The USB Promoter Group is hard at work developing the USB4 specification. We met with them at Computex this year, and the good news is that the spec is in its 0.96 version and things are proceeding quickly. The group believes that retail products featuring USB4 will be available by the end of 2020.

Update 16/6: The current USB4 spec is at 0.96.

Being based on Thunderbolt 3 technology and offering up to 40 Gbps bandwidth, USB4 promises to be more than that. In fact, so much more that the USB Promoter Group is considering a new logotype and branding scheme. The current one is already complex enough, so expect some kind of simplification on that front. Meanwhile, USB4 will be backwards compatible with existing USB Type-C devices.

When it comes to availability, USB-IF seems to be optimistic that the specification will be finalized this Summer and actual USB4-supporting devices will be available by the end of 2020. Since Intel knows how to build Thunderbolt 3 controllers, it will certainly use its expertise developing USB4 controllers eventually.

Related Reading

Comments Locked

69 Comments

View All Comments

  • TheUnhandledException - Friday, June 14, 2019 - link

    With 10 bit input you can get up to 3.75:1 compression using DSC.
  • repoman27 - Friday, June 14, 2019 - link

    Goes and reads the specification...

    Hmmm... 10 bpc may yield a 3.75:1 compression ratio, but AFAICT the decoder is effectively limited to 3 pixels per unit of pixel time. So with VBR you might be able to get some extra down-time and save a little power, but you can't transport more than 3x the number of uncompressed pixels. Having only skimmed through the spec once though, any additional insight into how this works in practice would be appreciated.
  • repoman27 - Wednesday, June 12, 2019 - link

    Thunderbolt controllers have connections for up to 4 lanes of PCIe to connect to the host or devices, but the signaling used for the Thunderbolt ports on the other side is a complete different thing, even though you can transport PCIe or DisplayPort packets over a Thunderbolt link. Updating the PCIe interface on the controller from PCIe 3.0 to PCIe 4.0 is completely straightforward, and would be quite welcome, seeing as a dual-port controller currently has two 40 Gbit/s ports with only ~31.5 Gbit/s of PCIe bandwidth available on the back-end between them. Historically, Intel has only bumped up the PCIe or DisplayPort revisions once their platforms actually supported those technologies. Which makes sense, seeing as it's their thing after all.

    Thunderbolt controllers also have connections for up to two DisplayPort main links, which is how the DisplayPort packets get on the bus; it has nothing to do with the PCIe side of things. With Titan Ridge Thunderbolt 3 controllers, those are at revision 1.4 with HBR3 support, providing 25.92 Gbit/s apiece, or up to 40 Gbit/s per Thunderbolt link. Updating to DP 1.4a with Display Stream Compression is also relatively straightforward, and even a 2:1 compression ratio would allow you to saturate both links of a dual-port controller.

    The logical next step for Thunderbolt 4 is to double bandwidth again by switching from NRZ to PAM4 signaling, which would allow up to 80 Gbit/s per link. However, feeding that on the back end without making the chip enormous and radically more expensive will require waiting for platforms that support both PCIe 5.0 and next generation DisplayPort.

    Without DSC or chroma subsampling, a 16:9 6K display at 10 bpc and 120 Hz requires between 72.11 and 81.96 Gbit/s (depending on how you define 6K). 8K at 10 bpc and 120 Hz is between 127.75 and 145.26 Gbit/s. Thunderbolt 4 might be able to handle some 6K resolutions over a single cable without DSC, but as it stands, you'd need at least three DisplayPort HBR3 main links on the back end.

    Incidentally, HDMI 2.1 Fixed Rate Link, which is not available on any shipping products yet, is only 42.67 Gbit/s, so not even close to being able to do 6K or 8K with 10 bpc at 120 Hz without DSC or chroma subsampling.
  • repoman27 - Wednesday, June 12, 2019 - link

    Ugh, ignore this bit, "and even a 2:1 compression ratio would allow you to saturate both links of a dual-port controller." Obviously the total DisplayPort bandwidth remains at 25.92 Gbit/s per main link, but compression allows you to squeeze more pixel data in there.
  • Dolda2000 - Wednesday, June 12, 2019 - link

    I cannot help but wonder if all this PCI-express hotpluggability is well considered. Hotpluggable DMA attacks were already a problem back with Firewire.

    Sure, in theory we do have IOMMUs these days, but I'm not even sure how many implementations have an IOMMU group for each Thunderbolt port, nor how well developed driver support is for applying IOMMU protections to devices. OS support for Thunderbolt access control also seems to very rudimentary at best.
  • timecop1818 - Wednesday, June 12, 2019 - link

    Ah, you're the type of shithead that cries wolf each time a new fuckclown-like "exploit" is announced and spews garbage on forums that we're all fucking doomed huh.

    I believe I speak for every normal desktop PC user when i say that none of that shit fucking matters whatsoever. I disable all these retarded "mitigations" as soon as disable mechanism is available and will continue to do so. None of this shit affects normal users and never will. Maybe you lunix dweebs worry about "hot plug DMA attacks" but to us normal users, this shit is just another cable to plug in and enjoy a device that works.
  • damianrobertjones - Wednesday, June 12, 2019 - link

    ?? Steady with the language. It adds nothing to what you're trying to say and makes you seem childish.
  • s-plus - Wednesday, June 12, 2019 - link

    What's wrong with you man?
  • AshlayW - Wednesday, June 12, 2019 - link

    Wow, rude.
  • id4andrei - Wednesday, June 12, 2019 - link

    Normal desktop PC users don't use and don't care about TB. They care more about USB-A than TB. Only Apple users and the blogosphere cares about the TB dogma.

Log in

Don't have an account? Sign up now