Announcement

Collapse
No announcement yet.

VESA Releases DockPort Standard To Compete With Thunderbolt

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • VESA Releases DockPort Standard To Compete With Thunderbolt

    Phoronix: VESA Releases DockPort Standard To Compete With Thunderbolt

    VESA has released the DockPort standard today from Computex that's an extension of the DisplayPort interface and allows for USB data and power charging capabilities as a royalty-free, industry standard...

    http://www.phoronix.com/vr.php?view=MTcwODM

  • #2
    Thunderbolt on Linux market is a...

    Originally posted by phoronix View Post
    Phoronix: VESA Releases DockPort Standard To Compete With Thunderbolt

    VESA has released the DockPort standard today from Computex that's an extension of the DisplayPort interface and allows for USB data and power charging capabilities as a royalty-free, industry standard...

    http://www.phoronix.com/vr.php?view=MTcwODM
    Niche in a niche market.
    Makes no economical sence whatsoever.
    How many Linux users are there? Unfortunately not so much. How many can afford Mac? Not too many. How many of them actually have Mac? Even less. How many would switch to Mac with Linux on board? Virtually none.

    So yeas, hope more open standard will take a place in OSS world.

    Comment


    • #3
      My understanding was that Thunderbolt is supported nicely, and only Apple refusal to support that standard on the host side (moving responsibilities from frimware to OS drivers), made it problematic for Linux support efforts.

      Comment


      • #4
        Originally posted by przemoli View Post
        My understanding was that Thunderbolt is supported nicely, and only Apple refusal to support that standard on the host side (moving responsibilities from frimware to OS drivers), made it problematic for Linux support efforts.
        That's correct. You know you've fucked up when even one of the top kernel maintainers can't get it working.

        Comment


        • #5
          Originally posted by dimko View Post
          Niche in a niche market.
          Makes no economical sence whatsoever.
          How many Linux users are there? Unfortunately not so much. How many can afford Mac? Not too many. How many of them actually have Mac? Even less. How many would switch to Mac with Linux on board? Virtually none.

          So yeas, hope more open standard will take a place in OSS world.
          I think that was not the main intent of VESA. This is more aimed at the hardware manufacturers to unify support. I hope intel will open-source their implementation or it can just as well turnout like nvidia G-sync.

          Comment


          • #6
            let me transalte that

            Originally posted by przemoli View Post
            My understanding was that Thunderbolt is supported nicely, and only Apple refusal to support that standard on the host side (moving responsibilities from frimware to OS drivers), made it problematic for Linux support efforts.
            to geek language.

            Apple decided to act assholes on that. As usual. I am really uncomfortable with LLVM, it's because of Apple.

            Devs that develop LLVM should really support GCC. Apple is known to be assholes when it comes well, to think of it, in pretty much everything.

            Comment


            • #7
              Originally posted by dimko View Post
              to geek language.

              Apple decided to act assholes on that. As usual. I am really uncomfortable with LLVM, it's because of Apple.

              Devs that develop LLVM should really support GCC. Apple is known to be assholes when it comes well, to think of it, in pretty much everything.
              Like any other non-opensource big corporation, which is everyone.

              Comment


              • #8
                I'm little disappointed that the non-video data link is USB. I guess that the video & USB data are multiplexed on the same lanes. Why couldn't they have just chosen PCIe..? (and create royalty free Thunderbolt alternative.)

                Comment


                • #9
                  Originally posted by Grawp View Post
                  Why couldn't they have just chosen PCIe..? (and create royalty free Thunderbolt alternative.)
                  Because USB already works over passive cables. Extending PCIe would likely require active cables like Thunderbolt has, thus eliminating most of DockPort's cost advantage.

                  Originally posted by phoronix
                  hopefully the Linux support for DockPort will get lined up quickly and be in better shape than Linux's Thunderbolt support.
                  DockPort switches the PC's existing USB and DisplayPort controllers, there is no extra driver or software needed. Unlike Thunderbolt which requires complex logic in the firmware or drivers to handle hotplug.

