Last week SemiAccurate told you a bit about SuperMHL, now it is time for more detail and two new chips. Both chips are additions to the two announced at MWC and the details are a bit on the technical side, just how we like them.
If you haven’t already, go back and read our earlier article here, we won’t repeat the things we wrote up earlier. That said one of the big questions was how many USB-C high-speed lanes does SuperMHL take to transmit 4K60 video, or more to the point can it do the quoted 8K120 over USB-C? The answer to those questions are one and no respectively but the reasoning is where it gets interesting.
The pinouts for USB-C that allow Alt Modes
As you know the USB-C connector has two pairs of high-speed bi-directional pins to use for high-speed data, that is USB3/3.1 up to 10Gbps per pair of pins. USB at the moment only defines one high-speed link for USB3/3.1 but there is no reason why you couldn’t do another if you wanted. The mess that is alt modes pulls one of the two pair and uses it for other purposes or both high-speed pairs while dropping data speeds to USB2 speeds.
The answer to the original question on SuperMHL is that they can do 4K60 4:4:4 with only one pair, IE transferring 10Gbps of USB3.1 at the same time. If you use two pairs you drop to USB2 data rates but the cable will support 4K120 4:4:4, not a bad tradeoff for this humble little connector. USB-PD is unaffected either way, 100W for the taking if your cable supports it. If you want 8K60 4:4:4 you need four pairs while the full 8K120 4:2:0 takes six pair.
But wait you say, USB-C only supports two high-speed lanes, how can you get four lanes to do the 8K60 from said humble cable? Well that is because the USB-C connector actually has four high-speed lanes on it. “Gasp!” you gasp, what conspiracy is there that hides these pins and cables from us, how did they do such a dastardly thing?
Easy, bi-directional needs a transmit and receive, two bi-di’s need two transmits and two receives. Data is two-way, video is only one. For SuperMHL Alt Modes the receive links are used as transmit links because you don’t need massive bandwidth coming back from the screen. That is how you get four unidirectional high-speed links out of one still humble USB-C cable. All is good, 8K60 4:4:4 and no laws of physics bent.
Until you get to the 8K120 4:2:0 bit that needs six high-speed lanes, USB-C still only supports four if you swap two receives around. How do you get six out of USB-C? That one is easy, you don’t. No kidding, you can ‘only’ get 8K60 out of a USB-C cable with SuperMHL, if you want 8K120 4:2:0 you need to get a real SuperMHL cable which has, wait for it, six high-speed lanes. It actually all does work out.
Until you once again question the bandwidths needed to do all of this over a USB-C cable which supports 10Gbps per link. If you run the numbers, that really isn’t enough to do 4K60 4:4:4 over one high-speed link and the numbers for two, four, and six lanes don’t work out any better either. How does this work out without bending the aforementioned laws of physics.
Here is where things get a little twitchy on my end, SuperMHL uses compression. When asked if it was lossy or lossless, the Lattice folk called it, “Visually lossless compression” which could mean anything from what it sounds like to far less desirable results. Lets wait and see what happens with production devices before we break out the pitchforks and torches but if done right, it should be unnoticeable. In any case this allows SuperMHL to transmit 18Gbps over one differential channel which does allow the math to work out.
To get really geeky, the compression used by SuperMHL is line based and has a latency of about 3-4 scan lines per side, so 6-8 in total. Since a 4K screen uses 2160 lines, that is a latency of about .37% of a frame in total, completely ignorable. Once you get to 8K and the line count doubles, you will be at about .19% of a frame. Even if you are only doing 4K60, that is a pretty small number in seconds but I won’t spoil the fun, you can do that math yourself.
Another bit we will throw in here, MHL and SuperMHL both have mandatory compliance tests for vendors. That means if they say it will work, it really should, probably, most likely, work. If not, vendors will get their sticker and logo taken away or something, and it will probably be very expensive for them too. If it has a SuperMHL logo on it, it should pass the requisite data at the requisite rates.
There are a few features we should probably mention here too, some are MHL carryovers, others are new. If your phone and panel both support SuperMHL, you should be able to drive a 4K screen with just about any mobile product over a USB-C cable. Even if you don’t use a USB-C cable, the MHL/SuperMHL cables can charge your phone from the panel even without USB-PD, 40W is a standard MHL feature. Last is that the wired connection will also pass remote control data, not a bad thing to have if your device is wired up already.
So that leaves the two new chips Lattice is coming out with today, the Sil8630 and Sil9396. The 8630 is a one channel transmitter that takes eTDMS signals and outputs SuperMHL. It obviously supports 4K60 4:4:4 and takes up only one lane so you can also pass USB3/3.1 data too. It is a 4x4x.8mm package device aimed at mobile devices.
You might have guessed that the Sil9396 is a one channel receiver, and if you did you would be wrong, it will take three channels. This one is meant for a TV/panel so it takes SuperMHL and outputs HDMI2.0 signals. It is relatively huge at 9x9x.85mm, a bit big for a modern phone. Then again a modern phone has very little need for three SuperMHL inputs so no big deal there. Think of this futureproofing for your next high rez panel.
So in the end SuperMHL can do 4K60 4:4:4 over one USB-C lane, 4K120 4:4:4 over two, and 8K120 4:4:4 over all four available lanes. There is a bit of magic used, or at least a bit of hopefully unnoticeable compression to make it all happen but the numbers do add up. Throw in a few more useful features with the two new ICs, both sampling now, and the upcoming 8K panels should have a working way to transmit video when they arrive. 4K is almost yestertech, give it a few more months.S|A
Latest posts by Charlie Demerjian (see all)
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018
- Intel can’t supply 14nm Xeons, HPE directly recommends AMD Epyc - Sep 7, 2018
- AMD reintroduces the Athlon name with two CPUs - Sep 6, 2018