UL Benchmarks Outs 3DMark Feature Test for Variable-Rate Shading Tier-2

[FM_Jarnis]

maybe the level of implementation will be adjustable.

I would be surprised if it was not adjustable in games that implement it. In VRS Tier 2 test you can already do this in the Interactive mode that allows you to pause and move along the timeline, plus modify the shading quality, take screenshots... all kinds of things to see how varying the settings affect quality and framerate.

 

[biffzinker]

Forget my previous reply, I thought you were implying something else. @cucker tarlson

 

[Testsubject01]

67% uplift in framerates when I tested it (2080) @4K!

If the aim is to keep framerates above 60 @ 4K at all times, then it works very well at doing that.

..I'm sure if you looked hard enough you'd see where the cutbacks are but it's not obvious when in motion.

Did you test it in a game where 67% of it is covered by shadows and are dark areas?

Yeah, my thoughts too. Let us see the whole thing close up, in motion, in dark scenes, in mixed scenes, with all bells and whistles of post-op processing and AA.
The tech inherently gets its performance through lowering image quality, the question is, if its improvement in performance can be had at a low price in said image quality.

A single promotional shot is not really proof.

If it holds true though, its good news for 4K and VR on hardware people can actually afford, in light of the pricing on the high end as well as enthusiast-level stuff in recent years.

 

[kabarsa]

It was already included in DX12 standard, but it should be implemented by engine developers. To get shading rate image devs are responsible for reprojection of a previous frame, velocity buffers and other things, as well as threshold values for tiles to decide how much derivatives is enough to maintain shading rate or drop it. So it is not automatic. Feature can be used with a completely different shading rate mapping based on lens matching in VR or motion blur even. And it can be used for supersampling as well. It's up to developers.

 

[Chrispy_]

I have to admit that in motion the differences are minimal at worst, and I can only see them because I know where to look.

The sad fact is that AMD don't support this DX12 feature and with AMD having current gen and next-gen exclusivity, I doubt many game devs will bother putting the extra work in for only a fraction of a fraction of a fraction of their target market.

I dunno, given the PC vs Console marketshare, and then given the AMD vs Nvidia PC marketshare, and then given the marketshare of Nvidia PC GPUs that support this feature, I suspect it's not even worth a second glance for most devs.

The minute AMD support this, it'll gain momentum because consoles are all about squeezing the best performance out of fixed hardware.

 

[FM_Jarnis]

I would not be surprised to find out that next gen consoles support this. I don't know either way, but considering this feature is from Microsoft...

 

[Vayra86]

I like the idea of the things nvidia is trying out for the user to have the option to somehow "cheat" the framerate and provide tools for trading a little bit of IQ for noticeable performance gains.

First,there was multi res shading in shadow warrior 2,I usually ran the balanced setting that rendered 20% on each side at 1080p and the central 60% at 1440p.Noticeable in static image,not noticeable during gameplay.the game ran at pretty high framerates anyway.

then dlss,which it seems is really hard to do properly.played the whole control with dlss on and while for the most part it's doing an amazing job of keeping the IQ reasonably high,it's still got frequent glitches and bugs too where it takes too long to load when entering a new area or just plain doesn't recreate some objects.it's really good when it's doing what it's supposed to do,the problem is how inconsistent it is.when comparing static shots 960p dlss produces a tiny bit less jagged image and much less flickering than 1080p while still delivering about 10% more performance.but still,more bugs to work out than most of us would like to see.

with vrs it's easier and the performance gain is almost as high.seems like the best of all three by far.

The whole devs have to implement part is really the risky bit. If that is on the engine level and the 'heatmap' generation can be automated and tweaked in a simple, transparent way then all is well, but if its forcing devs to do per-game sequence optimizations that can go to shit, fast. Then it'd just be DLSS all over again.

 

[Fluffmeister]

So does Navi support VRS or not? It would suck for them to hold the industry back.

 

[biffzinker]

So does Navi support VRS or not?

Apparently no, but RDNA 2 is rumored to have it.

Variable Rate Shading

Engineering specs for DirectX features.

microsoft.github.io

 

[Apocalypsee]

Remember earlier times when ATi get caught doing bilinear filtering on Radeon 8500, and during GeForce FX era nvidia cheating in 3DMark03 test to get higher performance? So comes the time where you can lower the quality to improve performance and being as an accepted feature. In benchmarking test no less. Time truly has changed.

