News Crytek CryEngine shows raytracing technology demo (runs on Radeon RX Vega 56)

[BFG10K]

So how far above 45% does it have to be before you can call it destroyed?

45% is about the equivalent of a generational performance jump in traditional rasterization. So if they did nothing except release faster 7nm hardware, they'd achieve the same thing.

If that's the best they can do in ten years, they've failed. It'd be like Intel saying "9700K is ten years in the making because it's 45% faster than Lynfield".

What complete nonsense.

Also BF5 was the case you talk about later on they just flipped the switch in the engine and where running uncontrolled raytracing,for all we know it could be the same deal as way back when with tessellation where they had tessellation active on the whole map even though in most of the map it was covered and wasn't even showing any tessellation.

Except leather-jacket-man never said tessellation "just works". He did for ray tracing, though. He also told us "just flip the switch" and "no developer effort or hacks".

Yet we see every patch that "improves" performance either dials back ray tracing and/or reduces quality elsewhere, with even more in-game hacks on top of the old ones.

 

[BFG10K]

I strongly feel we're being manipulated with some 1st class BS.

Oh I agree completely. Every time you see the DLSS label on a chart, replace it with "post-filter that upscales from a lower resolution that only works if nVidia blesses you".

It's a widespread public deception to label the resolution as native when it's intentionally running lower. 4K DLSS is not 4K, it's 80% of 4K + upscaling postfilter.

Likewise, it's a scam to label it anti-aliasing, unless nVidia thinks Bluray players and console games @ non-native are also "anti-aliasing".

DLSS is consumer fraud perpetuated on a massive scale, plain and simple. I'm honestly shocked nVidia hasn't been sued yet.

 

[jpiniero]

45% is about the equivalent of a generational performance jump in traditional rasterization. So if they did nothing except release faster 7nm hardware, they'd achieve the same thing.

If you look at Port Royal or the ProViz results, it's double the performance or more.

 

[Stuka87]

Still not sure how much trust we should put in a benchmark that nVidia worked on. Its in their best interest to make the RTX cards look better. The benchmarks may be fine, just don't blindly trust them.

 

[UNCjigga]

"Will it run Crysis...with RT on?" will be the new battle cry.

I wonder if Stadia games can take advantage of this tech...

 

[jpiniero]

If anything I would suggest that Turing doesn't have enough RT cores.

 

[coercitiv]

If anything I would suggest that Turing doesn't have enough RT cores.

Hehe, but this is in direct contradiction with the latest GPU activity graphs Nvidia supplied. According to those, in Metro Exodus RT cores activity covers only 10% of the frame time. Double RT cores count and you're likely to get less than 5% overall gains, diminishing returns at it's finest.

 

[Ajay]

Going back to the OP, if anyone wants to understand this more; there is a good discussion on B3D (post 1398 ff) : https://forum.beyond3d.com/threads/...raced-and-everything-else-spawn.60873/page-70

 

[Ajay]

If anything I would suggest that Turing doesn't have enough RT cores.

It doesn’t, the RT cores are bottlenecking performance. The rasterizer can shade more pixels the RT can produce. However, if more die space was given to RT at the expense of more generalized compute, Nvidia would have underperformed Vega20 in traditionally rendered games.

 

[Timmah!]

It doesn’t, the RT cores are bottlenecking performance. The rasterizer can shade more pixels the RT can produce. However, if more die space was given to RT at the expense of more generalized compute, Nvidia would have underperformed Vega20 in traditionally rendered games.

Are you sure about that? Why then in Octane render the performance improvement, when compared to non-RTX, is bigger in case of more geometry-heavy scenes? Would it not suggest its the shading holding things back?

 

[Ajay]

Are you sure about that? Why then in Octane render the performance improvement, when compared to non-RTX, is bigger in case of more geometry-heavy scenes? Would it not suggest its the shading holding things back?

Interesting point. Haven’t looked at Octane results, but performance is surely scene and implementation dependent (including how BVHs are built). There are changes in Turings shaders compared to Pascal.

 

[Ajay]

Speaking about the variations in performance based on implementation details, here’s a great read: https://devblogs.nvidia.com/rtx-best-practices/