Originally Posted by epic1337
yeah, i know you meant it as sarcasm but the key point is existing GPUs doesn't need RT cores to get a leg in raytracing, and its already a practical alternative to Nvidia's RTX GPUs.
so what would you prefer, push forward an Nvidia-exlcusive function and shut-off everyone else? or let them develop a different approach for the same result that applies to everyone?
even Physx went through this phase, Havok proved that physics feature isn't Nvidia-exclusive.
No you don't need RT cores for RT, most tracing renderers run on CPU although in recent years they do use GPU heavily for faster parallel compute and now even RT HW accelerations unless they want to be left in dust.
I would hardly call a GTX 1080 practical alternative of RTX 2070 for ray tracing. You can check the performances of 1080Ti even Volta vs Turing RTX cards.
It's not Nvidia exclusive, it's not shutting everyone else off. It's standardized in DX and Vulkan for anyone to use. If you don't want to use these APIs and their tracing "extensions/APIs" then you can do it in shaders on any other API but obviously without HW acceleration and much slower.
This approach does apply to everyone. It's not an Nvidia only API.
Anyone can make a GPU and write a driver that supports DXR and Vulkan tracing (either https://www.khronos.org/registry/vul...NV_ray_tracing
or make their own extension, OGL always had features as extensions and the same now goes for Vulkan, new things are added by those who make them "kinda" instead of having to wait decades for M$ to update DX, see what AMD had to do with Mantle to even get M$ to wake up and start making DX12 finally).
Why can't we use AMD/Intel GPUs for ray tracing, we can, why not, but they did not add driver support for DXR or Vulkan ray tracing, so anything that uses those two standardized APIs that offer HW acceleration on selected NV cards and run on shaders on other NV cards won't run at all on AMD/Intel because the driver is outdated and doesn't support these APIs. So if you wanted to say play Q2VKPT/Q2RTX you would have to go and write a wrapper for AMD/Intel and make the calculations on shaders/CPU instead of having it fail due to lack of driver support.
Why is AMD not rushing on board? Well they don't care, they always focus on server compute instead, that's what their cards are designed for, server/cloud use. So they have extensions/APIs for ray tracing for non gaming purposes on their GPUs, etc. They don't really invest as much as NV into their software and education/website/tutorials ==> most developers are familiar with NV's DX6-11/OGL tutorials since they've been there for ages, while AMD only recently has revamped their web to provide access to any developer relevant information. You search for graphics and API guides and you're gonna end up on NV's site sooner than on AMD, for over a decade. AMD now also has GPUopen and neither of their sites are exactly easy to navigate, search and use, confusing "modern" UI with article/news style list and that's it. No tree structure, nothing, no easy way to find guides, tutorials, documentation.
NV invests a lot into support as well despite AMD's solutions indeed can be faster and better overall, developers get support from NV and can actually find stuff they need, not spend days trying to dig out of AMD's pages on their own.
Programming and graphics are hard enough already without having to deal with poor resources.
AMD is playing a waiting game because they know their products are very slow (same as GTX are) with tracing anything, so why ruin their reputation even more by enabling a feature even 1% won't use because it's too slow on their hardware. They need to add HW acceleration first, then they will enable the support in driver/software. Till then they better finally fix their never ending HW encoder issues.