I'm looking at it more like CUDA with a hint of G-Sync, from a marketing perspective.
CUDA in terms of NV put forward their hardware version specification first. As far as I know DXR is hardware agnostic. However, NV getting RTX out basically around the same time the standard (DXR) was announced, it gives them a 1-2 year lead over their competitors version that would use a more standard spec (ie variable refresh existed long before G-Sync was a product being marketed but most people would know it as G-Sync before Freesync/Adaptive Sync.) By getting developers to work with RTX I'm betting NV is hoping of having a CUDA effect where they'd stick to use NV's RTX extensions which wouldn't affect console games, but like we've seen for years with GameWorks throw huge monkey wrenches into AMD's product performance.
Just NV playing their usual games, and it's clearly working.
EDIT: typos galore!
It's way too early to argue that Nvidia's implementation of raytracing will become the standard just like how CUDA become the standard for GPGPU programming. CUDA for years had no effective response against it unlike raytracing which will almost certainly see a competing implementation against it in less than 2 years since Nvidia's introduction ...
Also each vendors likely have very different HW implementations of raytracing as well under the hood so don't get the idea that either AMD or Intel will follow the DXR specifications as is without any of their own unique traits. Based off of AMD's patent alone they do hardware accelerated intersection testing in the TMUs and handle the BVH traversal in the shaders. Intel looks to be seriously pushing the idea of programmable traversal shaders from one of their research papers so it's very likely that they're customizing their shader/ISA design to handle this as efficiently as possible. As for Nvidia from extensive analysis, their RT cores do fixed function BVH traversal while the ray intersection tests are done on the tensor/shader cores ...
Intel's possible implementation vs NV's current implementation look to be extremely polar opposites of each other with wildly different performance characteristics depending on the workload design ...
Each vendor's design comes with their own set of strengths/limitations and by no means is Nvidia's implementation is impervious like you seem to imply it is. With AMD's RT implementation, doing customized intersection test shader programs would be a very bad idea since they can't use the intersection units built into their TMUs. Intel's traversal shader concept is good for reducing bandwidth consumption but there is some overhead involved with the additionally generated shader invocations. With Nvidia, ray traversal is totally fixed function so traversal shaders would end up being a bad idea on their hardware since they have to emulate this without being able to use their RT cores ...
Come DXR 2.0 specification and Microsoft chooses to standardize Intel's traversal shaders, it could very well end badly for Nvidia because then they'd be forced to significantly reachitect their HW designs for ray tracing compared to Turing or face a huge performance cliff if god forbid developers decide to write custom ray traversal shader programs ...
Facebook
Twitter