The Monkey House

< Writing : Other : Pro Display XDR >


Apple has announced the Pro Display XDR, a high-performance display allegedly nipping at the heels of reference displays. Scroll to the end to see how accurate my speculation was.

Reportedly, Ming-Chi Kuo says Apple will be releasing a "6k3k" monitor soon. Speculation abounds. My speculation concerns the video interface on such a beast.

The one reported specification, "6k3k," could mean a variety of things in terms of actual pixel dimensions. The two most likely seem to be 1.5x UHD, giving a resolution of 5760 x 3240 pixels, and 1.5x DCI-4K, giving a resolution of 6144 x 3240 pixels. Other options exist, but for the moment let's assume the larger pixel dimensions, giving us 19.9 Mpixels per frame. At 60Hz, that's 1.194 Gpixels per second. At eight bits per color channel (8bpc), that works out to 30.0 gigabits per second after you take into account CVT-R2 timings. At 10bpc, that's 37.5 Gbps. Add in some packet overhead (it's not immediately clear how much) and room for control signals or HDR metadata, and your numbers go up a bit more. But we're looking for about 30-40 Gbps for the video.

Apple has a variety of potential interfaces to choose from.

From a purely technical standpoint, HDMI 2.1 would seem to be the obvious choice, as it offers the highest bandwidth, at just under 43 Gbps. Furthermore, it offers the feature support, primarily in the form of HDR and 10bpc color support, which would seem to be a prerequisite for launching a high-end PC display in 2019. Even so, I can't see Apple switching back to a single-function HDMI connector after pushing single-cable attachment options for so long. Furthermore, HDMI doesn't support the P3 colorspace that Apple is pushing.

DisplayPort 1.4 is another unlikely option. It offers only about 25 Gbps, which means either reducing the refresh rate below 60Hz which is untenable on a PC display, or enabling display stream compression. That's a possibility, as it would allow 10bpc video, and DP otherwise hits all the feature boxes: HDR, P3 colorspace. But Apple certainly isn't going to be putting a big polarized locking full-size DP connector on anything, no matter how nice it is to have a cable that solidly attached to desktop hardware. And even mDP has disappeared from Macs.

If Apple were going to use DisplayPort, it would be via ThunderBolt 3, which encapsulates 8 lanes of HBR2 DP, with about 34.5 Gbps of usable bandwidth. That's enough to run 8bpc video. The problem is that TB3 supports only DP 1.2, meaning there's no HDR. Coupled with the slight bandwidth shortfall which prohibits 10bpc color - and no protocol support for DSC to make it up - I can't see Apple going DisplayPort either. They can't simply run it faster than spec either, like they did on the 5k iMacs, because it's an external device and has to adhere to the standards. Theoretically they could use two TB3 cables. Plenty of bandwidth, but it still has the DP 1.2 protocol problems, and requiring two cables is inelegant and offers more than enough sources of potential problems.

Then there's the proprietary option. A raw TB3 link has enough total bandwidth, though none of the individual interfaces it encapsulates can use all of it; Apple might find a way to use it all at once. Conceivably Apple / Intel could unveil a minor rev to TB3 that would allow some combination of higher DP bandwidths (8 lanes of HBR3 DP would more than saturate TB3) or higher DP spec versions, potentially bringing HDR, DSC, and 10bpc into the mix. Apple's new pro Mac and pro display could be the launch devices for this new "TB3.1" revision. Given where we are in the adoption curve of TB3, I can't see a TB4 with doubled bandwidth coming out soon, so this sort of minor spec update would be the only plausible TB option I can see. There are plenty of other theoretical options - Samsung has their single-optical-cable interface, for example - but they're all quite inelegant and would require Apple to place a single-purpose connector on their new computers, one which none of their installed base has. Even as a halo product, that's a big ask: buy a brand new >$5k desktop to run this >$5k display, because it doesn't work with anything else.

So what will they do?

I don't know! I think they'll try very hard to make TB3 work. It's a port that's useful even when you're not hooking up a display. It lets a display power a laptop and also operate as a port expander, which is an obvious weakness in their product lineup. And it's Apple's port of choice these days. If anyone can get just-a-little-bit of extra protocol support shoved into Thunderbolt silicon, it would be Apple as the interface's co-developer and a company not shy about custom chips.

If they can't get past the bandwidth and feature support problems with TB3, in other words if they're beholden to the same interface constraints that any other company would be, then I'd say HDMI 2.1. If the display really is a desktop-only affair, then a data-only cable is okay: a desktop will draw more power than TB3 can supply anyway, so both devices will be individually AC-powered regardless of the cable between them. But of all the things Apple might give up in order to have a halo 6k display product, I really struggle to see them giving up DCI-P3: everything Apple makes, from the iMac Pro down to the iPhone, supports DCI-P3. The only current Apple hardware which doesn't is the Macbook Air. And there's no way this rumored display wants to find itself in the same sentence as the Air.

My bet: TB3 with just enough Apple-y special sauce to make that work.

Update 20190303: What about eGPU?

Maybe. Finally. Having a big external display, with a high-powered eGPU and ports for all your peripherals and which also powers the laptop you plug into it, always seemed to be the point of Thunderbolt. But then Apple never made it. And then they stopped making displays entirely. And then abandoned desktops. Maybe they're finally getting around to using TB3 properly. Using PCIe through TB3 rather than using the DP lanes would solve the bandwidth problem: it's fractionally less total bandwidth, but can send compressed data and 'natively' supports partial refresh on the display if only one window's contents need updating. It's not an unreasonably difficult software proposition to have a laptop power down its internal GPU to save that power and thermal headroom for the CPU when plugged into a hypothetical eGPU/display, but have a desktop in the same situation recognize that its internal GPU is more powerful and should be left on for compute while the eGPU is dedicated solely managing to the video output.

