Valve’s Standalone XR Headset ‘Deckard’ Reportedly Launching This Year at $1,200, Leaker Claims

It's okay if you want Valve to primarily grow the VR market, but that doesn't mean that's what they want, have to or will do. The Index wasn't aimed at the mass market either, instead targeted enthusiasts with a high end device showing what is possible with VR. HL:A similarly demonstrated how to do good VR game design, not how to design a VR game that makes money in a still small market.

I doubt Valve ever considered releasing a Quest 3S like HMD priced low enough to significantly grow the market, even though Steam Deck did exactly that for handheld gaming PCs. Instead they are going for the high end again, but this time making the HMD also attractive to non VR gamers with a 4K microOLED display (2022 eMagin "Steamboat") for a huge virtual display to work/play on, similar to AVP.

It will slightly grow the number of VR users, might turn out to be great for DIY hybrid games using UEVR, and longterm get developers to release more hybrid games, similar to PSVR2's hybrid AAA approach, that later may draw more users. But short term it can't grow VR anywhere as much as Quest. But that's okay too, as Valve's goals and benchmarks for success don't have to match yours either.

TL;DR: Valve won't save VR; Meta won't save VR; technology improving will save VR, but it will take time, and an approach that relies more on extending what exists instead of creating something completely new at unsustainably high cost.

Valve isn't the savior of VR, but neither is Meta, because user acceptance cannot be bought only with subsidized hardware or games. People will come to VR once the cost/benefit/hassle ratio is good enough, but bad ergonomics limited by lacking technology will prevent mass market adoption for years to come.

Meta now focusing on free-to-play and Horizon Worlds is one indicator that throwing money at AAA didn't work, AAA titles Meta paid for quickly vanishing from the Top 20 is another. Most people use standalone HMDs not in the way enthusiasts want them to, and billions invested didn't change that, so Meta is adjusting its strategy. And I think that is okay, because nobody, neither Meta nor Valve, is obliged to burn more money just because we want nicer toys.

I like the concept behind AVP, PSVR2 and Deckard: pick up new user where they already are, with using iPad apps, adding VR modes to console AAA titles or letting them play flat games on virtual screens, providing a use from day one without having to radically change their behavior. IMHO more likely to win over non-enthusiasts than asking them to switch to a completely separate ecosystem lacking all the tiny things they use.

VR growth will be slower with extending existing use cases to XR, lacking true XR killer apps, but no amount of AAA will change that. The Steam Deck exemplifies this: it doesn't create a new platform, it extends an existing one to a new form factor. And despite the constraints, it proved popular, similar to the even less capable PlayStation Portal many expected to flop. They win not by providing a grandiose new experience, but by providing more convenience that is worth the price and hassle for a (small) number of users.

If regular people like something, they buy and use it, and convenience is often way more important than quality. Which is why still very inconvenient VR cannot be forced into the market by throwing more money at games. For the foreseeable future it's a niche for those valuing immersion over convenience, gradually gaining more uses and content while tech and comfort improve. The trick is finding compromises that allow for a slow, but sustainable development. And currently Apple, Sony and Valve all seem to have more viable concepts for this than Meta who's strategy mostly bet on having lots of money to quickly buy user acceptance.

Don't underestimate how much more powerful the newer, yet to be released, AMD x86/GPU socs are. Yeah they won't be able to run hignend needed PCVR games, but if it is as powerful (or better) as an RTX2060, it will be able to run a lot, if not most PCVR games. But I'll bet, with most standalone headsets these days, it will shine with wireless streamed highend PCVR.

TL;DR: process improvements, ETFR, higher TDP and upscaling should allow a 4K Deckard based only on average progress and already existing technology.

We need 500% better efficiency.

Not really. The 2024 AVP is running on a 2022 M2, the Steam Deck on a 2022 AMD Van Gogh/Aerith. AVP draws up to 40W, while the Steam Deck is limited to 15W, so one way to 4K HMD performance is adding more battery to allow for a larger/faster SoC. The M2 alone still wouldn't be anywhere near enough, so it relies on eye tracking and ETFR, which is apparently more aggressive than on PSVR2, where gamers are unable to notice the difference, while AVP users see somewhat blurry regions during eye movements.

GPU power increases by roughly 1/3rd each year. With Deckard expected late 2025, it would be almost four years younger than the Steam Deck released in 2022-02, so just based on time passed, it would offer 3.1x the GPU performance. AMD just announced an Aerith Plus supposed to use 12 instead of 8 CUs, getting us to 4.7x even before ETFR. The original Aerith used RDNA2, while a Deckard APU might use RDNA4, which introduces a new FP8 data type to speed up AI that is used for the first time in FSR4.

FSR1 was a spatial only upscaler, FSR2 added temporal data, FSR3 frame generation, and FSR4 ML similar to DLSS for much improved results. And since SteamOS comes with the Gamescope compositor that already does system wide FSR1, with OpenVR providing the motion vectors necessary for FSR2/3/4, upscaling on Deckard probably won't depend on developer support and instead work for any PCVR game.

So a late 2025 AMD APU with ~ 1/3rd higher TDP (Aerith Plus 20W) than the Steam Deck could get you to almost 5x the raw performance. If Valve's ETFR is as efficient as Sony's/Tobii's saving 72% render time, we see a factor of 17x without the benefits of RDNA4/upscaling. Which Deckard will need, because the Aerith APU had to render 1280*800@60Hz, while the HMD will have to render 2*3600*4000@90+Hz, more than 40x the pixels per second. But that's actually doable without any magic involved, only new technologies that so far haven't been standard in HMDs, but are already in use, so not only theoretical options.

– Don't remember the source, the 40W were supposed to be a peak value while charging.

– Qualcomm had a generational GPU performance grows of ~32% per year following the SD820 for several years, Apple closer to 40%. Desktop GPUs gain less. Yes, this is raw peak power that will not translate directly to real world performance, but that doesn't matter for evaluating relative generational changes.

– The Steam Deck is supposed to run for 8h under low load with a bright 8" display and loud speakers, though you can drain the battery in 90min by cranking everything up in a demanding game. The Quest 3 battery weighs 69g, the AVP battery pack 353g. Deckard could use an AVP sized battery as counter balance, so this isn't really an issue.

– DFR on Quest Pro hasn't shown huge performance gains because it has to run on a rather weak XR2+ Gen 1and consumes so much compute that apps can no longer activate the highest CPU performance. AVP is a mobile headset with a much more effective implementation, very likely due to having way more compute power to throw at it.

AMD has published lots of roadmaps, so it has been known for years that APUs 15W TDP with the raw performance of a GTX 1060, so pretty much the 2017 PCVR entry level performance without reprojection, would become available around late 24/early 25, so this was the first moment when an x86 Deckard capable of running most PCVR games at low settings would become possible.

But Valve will very likely aim for a better experience than a Rift CV1 or Vive provided, so the APU will have to be more powerful and a lot of tricks will be needed to increase the effective render resolution. And of course this is all speculation based on rumors about a headset that may or may not appear in almost a year, so all the calculations are mostly about general feasibility, less about current implementation.