 

[Nater]

I would not be surprised to find out that next gen consoles support this. I don't know either way, but considering this feature is from Microsoft...

Apparently no, but RDNA 2 is rumored to have it.

Variable Rate Shading

Engineering specs for DirectX features.

microsoft.github.io

So we can bet it'll for sure be on the next Xbox, and more than likely will be on the PS5 as well.

 

[Fluffmeister]

Apparently no, but RDNA 2 is rumored to have it.

Variable Rate Shading

Engineering specs for DirectX features.

microsoft.github.io

Interesting, so Navi is nothing more than a stop gap, with talk of RTRT coming to consoles too, it seems RDNA 2 will actually bring feature parity with Turing at least, but no doubt Nvidia will bring plenty of improvements when they drop down to 7nm too.... so they seem will prepared for the future of gaming.

 

[Hotobu]

Since this only affects certain portions of some scenes couldn't this lead to some wildly inconsistent framerates?

 

[Nater]

Since this only affects certain portions of some scenes couldn't this lead to some wildly inconsistent framerates?

Doubt it, or we wouldn't be reading about it being implemented already. Some reason you jogged my memory, I remember reading that this VRS is going to be combined with eye tracking in VR to really take things to the next level. Quick google found this:

Easy VRS Integration with Eye Tracking

 

[xkm1948]

75.4% gain with 2080Ti. Hmmm if implemented well this means 2080Ti can handle stable 4k60Hz gaming maybe~

 

[Nater]

^ See. We DO need 8K monitors/TVs sooner rather than later.

 

[Zubasa]

I have to admit that in motion the differences are minimal at worst, and I can only see them because I know where to look.

The sad fact is that AMD don't support this DX12 feature and with AMD having current gen and next-gen exclusivity, I doubt many game devs will bother putting the extra work in for only a fraction of a fraction of a fraction of their target market.

I dunno, given the PC vs Console marketshare, and then given the AMD vs Nvidia PC marketshare, and then given the marketshare of Nvidia PC GPUs that support this feature, I suspect it's not even worth a second glance for most devs.

The minute AMD support this, it'll gain momentum because consoles are all about squeezing the best performance out of fixed hardware.

So does Navi support VRS or not? It would suck for them to hold the industry back.

The biggest thing holding the industry back right now is Pascal.
Given how popular it still is and the fact that it hardly has more rasterization features than Maxwell.

 

[Xaled]

A new tech by Nvidia always means higher prices, therefore same or worse performance/dollar.

So nothing really to get excited about

 

[Zubasa]

Since this only affects certain portions of some scenes couldn't this lead to some wildly inconsistent framerates?

Doubt it, or we wouldn't be reading about it being implemented already. Some reason you jogged my memory, I remember reading that this VRS is going to be combined with eye tracking in VR to really take things to the next level. Quick google found this:

Easy VRS Integration with Eye Tracking

The only major games I know that implements VRS so far are WOLFENSTEIN using the Vulkan API.
They minized the overhead and optimized pipeline via Async Compute.
http://www.leiy.cc/publications/nas/nas-gdc19.pdf

 

[FM_Jarnis]

In case you missed this earlier this year, this is a very good explanation from Microsoft DirectX guy as to what the feature is:

 

[biffzinker]

Reading material from Microsoft about VSR in DirectX 12

Variable Rate Shading

Engineering specs for DirectX features.

microsoft.github.io

 

[Vayra86]

Interesting, so Navi is nothing more than a stop gap, with talk of RTRT coming to consoles too, it seems RDNA 2 will actually bring feature parity with Turing at least, but no doubt Nvidia will bring plenty of improvements when they drop down to 7nm too.... so they seem will prepared for the future of gaming.

Isn't everything a stopgap when you're always late to the party? That's AMD's dGPU reality for a while now, and it suits the console crop well. They've even said as much about RT; 'we'll do it when midrange can run it'.

With RDNA2 they'll again have stuff we already knew. The real question is what Nvidia will have by then...

 

[cucker tarlson]

it seems RDNA 2 will actually bring feature parity with Turing at least

I don't know about that
RT cores
Tensor cores
simultaneous int32+fp32 (amd can do 16+16)
vrs
mesh shading

 

[lexluthermiester]

Soo... The first two pictures are 110% the same only with on/off in the bottom?

You missed the point. You will not see with your eyes the difference. You will see it with performance differences.

 

[medi01]

Let us all jump to "it surely is always indistinguishable in all games" based on one static screenshot, shall we?