Slashdot Mirror
Oculus Unveils the Rift S, a Higher-Resolution VR Headset With Built-In Tracking (theverge.com)
Oculus VR unveiled the Oculus Rift S, a higher-resolution pair of virtual reality goggles that remove the need for external cameras by incorporating built-in tracking. The company partnered with Lenovo "to help it speed up manufacturing and to improve upon the design of the original Rift," reports The Verge. From the report: The result is a new VR device that is more comfortable, sports 2560 x 1440 resolution (or 1280 x 1440 per eye), and features the same inside-out tracking system that will ship on Oculus' upcoming standalone Quest headset, which the company calls Oculus Insight. That way, you won't need cumbersome cameras to enable full-body movement. In another twist, both the Quest and Rift S device will cost exactly the same at launch: $399, with the same pair of slightly modified Touch motion controllers included and the same integrated audio system (plus a headphone jack for external audio). That decision makes it clear that Oculus wants its VR platform to offer a choice not between two vastly different pieces of hardware, but by the more simple determination of whether you have the hardware to power PC-grade VR. The Rift S will support every existing and future game on the Rift platform. "The company is also enabling cross-buy and cross-play features," the report adds. "That way, you can buy a Quest and, at a later date, upgrade to a Rift S and still have your entire library intact. Additionally, multiplayer games that support both platforms will let players play one another, regardless of whether you're playing on a Quest or Rift device."
The Rift S and Quest will be shipping this spring.
The Rift S and Quest will be shipping this spring.
Sure, it has more inside-out cameras (I think), but it no longer has mechanically adjusted PD and no ear phones.
In some ways, it's a step backward. I'd rather have seen a hybrid approach to sensors - the inside out is great, but a couple sensor pods behind wouldn't be too bad. Oculus "partnered" with Lenovo on the "S" - and they basically swallowed up the Lenovo Mixed Reality headset (which can usually be had for a lot less than $399), and passed none of the cost savings onto consumers.
Where's the wireless option? Go and Quest are standlones, and have their own issues... how about something like Vive's wireless option? Instead of pushing VR tech forward, they've just sidestepped into Microsoft's MR standard. It's not a terrible thing... the MR headsets are very good, and inside-out tracking is very slick, but it isn't a step forward.
I want no tethers and galvanic stimulation to ward off motion sickness (and feel motion). I want OLED displays. I don't like losing the earphones.
hence the "Built-In Tracking"
Interesting note... the Quest uses a Snapdragon 835 SoC, which was designed to cheaply support 60ghz 802.11ad with the addition of a single chip and some passive components. While that chip obviously isn't in the Quest, I can't help but wonder whether Carmack might have been able to pull some strings and find a way to expose the pins needed to interface with that chip so it could be implemented as an external add-on.
In most Android devices, the USB port is actually connected to the USB root hub through a crossbar chip that enables the port's literal pins to be connected to multiple different circuits inside the phone, and used for purposes that have nothing whatsoever to do with USB. For example... JTAG, I2S digital audio, analog audio, a TTL serial port, whatever.
I don't have an Adreno 835 datasheet in front of me, nor do I have any idea how the Quest is wired up internally. But... given how badly Carmack WANTED the Quest to support Rift-like usage, I'm going to go on a limb and theorize that if there's any way the engineers could have exposed the necessary pins to connect the Qualcomm 802.11ad chip to the 835 through the crossbar and USB-C port without raising the manufacturing cost... they almost certainly DID. Carmack is someone who absolutely HATES to slam the door on a future possibility if there's any possible way he can leave it propped open for free. He might have lost the battle to add the extra hardware to support something like DisplayPort-over-USB-C, but exposing a data bus through a crossbar that's already "there" using lines that would otherwise have gone unused is another matter ENTIRELY.
Another possibility: if at least a few of the pins on the USB-C interface are connected to pins on the 835 that can be tri-stated, and the pins on the 835 that are needed to connect to Qualcomm's 802.11ad chip can be tri-stated, Oculus could have conceivably dispensed with the crossbar entirely and just connected the two sets of pins in parallel. It would complicate the software initialization slightly (if something caused the 802.11ad-interface pins to go high or low, like a firmware bug, it could cause the USB interface to malfunction), but in hardware terms, it would basically be free. Worst-case, they might have to add some pull-up or pull-down resistors that they might have gotten away with omitting if the lines were single-use, but officially are required for standard-compliance ANYWAY.
On the downside, I'm not sure someone actually COULD viably sell such an 802.11ad interface, even IF the signals were exposed as described. The problem is FCC certification. If someone built a module that used an already-certified mPCIe 802.11ad module and connected to the Quest via literal USB 3.1, getting it past the FCC would be easy... they'd just have to prove it didn't emit excessive RFI, and that the 802.11ad module itself was already certified. In contrast, if they tried to roll THEIR OWN 802.11ad module using Qualcomm's bare chip, they'd have to go through a much more rigorous and expensive certification and approval process that would likely result in an 802.11ad add-on interface that cost $25 to manufacture and had to sell for $600 to cover the cost of getting it certified in the first place.
Compressing with something like h.264/265 that depends upon having a few frames to reference adds latency. Doing things like HuffyUV... YPbPr with 4:2:2, RLE, etc, is fast, because it doesn't depend upon knowledge of anything besides the current frame.
The big delay with most codecs isn't literal calculation time, it's the need to wait until you have at least 2 or 3 frames in the pipeline before you can even START compression. For a modern GPU, matrix transformations between RGB & YPrPb or between 4:2:2 and 4:4:4, are practically instantaneous.
The key to making something like this work with a 1-2gbps link is to forget about codecs designed to be maximally-efficient, and instead look at codecs designed to be fast & "efficient ENOUGH".