During the GTC, NVIDIA’s Don Palomo conference, they have presented the technologies that will allow apply Path Tracing in games, although for the moment they have shown it only for Cyberpunk. Obviously, since it is an event created by them and the company’s marketing is very powerful, you have to separate the wheat from the chaff. And that’s what we’re going to do.
For some time now, Jen Hsen Huang’s company has been choosing to implement new technologies not in terms of hardware, but to take advantage of the enormous computing power left over from its graphics chips to do so using software. But not the one that works on your processor, but libraries and programs based on CUDA that are executed as computational shaders, that is, general purpose and not for manipulation of graphical primitives, to solve problems that could be solved with fixed function accelerators. that would do the job automatically, without developer involvement.
CUDA to tie them all to NVIDIA
We have an example of this in Shader Execution Reordering, a function that NVIDIA has introduced in its RTX 40 that is based on the ability to reorder the execution order of the different threads that go to the GPU cores according to the units of execution that are available at all times. This is a job that should fall to the kernel scheduler, but NVIDIA has decided to make it a feature of their own libraries.
That is to say, there is no hardware implementation that performs this work and so on with a host of functions that are being promoted lately. What they have done is create a GPU, the AD102, and its graphics card, the RTX 4090, with so much power that no current game really uses it 100%, and then they have launched a series of justifications in the form of proprietary software to justify the existence of it and tie developers to its ecosystem.
NVIDIA Path Tracing is not a standard
All of this brings us to the issue of the NVIDIA Path Tracing SDK and its big difference compared to Ray Tracing in 2018. At that time, the initiative came from Microsoft to create a version of DirectX that allowed us to use what we call hybrid rendering, that is. That is, using ray tracing algorithms to solve a series of visual problems such as reflections, transparency, light transport, etc. that are not possible in any other way. Therefore, a series of norms agreed upon by several manufacturers for the creation of a standard originated.
Now, although Path Tracing is an algorithm to generate 3D scenes in real time that has been defined for decades, what NVIDIA has presented at the GTC is not and it is the umpteenth attempt by NVIDIA to tie users to your platform. Like the brand’s DLSS, it is software that runs only on its graphics cards. That while it brings benefits, it has made NVIDIA a de facto monopoly in both high-performance computing and artificial intelligence, and this is something that has lasted for years.
One only has to look at the technologies that make up the NVIDIA Path Tracing SDK, all of them proprietary and impossible to run on a competitive GPU. Which are:
- DLSS 3.
- RTX Direct Illumination (RTXDI).
- NVIDIA Real-Time Denoisers (NRDs).
- Opacity Micro-Map (OMM).
- Shader Execution Reordering (SER).
- Radiance Cache.
All of them proprietary technologies not implemented in the hardware but running under CUDA and requiring an RTX 4090 to give good performance.
A technology only for the privileged
Since game developers want to sell as many units as possible, it doesn’t make sense to adopt a technology that only 0.31% of users who play games will be able to enjoy at the moment. One thing is DLSS, which is still a banal addition, and another thing is what NVIDIA is trying to do with the Path Tracing kit that they presented at the GTC and that currently only works in Cyberpunk Overdrive.
We should view its implementation as a mere curiosity at the moment, no DXP-style standard, DirectX Path Tracing, or Vulkan extension has been created that everyone can use. In any case, the steps to follow will be the same as those given by NVIDIA, but without tying developers and users to a single brand.
Although given the necessary power, we will still have to wait a few years. In any case, it is most likely that we will see a good part of these functions being applied within the hardware, with new units specialized in said tasks and improved versions of the current ones.
There is no doubt that the implementation of Path Tracing is the next evolutionary step, but it is still years away from its massive standardization and not a single generation of graphics cards.
