Realtime raytracing future

Why exactly didnt AMD go MCM with GPUs already? Correct me if I am wrong, but its much easier to parallelize rendering across chips than CPU tasks. Yet Ryzen is MCM (with associated cross-CCX latency increases), but Vega is not..

This, Nvidia is taking all sorts of shortcuts. It is nowhere near on par even with real raytracers.

I think nVidia is primarily concerned with render farms. In realtime yes, they take all sorts of shortcuts.. But those can be turned off for offline rendering.. The speedup compared to a card without dedicated raytracing hardware is still massive, and render farms will eat it up.

GPUs are considerably more sensitive to latency than cpus

DESU, I wonder how fast would 2080 Ti be if they removed all rasterization shaders and dedicated entire chip area purely to raytracing cores + tensor cores for denoising.. Could we have good quality realtime raytracing now if we sacrificed backwards compatibility?

Not that much.

>what is SLI/Crossfire

And MCM would definitely have lower latency than that.

>Will this enable photorealistic realtime graphics?
More importantly, does this really matter? I mean, I don't want to be a graphics luddite, but is it worth 20x the performance to fill in those few last missing percents that are missing from photorealism? For visualization, who really cares? For games, arguably stylized graphics are better anyway. I'm sure it's fun to do as a cool technical demo, but for more practical applications, does it really matter?

>photorealistic
>true
Raytracing is still a vast shortcut to real global illumination.

What? The very idea with GPUs is that they prioritize throughput over latency.

Raytracing will also make work a lot easier for developers. Its much simpler compared to incredible complexity of modern high-quality rasterization renderers.

You still need normal shader units to do more general computation, so probably the end result wouldn't be all that different if you optimized purely for ray tracing.

That is perhaps true if we're speaking of raytracing more in general, but it is almost certainly not so in terms of nVidia's specific implementation, which is a very specific hardware implementation of a very specific algorithms, requiring developers to lock their whole architecture in to a very specific conception of whole-scene geometry, which must be organized in such a way that the GPU's fixed-function hardware can parse it. It is far more restrictive than general-purpose rasterization acceleration, and I don't expect any majority of developers to adopt it, really.

I think it would have been generally better -- especially seeing how the general demand for raytracing is... modest -- for developers to keep implementing raytracing in software (CPU- or GPU-side), and for hardware vendors to try to find more general-purpose ways of accelerating the algorithms so used. Right now, I'm resenting nVidia quite a bit for forcing a specific ray-tracing acceleration scheme through all the standards bodies as an unnecessary gimmick just to sell more cards this generation, failing to find ways to improve general performance.

Please leave.

Front end is the bottleneck that AMD needs to fix first.

>True shadows, reflections and lighting.
>Watch Nvidia presentation (aka Starcraft 2 intro rip off)
>Everything looks like it's made of shiny plastic
Realistic indeed