RTX is Nvidia's implementation of DX12 DXR. AMD can't run RTX for what I understand. AMD has to implement it into their drivers seperately (if not already done in some form).

Fox2232:

I am talking about implementation. Game either uses DX12 code written by someone in studio, or uses nVidia's proprietary library optimized for their HW. (With usual consequences for other companies.)

Or, AMD partners to bring their own optimized library for Vega / Polaris / Navi.

Fox2232:

That's the thing. Developers are not going to make 2 separate code paths. DX12 was meant to be above that and there could be different visuals as result too which would be very undesirable for any multiplayer game. Then there is fact that AMD's "library" would be open source and nVidia could optimize quite easily if dev used just theirs.

Sadly that story has been going on for years now, exactly like that. If anything it's Nvidia's money making a closer / earlier adaption into RTX possible as they spend $ (manhours) into working together with m$. Only shows how DX12 is kind of lackluster as to what it was hailed as, and what it really is, YEARS later.

The goal behind DX12, aside from the CPU overhead stuff, was just to give developers lower level access to hardware. Picture a machine with 20 layers, in DX11 developers had access to layers 20-10, 10-1 was a complete mystery to them.. that's where AMD/Nvidia engineers reside and write driver code. With DX12 developers now have access to layers 20-5. The drivers still play a role but the developers can see/change much more going on at the very core of the system. The lower you go the more complicated things get, the more you need to get in there and write things for specific vendors, architectures - things Nvidia/AMD driver teams have been doing for decades now in the hands of game developers.. but the lower you go with optimizing the more performance you get out of those systems. The thing is most devs don't care to do that work. They have a budget for their project and why would they dedicate 7-10 developers to re-architecting low level code that Nvidia/AMD already handles pretty well? Like at most you get 10-15% performance in certain scenarios, pretty cool, but you can do the same by just like cutting a few poly counts down with a slider, which doesn't require a dedicated team. The big engine guy's - companies like Epic/Unity/DICE with Frostbite, etc.. those guys have a reason for it. They are architecting engines which many games utilize.. so they already have those teams in place with developers that are familiar with that level of development. But they still can't optimize for every scenario that random company A utilizing their engine is going to do. So even there they aren't going down to the metaphorical "level 5" of optimization. Most of Nvidia's GameWorks libraries are open source now - RTX libraries use DXR as the base so as long as the RTX library is open AMD shouldn't have a problem accelerating. I don't think Nvidia has anything to gain by keeping the RTX libraries Geforce only - I think they are going to rely on the fact that they've been dedicating the last 10 years of R&D to deep learning/self driving, are super far ahead there and can use that to make their hardware better at accelerating. The goal for them should be mass-scale adoption of DXR, which seems to be what they are trying to push.

Fox2232:

I am talking about implementation. Game either uses DX12 code written by someone in studio, or uses nVidia's proprietary library optimized for their HW. (With usual consequences for other companies.)

If the Tomb Raider RTX performance reports are anything to go by, then tbh I could care less about proprietary code. The performance needs to be increased by 100%+ to make it even viable and if that means Nvidia needs to take it in-house, then, so be it. Long-term is another story, but, seeing as AMD and Intel can't run RT at anywhere near the performance of Nvidia with their hardware, then, it's meaningless to them anyway.