                  Comment


                  • #10
                    Originally posted by Grawp View Post
                    I'm little disappointed that the non-video data link is USB. I guess that the video & USB data are multiplexed on the same lanes. Why couldn't they have just chosen PCIe..? (and create royalty free Thunderbolt alternative.)
                    Because exposing PCIe lanes to the outside world and giving everything from a random flashdrive to mice DMA access, where they can fuck over the hardware itself much much more easily is a bad thing and further for everything except for storage there's no benefit to not using USB instead.

                    Comment


                    • #11
                      Originally posted by Luke_Wolf View Post
                      Because exposing PCIe lanes to the outside world and giving everything from a random flashdrive to mice DMA access, where they can fuck over the hardware itself much much more easily is a bad thing and further for everything except for storage there's no benefit to not using USB instead.
                      PCIE has a lot of technical advantages over USB. It is a point to point interconnect, so its stream based rather than packet based. You get reduced latency that way. Also, it is a parallel vs serial bus, so of course the parallel one will have more bandwidth.

                      I would not want to try to run a GPU over USB3, for example, but over PCIE x4 over a Thunderbolt cable it is a much more reasonable proposition.

                      I see dockport being useful as an all in one.. um.. dock port. So you can have one displayport conncetion to a monitor that has a USB hub and ethernet switch on it.

                      Comment


                      • #12
                        Originally posted by zanny View Post
                        PCIE has a lot of technical advantages over USB. It is a point to point interconnect, so its stream based rather than packet based. You get reduced latency that way. Also, it is a parallel vs serial bus, so of course the parallel one will have more bandwidth.

                        I would not want to try to run a GPU over USB3, for example, but over PCIE x4 over a Thunderbolt cable it is a much more reasonable proposition.

                        I see dockport being useful as an all in one.. um.. dock port. So you can have one displayport conncetion to a monitor that has a USB hub and ethernet switch on it.
                        Reduced Latency, except that latency really isn't a problem for pretty much anything that uses USB other than arguably storage. Also you clearly don't care that much about latency or bandwidth if you're focusing on the idea of using the monitor as a USB hub and ethernet switch.

                        Running an external GPU is and has always been a stupid idea. Why? Well exactly because of latency. Going across a PCIe lane on the motherboard is bad enough, having to traverse most of the motherboard to get to an external port, then going across a cable to another port and finally to the GPU is going to cause excessive latency, plus unless the cable is appropriately shielded you're going to get all sorts of fun interference right where you don't need it.

                        All of this however does not change the fact that using Thunderbolt means trusting random devices with Direct Memory Access which is really not a smart thing to do. The harm far outweighs the benefits.

                        Comment


                        • #13
                          Originally posted by Luke_Wolf View Post
                          Running an external GPU is and has always been a stupid idea. Why? Well exactly because of latency. Going across a PCIe lane on the motherboard is bad enough, having to traverse most of the motherboard to get to an external port, then going across a cable to another port and finally to the GPU is going to cause excessive latency, plus unless the cable is appropriately shielded you're going to get all sorts of fun interference right where you don't need it.
                          You know, Thunderbolt isn't just simply an "riser" (i.e.: an "extension cord" at the electrical signal level of PCIe). They didn't add any new contacts inside the connector, no any additional conductor inside the cable.
                          Also remember that we're speaking about DisplayPort. Which is a protocol already used in production to push image raw-data, i.e.: uncompressed pixels at a high frame rate, and does this corruption-free. This it is already very good a pushing digital data packets reliably and corruption-free (otherwise you'd have garbage on the screen), at a high-bandwidth (it needs to push enough data for every single pixel at a reasonable frame rate. Double it if you need 3D Stereo. Therefore standards like DisplayPort 1.3 have 32Gbits/s throughput) and low latency (it drives an *actual* display. It would be a shame if visual feed-back lagged a lot behind your actions).
                          Compared to the requirement of a *display* interconnect, PCIe is dwarfed by comparison (PCIe 3.0 is only around 1Gbit/s per lane)

                          [ Note: Again I'm completely ignoring the actual electrical limitations, as PCIe isn't carried on extra conductors in Thunderbolt. Also, speaking of electrical limitations, note that both of them, DisplayPort and PCIe, use bunch of several serial link next to each other, instead of a single wide parallel link = same number of wire (so multi-serial and parallel have roughly the same speed) but a lot less sensitive to electrical distance (each serial line is a separate independent channel, so less problems of signal not arriving at the same time). ]

