Valve said their foveated streaming results in only about 1/10 the bandwidth requirements, and AFAIK they had a special demo where all the image parts currently not in view were blacked out, without the player noticing (which shouldn't really work).
To get to 1/10th, you would have to reduce the transferred image to something close to 30%*30% of the full image, so 30°-40° h/v FoV. It is true that current DFR implementations on PC only provide ~30%-40%, while Quest Pro often struggled to get 20% performance boost. The limited overall effect, and much lower gains on mobile is not only due to simple, fast shaders, but also due to a still rather large area rendered to compensate for tracking/prediction errors basically requiring a larger safety buffer.
ETFR should see more performance with more precise tracking, more complex geometry separated into smaller parts, more complex shaders and higher FoV. The worst case is basically one large object covering the background using an uber shader that still always has to be fully rendered, no matter where you look. So increasingly more complex games with finer grained geometry expecting more capable GPUs on standalone HMDs with faster SoCs and larger FoV might see much higher performance boosts.
Given that the eye tracking on Frame has to be rather fast to allow for the massive bandwiths gains they claim, there is a decent chance that ARM ports of PCVR games using ETFR will benefit a lot more than on other standalones. What we got so far is hard to compare, with Quest Pro running very optimized Meta software on a now rather slow XR2+ Gen 1, the Play for Dream using eye tracking of unknown origin on the faster XR2+ Gen 2, and AFAIK the basic tracking on the Pimax Crystal being done by the integrated XR2 Gen 1, which then feeds Tobii eye tracking running on the PC. With massively varying parameters, it is hard to tell how much of the (lacking) performance gains from ETFR so far are due to fundamental issues, and how much due to implementation limits.
Espace publicitaire · 300×250