Integrated peripherals and ports would let the display operate as a docking station without the un-Apple-like 'docking' part. A fast card reader, 10GbE, and a bunch of USB-A and USB-C ports would probably be all anyone would want. Maybe audio I/O too. And, as Siracusa notes, a FaceID camera. That would be the best-case scenario: the best standalone display for a desktop, plus the best possible display for a laptop, in a $5000 monument to Apple engineering.

Update 20190706: Display announcement

Apple's Pro Display XDR Specs give us a glimpse of what this display actually is. My guesses - admittedly not very hard guesses - were fairly accurate, though I still have some questions.

With pixel dimensions of 6016 x 3384, it is the same aspect ratio as UHD displays, but is almost 57% larger in each dimension than UHD, rather than the 50% that I had guessed. I'm still calling this one a pass, as I guessed 19.9Mpix total, versus the real display's 20.4Mpix, less than 3% off the true value.

Apple's display is aiming quite a bit more "pro" than I expected, but the basic assumptions I made were all spot on: P3 color space, HDR, 10 bits per channel color depth. The very high peak and sustained brightness values, and the use of full-array local dimming (FALD) put this display at the peak of the non-reference-display market.

As for price, another win for me: $5000. Though in fairness, I did assume that would include a display stand!

My final guess, TB3 with just enough Apple-y special sauce to make that work was apparently correct as well: Apple advertises the display as being a Thunderbolt 3 device despite the fact that it can't be, due to the limitations I noted above. Perhaps this fall when people start getting their hands on them, we'll find out more.

But I think I was actually wrong about Thunderbolt 3, despite what Apple's spec page says. There's no way Apple would not have included a USB 3.x controller in the display if it were actually using TB3 and thus had spare bandwidth for PCIe devices. I think those downstream USB ports on the Pro Display XDR are telling us something. USB 2.0 is distinctly down-market compared to everything else about this display, and there has to be a very good reason why. I think the Pro Display XDR is actually using the just-announced DisplayPort 2.0 standard which Apple erroneously calls TB3. DisplayPort 2.0 does use the TB3 physical layer but turns around the normal upstream lanes in the bidirectional TB3 port to get twice as much downstream bandwidth for video data. I suspect the USB ports are being fed by the DisplayPort AUX channel, which has run at 720Mb/s (Fast AUX) speeds in recent versions of the spec, and which explicitly can encapsulate USB 2.0 ever since DisplayPort 1.2. It's also possible that DP 2.0 contains a dedicated USB 2.0 port on the USB-C connector on pairs 13 or 14. This avoids the problem that even a non-standard variant of Thunderbolt 3 which uses the DP1.4 protocol has: there isn't enough video bandwidth without resorting to display stream compression, and that's not a high-quality option. With DP 2.0 there's plenty of bandwidth to run everything uncompressed at full color resolution with support for all the HDR metadata and color spaces that Apple advertises. This assumes Apple has custom versions of the Radeon Pro for the MPX modules in the new Mac Pro which include DP 2.0 outputs, which is a stretch (it's about a year before the market is generally expected to see such devices) but within the realm of possibility. The Mac Pro Specs page refers to its outputs as Thunderbolt 3 as well, but this may simply be a case of Apple marketing believing that the Thunderbolt brand has more cachet than DisplayPort.

Update 20191210: The beast released

An Apple KB article was just published today with further specifications. Additionally it has been brought to my attention that Intel's Titan Ridge TB3 controllers actually do support DP 1.4. So that resolves some of the feature discrepancies I'd noted above: Display Stream Compression and HDR support were both added in version 1.4. I still think DSC is a poor choice for such a high-end product, so Apple needs to find a way to send ~36.6Gbps of raw video data (and probably 38.3Gbps, with CVT-R2 timing) through a cable.

The supported Macs section offers some clues. Old macs which can only drive the display at 5k resolution get USB 3.1 gen 1 support on the downstream USB-C ports. And that makes sense: TB3 has enough DP bandwidth to support 5k resolution, DP1.4 provides HDR feature support, and there's 5Gbps spare for USB 3.1. But the new 16" MBP, newest 15" MBP, 2019 iMacs, and of course the Mac Pro, can all drive 6k resolution. The old MBP and iMacs all seem to have Intel's Titan Ridge controllers, and perhaps they use DSC to drive the display. BlackMagic's eGPUs also have Titan Ridge controllers. I still don't see any way, compliant to the standards as I understand them, to drive 6k at 10 bits per color channel through TB3 without DSC. The total bandwidth is (just barely) there, but as far as I can tell is not completely allocable to DisplayPort.

So there are three options. Apple may use DSC to reduce the required bandwidth. I may not adequately understand TB3, and there is some way to get more than ~35Gbps of TB3's bandwidth allocated to DisplayPort. Or Apple has cheated, perhaps by detecting the Pro Display XDR and running the TB3 link or its encapsulated DP links at slightly higher-than-spec speeds.

Version 1.6     |     Originally written: 2019-03-03     |     Latest revision: 2020-03-11     |     Page last generated: 2020-03-11 22:12 CDT