                          The idea behind thunderbolt is, as we already have this ultra-high performance display interconnect, why not using it to carry also other type of data like PCIe?
                          And thus, there's an extra chip that does multiplex PCIe data among the data packets sent down the display-port cables.
                          No extra wire, just add a bit more data on the logical level, while sending it on a connection that nevertheless is built to carry extreme high-bandwidth data.
                          (Currently, Thunderbird 2 is PCIe 4x over a 20GBit/s DisplayPort 2.0)

                          Thus connecting a GPU through Thunderbold does make sense: there's plenty of bandwidth to use (and uploading textures and commands is certainly *not* going to use more bandwidth that sesnding whole uncompressed frames), latency isn't that much catastrophic (it's already capable of pushing images to the screen without excessive lag) so GPU could be used within a reasonable limit, and there are several use-cases where it would really make sense (have an [external] GPU when the laptop is docked, have only a low-power embed GPU while on the move. The same kind of advantage that things like Optimus enable, except that you're avoiding the weight of the extra GPU in ultra-light notebooks like Mac Air)

                          Originally posted by Luke_Wolf View Post
                          All of this however does not change the fact that using Thunderbolt means trusting random devices with Direct Memory Access which is really not a smart thing to do. The harm far outweighs the benefits.
                          This is already a (partly) solved problems.
                          - PCIe is used as interconnect in some clusters (PLX, IDT, Dolphin, etc.) which all use *bridges chip*, which can also isolate what one CPU can see of the PCIe tree from a different device.
                          - IOMMU enables to do memory protection also against peripherals on a bus. AMD's IOMMU is available on all current CPU/Motherboard combos. (Though Intel's is only available on high-end CPUs). This can severly limit exploitability by having peripherals only accessing the subset of address they are confined to (and thus a rogue thunderbolt hardware, won't be allowed to access encrpytion key stored in RAM)

                          Comment


                          • #14
                            Originally posted by DrYak View Post
                            *snip*
                            The problem is that you want the interconnect between dependent compute units to be as fast and as low latency as possible, and IMO you're really pushing it to do something as latency dependent as GPUs off board, to then hook up to a monitor and expect things to turn out nicely. AFAIK all of the demo ones that have been played with have had extreme lag issues, which is part of why they never really went to market with them.

                            Originally posted by DrYak View Post
                            This is already a (partly) solved problems.
                            - PCIe is used as interconnect in some clusters (PLX, IDT, Dolphin, etc.) which all use *bridges chip*, which can also isolate what one CPU can see of the PCIe tree from a different device.
                            - IOMMU enables to do memory protection also against peripherals on a bus. AMD's IOMMU is available on all current CPU/Motherboard combos. (Though Intel's is only available on high-end CPUs). This can severly limit exploitability by having peripherals only accessing the subset of address they are confined to (and thus a rogue thunderbolt hardware, won't be allowed to access encrpytion key stored in RAM)
                            Weren't they still having issues with their IOMMU implementation? Also AFAIR you still have to manually turn it on in the kernel on Linux which means most people aren't going to do it, I'm not sure how that all works on Windows, but if it's not turned on by default people aren't going to use it, and if people aren't using it then that means that DMA is still something you should never allow outside of a small set of known trusted components.

                            Comment


                            • #15
                              Well, this totally makes sense, since DisplayPort already had a data line.
                              I am O.K. with an USB data line since it's cheap, transmitted over passive lines, extremely common and with a large bandwidth.

                              Hopefully, a CEC-like standard will be released for it. Meanwhile, we could just use HID commands, but CEC would just be better.
                              I already see multiple advantages : a USB soundcard integrated into the monitor, for instance, or a webcam, or transmitting IR commands the monitor receives (like CEC), but I think most of those were present with thunderbolt.
                              USB is just sufficient (for now) to handle all this, and both USB and DisplayPort standards are extensible.

                              I may finally wait a little more before changing my laptop/desktop :P

                              Comment

                              Working...
                              X