Comment by spicyusername | Hacker News Reader

Comment by spicyusername | Hacker News Reader

Parent Full thread

The last paragraph really sums it up

    In other words, Nvidia has earned the right to be hated by taking the exact sort of risks in the past it is embarking on now. Suppose, for example, the expectation for all games in the future is not just ray tracing but full-on simulation of all particles: Nvidia’s investment in hardware will mean it dominates the era just as it did the rasterized one. Similarly, if AI applications become democratized and accessible to all enterprises, not just the hyperscalers, then it is Nvidia who will be positioned to pick up the entirety of the long tail. And, if we get to a world of metaverses, then Nvidia’s head start on not just infrastructure but on the essential library of objects necessary to make that world real (objects that will be lit by ray-tracing in AI-generated spaces, of course), will make it the most essential infrastructure in the space.

    These bets may not all pay off; I do, though, appreciate the audacity of the vision, and won’t begrudge the future margins that may result in the Celestial City if Nvidia makes it through the valley.

It’s a very good summary. Nvidia to me gets a lot of flak (sometimes rightfully so) in large part to being the big risk-taker in the GPU/graphics/AI space that pushes forward industry trends. AMD, meanwhile, is largely a follower in the GPU/graphics space with few contributions to rendering research. Even worse, they even use their marketing to angle themselves as consumer-friendly when I reality they’re just playing catch-up. For instance, they started off badmouthing ray-tracing as not being ready back when Turing launched (when is any new paradigm ever “ready” when it requires developers to adapt?) before adopting it with much worse performance the following generation.

Well, if that "consumer friendliness" is usable Linux drivers, then I don't give a shit about NVidia pushing the envelope, even if it means the graphics holy grail of real time raytracing.

And is NVidia really the primary pusher of graphics algorithms and AI? Yeah, I doubt it. SUre they might contribute some things, but modern fabs have so much available transistor real estate that really NVidia is just trying to maintain market control by defining the computing interfaces rather than some fundamental achievement in computing theory.

Raytracing algorithms aren't new, and they are "embarrassingly parallel" like most graphics algorithms, very well suited for modern chip fab design. If NVidia went away, Sony/Microsoft/Nintendo would just find another graphics vendor for real time raytracing. There were "processor per pixel" raytracer experiments in the mid 1990s when 66 MHz was "fast". If anything, real time raytracing likely has been suppressed by the massive investments in rasterization toolchains.

I mean one of the pain points are mainly that they treat other companies so poorly many in the industry just kinda hate their guts, not just AIB partners as shown by EVGA's messy divorce, but also others such as Apple or the Linux Foundation. While this happens their consumers are ambivalent towards them due to the fact that all their "envelope pushing" mostly being proprietary tech that drags the performance of other competitor's cards and the fact that their high end cards are price gouged akin to Intel CPUs prior to 2017 due to a lack of meaningful competition. All that and their recent negative pr from price gouging the mining boom, the "4080" 12gb, and Jensen being publicly out of touch about the price of computer hardware "unfortunately never going down" in this context has dragged their goodwill a bit lately.

The author is not doing a good job of differentiating ordinary ray tracing with path tracing. It is path tracing that actually gives all of those wonderful render effects "for free." Ray tracing by itself isn't that impressive. But we are a long time away from path tracing being a widespread and practical rendering method.

The big worry here is that nVidia is pursuing the equivalent of internet streaming video in the 1990s. Sure, eventually the technology will get there, but that could far enough in the future that nVidia will be disrupted or displaced long before it is a reality.

It also accuses rasterization of being "a hack", as if 1 spp with the denoiser coming to the rescue was nothing of the like. Also no need to label things as "hack"; they are all approximations, and good ones at that.

It also then argues that developers get RT "for free", which is nothing of the sort since there are many details involved to get it to work decently.

Current RTX hardware is rather lame to only get to 1spp isn’t it?

It’s not even clear that it’s the right approach.

A graphics engineer I know has this to say about it. “Telling me you made hardware accelerated BVH is like telling me you implemented hardware accelerated bubble sort.”

I don't think real-time ray tracing is lame. Diffuse is low-frequency and redundant, denoisers work magically well. It's 1spp per frame, but the denoiser is essentially allowing the application to distribute the workload across frames. For perfect specular a single ray is sufficient, and glossy is somewhere in between the two. The limitation doesn't have to do much with firing more rays than handling more complex materials; what I just described decouples diffuse/specular and is how games are doing it these days. But if the current materials are plausible, then what's wrong with this approach?

I do think it is the/a right approach... when used judiciously and when an alternative rasterization approach doesn't already exist or has obvious flaws. For example, some games use a hybrid SSR-RT reflection; do the SSR first, then fire rays where the SSR misses (insufficient information in screen-space.) I think the UE5 Matrix demo does this. You kind of get the best of both approaches.

The article in this sense proposes a false dichotomy; to choose between the complex "hacks" of rasterization, or the easiness of RT. Except that rasterization has decades of research and solutions are well-known and already implemented, and RT is not 'do the dumb thing and let the hardware do the work'.

I don't think BVH construction is as bad as your friend paints it. There are semi-parallel approaches available, i.e. https://www.researchgate.net/publication/327216026_PLOCTree_...

I am no fan of nVidia, but yes, rasterization is a hack. I like to call raster GPUs overglorified sprite rotoscalers. Every little thing that light does in reality, except perspective, needs to be painstakingly implemented in dedicated code. Path tracing is much closer to just modeling what light does, and all the optical phenomena fall out of it naturally.

One man's "hack" is another man's "elegant solution".

Reply on news.ycombinator.com