection uses PCIe tunnelling to the Titan Ridge USB 3.1 gen 2 controller of the XDR display's Titan Ridge controller. But HBR3x2 reduces bandwidth available to USB 2.0 speeds. The USB ports and devices are connected internally with a USB 3.0 hub. How is the USB speed limited? Is the hub connection reduced to USB 2.0 mode or is it only limited by the reduced available PCIe bandwidth? For example, a USB 3.1 gen 2 (10 Gbps) PCIe card can work fine (but with reduced bandwidth) in a PCIe 1.0 x1 slot (2 Gbps). Nvidia RTX, AMD 5300M, 5500M, W5700, W5700X, 5700 XT support DSC. Until recently Intel graphics have been limited to DisplayPort 1.2. Now there are Intel CPUs (10th gen CPU, Gen 11 GPU, Ice Lake) that support DisplayPort 1.4 and DSC, so now the Surface Laptop 3 can support 6K. RTX has a USB-C port that supports DisplayPort alt mode so it works with the XDR display. The USB-C port of the RTX also supports VirtualLink alt mode. Do any available VR headsets support VirtualLink? Anyway, the XDR does not use VirtualLink. The W5700 also has a USB-C port but I don't know how it works for macOS/Windows, or if it supports USB with DisplayPort alt mode. For GPUs that don't have USB-C, a bidirectional USB-C to DisplayPort cable (such as the one sold by Moshi) will work but you'll be missing the USB features of the display. Maybe an app can set brightness using DDC/CI? If the GPU doesn't support DSC, then you'll be limited to 5K or 4K unless the GPU has two DisplayPort HBR3 connections to a Titan Ridge Thunderbolt 3 controller. However, the firmware of the controller might limit one or both connections to HBR2 or HBR which makes 6K impossible (this requires more testing). There was a problem with the Blackmagic eGPU where it would only allow two HBR2 connections to the XDR display (supporting only 5K). They released a firmware update to fix that. With a GC-TITAN RIDGE, when I connect two HBR3 displays, one of them can only connect at HBR speed. In macOS, the AGDCDiagnose command is used to get connection information (DisplayPort lanes, speed, DSC, HDCP, MST, etc.). Even if you get dual HBR3 over Thunderbolt 3 to work, I haven't seen an EDID that includes the 3008x3384 timing that would be required to support the tiled mode but that might be just because I haven't seen the AGDCDiagnose output from this connection mode yet. 6K is a lot of pixels (more than dual 4K which is more than 5K). A card that supports DSC and has 6 DisplayPort outputs (such as the W5700) might not support six 6K displays. Apple's Mac Pro tech specs says the AMD Radeon Pro Vega II can only support two XDR displays instead of the expected 3 (only 3 because it doesn't support DSC). The Mac Pro tech specs and the support document for the Radeon Pro Vega II Duo MPX Module have conflicting number of supported XDR displays. Is there a problem with 6K transport from the MPX slot to the I/O card or top Thunderbolt 3 ports? I don't think so, since that is the only way for the MPX 580X to support 6K.

Thunderbolt 2 supports dual DisplayPort 1.2, so shouldn't the max bandwidth be 20 Gbps instead of 17.28 Gbps, like you did for Titan Ridge with dual DisplayPort 1.4? You already have a DisplayPort 1.2 option so there isn't much reason to keep Thunderbolt 2 the same (unless the extra 2.72 Gbps cannot be used). With Titan Ridge, the extra 5.44 Gbps allows the Apple Pro Display XDR to run at 6K 60Hz 10bpc RGB using a dual link SST DisplayPort 1.4 HBR3 connections without DSC. Computers that support DSC (macOS/Windows: 5300M, 5500M, W5700X, maybe 5700XT?; Windows: RTX) use a single link SST DisplayPort 1.4 HBR2 connection with DSC at up to 12 bpc (though the display is 10 bpc). I admit that with Thunderbolt 2, there's not much you can do with an extra 2.72 Gbps. A 5K dual link SST display could increase refresh from 46Hz to 53Hz (CVT-RB 8bpc RGB). But first you would need to find a display that can support such a timing (I think the LG UltraFine 5K and Dell UP2715K only support 60Hz?) and an OS that will allow that timing twice on a single Thunderbolt 2 connection. In macOS, if you connect a 4K display then there is no bandwidth remaining for another display. To connect two 4K60 displays to Thunderbolt 2, you first have to connect a low bandwidth display (such as a 1440p display or a DisplayPort to VGA adapter or HDMI 1.4 adapter). With the 1440p display, some bandwidth remains to connect a second display. Connect the first 4K display. Then disconnect the 1440p display. The bandwidth reserved for the 1440p display is freed so it can be used by another display (it is not reassigned to the already connected 4K display). Finally, connect the second 4K display. I can only get the two 4K displays to work at 1440p73 over Thunderbolt 2. This is not sufficient to prove that more than 17.28 Gbps can be used. Maybe it's a limitation of the macOS driver. Lowering the refresh of one display does not allow the refresh of the second display to be increased because the bandwidth is assigned when the displays are connected. Maybe behavior in Windows is different for Thunderbolt 2 connected displays.