4 ms·
The vast majority of those problems sound fixable by software. In the absence of MSAA, supersampling goes a long way to improve crispness. With proper text rend
by DiThi 10y ago
The vast majority of those problems sound fixable by software. In the absence of MSAA, supersampling goes a long way to improve crispness. With proper text rendering (i.e. NOT E:D) you can squeeze much more information in your field of view, even when the game runs at a lower resolution. Not seeing IRL objects or not being able to do anything else at the same time are also software problems. E.g. with OVDDP[0] you can have a virtual "tablet" inside E:D with a window of your choice, in the world or in a controller so you can grab it. Double clicking system or the HMD button (on the left) enables the camera pass through (must be enabled in settings). Using a regular headset instead of the included earbuds to hear the people outside. Etc etc.
I find turning your neck in E:D an advantage: being able to look up (and down in some ships).
[0] https://github.com/Hotrian/OpenVRDesktopDisplayPortal/releases https://github.com/Hotrian/OpenVRDesktopDisplayPortal/releas...
- dogma1138 10y agoI find the free look nipple on my stick considerably faster than turning my head, but then again you might be better at summoning your inner owl than me. But overall I had no issues with the graphical fidelity, I can run Raw Data on the highest settings on the dual Titan X's. As far as the FOV it's the FOV it self, in sims with a UWD or screen spanning you can easily do 150 FOV, VR's limit you to 90-110 FOV, with some games even running effectively lower FOV which can be closer to 60-70 (e.g. Adrift). Overall for me it's the experience, cockpit games feel better without VR, VR specific games work very well, which is why I would like to see AR being pushed harder. If i can get the best of both worlds - 3 monitors and AR HUD that would be awesome for me.
- DiThi 10y agoI agree completely on the FoV. More than 110 wasn't possible because of how many pixels are wasted with a linear projection matrix. That is, until now, where the latest generation of GPUs have viewport multicast and we can render parts of the screen in different resolutions and projection matrices without having to re-render each mesh for each viewport. Between this and the VR software in development, the next 2 years will be very interesting.
- dogma1138 10y agoviewport multicast atm is somewhat vendor specific tho :\
- DiThi 10y agoIIRC AMD has it too, with a different name.
- dogma1138 10y agoAs far as I don't it doesn't and I can't find it for OpenGL. NVIDIA exposes it via both OpenGL (IIRC NV_viewport_array2) and DX (via NVAPI). With Pascal NVIDIA has SMP (to MPA) which is adds even more. There is a reason why the 1060 beats the Fury X in the most demanding VR titles these days. Yes they both run at 90fps but 90-98% of the time the AMD hardware resorts to reprojection.
- DiThi 10y agoOops, my memory failed. You're right, there's no equivalent with AMD, yet. There's equivalent for other VR things like ATW and direct mode, which I tend to forget they're driver specific features.
- dogma1138 10y agoYeah direct mode is "easy" it's a simple driver hack, ATW is kinda well meh, ATW is the oculus term for reprojection which means you don't hit the frame target so you have to resort to reproject existing frames, ATW reduces the judder by a bit but it still generates pretty bad artifacts this isn't something you ever want to use unless you really have too because your hardware can't hit the true 90fps target. Viewport multicasting and other true load reduce techniques are huge, NVIDIA had a demo with 2 VR headset running of a single GPU (I believe it was one of the newer Pascal Tesla's) and they showed that under worse case scenarios you get about the same load on the GPU as a single headset on a GPU without VPMC and multi-viewport shader pass through and under most circumstances they only were running at 30% additional graphics load because they could stream the entire 3D scene space only one, transform it once, and then do their pixel reduction screen lense transformation to reduce the amount of pixels that need to be calculated and drawn by about half.