So think I’m missing something here. At the moment the server to client for Windows needs rtx cards at both ends which precludes use of low end none rtx hardware at the client side that can still handle the direct hmd connection.
Is this on someone’s radar or something that we need to build out?
You don’t need an RTX GPU on the Windows Client. You just need a Pascal or newer GPU and either GeForce or Quadro are fine.
@MrGRID thanks. i could see that in the docs. The main issue here is the use of CUDA which is a specific NVIDIA solution. To make the client work with none nvidia pascal (or newer) gpu we’d need to replace the cuda code with opencl code which is supported by other vendors (eg my laptop uses Intel HD graphics and supports opengl and opencl). I’m not an expert on this, and i’m not expecting Nvidia to do the work - wondering if anyone else has the skills to port the client to opencl?
I don’t think it will be possible CloudXR utilises the NVENC encoder built into the GPU for decoding and encoding as far as I am aware. It probabably uses this under the hood: NVIDIA VIDEO CODEC SDK | NVIDIA Developer
The Windows client SDK only supports NVIDIA GPUs (Pascal chipset or later).
Reminder the client PC/laptop still needs to drive a high-resolution stereo HMD at say 90hz, including HEVC dual stream decoding, plus frame post-processing such as colorspace conversions or de-foveation, and then normal VR warp and render.
Thats fine if its a limitation of this version of the code - but that also precludes use of Radeon GPU. Also the client can run on Android devices (Quest 2 or Samsung Gear) so this cant be something thats baked in. It just needs the client to use the openCL rather than the CUDA aspects of the GFXplatform code