Just curious if any games right now are using ALL of the following:
VRS, SFS, Mesh Shaders and DXR
As far as I know, Gears 5 (according to Digital Foundry) pushes Series X harder then any other game right now, and that uses VRS. We know that Watch Dogs Legion and CoD are using DRX.
Are any upgraded games using SFS and Mesh Shaders as well, along with VRS and DXR? Its fun at this stage to speculate on how much more performance developers can extract from Series S|X hardware. All things considered, with a half-baked SDK, its not doing too badly now.
To my knowledge there are no games that use mesh shaders or SFS. I also think it is going to be a while before we do since from what I have read and understand it requires developers to change how they go about engineering their games and it is really not something that can be ‘bolted on’ ala Gears 5 with VRS.
The full feature set? None. Right now there is no game using mesh shaders or sample feedback (and even less sample feedback streaming). I think velocity architecture is pretty much not used atm as well.
Gears and Dirt 5 are using VRS. And CoD, Watchdogs and DMC5 SE raytracing, with Call of the Sea and The Medium as upcoming games that will use RT.
Note: It is not clear if mesh shaders or SFS are legitimately key features of RDNA2 (they likely aren’t core features). SFS actually for sure isn’t (it isn’t available on PC, just SF is). I agree that given GDK is still kinda busted and devs are hamstrung til it is fixed, XSX is doing great. Especially in the area of loading times, where it is negligibly behind PS5 in typical use cases even without SFS in use.
Mesh shaders are definitely one of the pillar features of rdna2 and dx12u.
Ms and Nvidia were touting this for a while as it allows orders of magnitude of more geometry than current games without tanking your framerate.
Sampler Feedback I agree that as it is, the benefits are muted. I can see it being used in tandem with VRS for performance reasons (you can get sampling data from intermediate buffers and adjust some quality based stuff based on the screen size each tile actually occupies on screen), but for streaming I think only when SFS is implemented by hardware on Pc that it will be used extensively there.
Nvidia has nothing to do with RDNA2 and AMD never mentioned it as a pillar up until their conference and even then it’s been muddied with cross talk on features (RDNA2, XSX, 6000 Series, DX12U). Earlier in the year AMD noted the key features in slide presentations and never were they highlighted. They instead chose to highlight RT, VRS and perf/wat as the key features.
Auto-HDR – It was based off work done from machine learning algorithms, but does not use ML in the consoles. The Series X|S Auto-HDR implementation is in the video display adapter and does not use GPU resources, not even the ML INT4/INT8 instructions. There’s a YouTube video by EvilBoris that goes over that aspect of it.
The only part I remember they talking about VRS and not Mesh shaders was on their Fidelity FX effects, but that’s the equivalent of Nvidia gameworks as in: implementations of stuff that is quick for the developer to use.
I dunno how Mesh shader would be highlited there because it requires some shifting on how geometry is processed, and one of the key advantages is that the developer can customize the geometry processing for their pipeline instead of having a fixel model where the content must be aligned.
But, they did mention Mesh shaders as one of the key RDNA2 features:
That is just showing off the 6000 Series’ DX12U features…not base RDNA2 features. The AMD conference really screwed up everyone’s understanding of what features related to which classification. 6000 Series are not reflecting only the baseline RDNA2 required features, it has extra stuff too (like Infinity Cache). XSX also has extra features beyond baseline RDNA2 (as does PS5, though i lacks a key RDNA2 feature in VRS).
This is from earlier in the year, before they (and MS) muddied the waters:
Yeah, that SFS demo Ms showed in the SS announcement showed it loading on the fly GB/s of texture data as the camera flies around the room. A game designed for that in mind surely will have trouble porting back to HDDs, so likely a few years off.
Interesting to note that the INT4/8 support for shaders in Series S|X are also in the 6000 Series cards, according to Digital Foundry. That’s good, that potentially means any AMD super-upscaling tech can be ported over to Series S|X.
Great feedback all, so no game is using Mesh shaders and SFS yet. Its impressive what the hardware is doing now, perhaps more so on the Series S. When devs start using SFS, they should be saving memory and bandwidth. Understandable that early games aren’t using that and Mesh shaders but all in all, I think the future is bright.
Hmm I see your point now. Could it be perhaps because mesh shaders requires a shift in the asset production pipeline to really take advantage of increased polycounts so adaption /benefits will likely be a few years off?
AMD’s super resolution is based on Ms, likely they are working together to offer a easy to use package for developers. And once it’s done I totally expect to be used on SS/SX
At the 18:36 mark, Richard talks about Amds supersampling tech and how it is not the same as ai upscaling ala DLSS (i.e. supersampling done using ai accelerating hardware).
Xbox on the other hand is actively working on ai upscaling using hardware as shown in the video too. If Xbox manages to get this feature out quickly, it will be a massive performance boost especially for Series S, if it is anywhere as good as DLSS it could deliver better results than native rendering at way lower performance costs.
I don’t think PS supports int4 and int8 extension on the shaders to implement this type of ai upscaling, it could be the big thing that separates Xbox and PS this gen.
So Amds Super resolution(temporal supersampling) and Microsofts AI upscaling could even work together.
MS could allow its implementation to carry over to pc if they want to.