Current Generation Games Analysis Technical Discussion [2020-2021] [XBSX|S, PS5, PC]

[zed]

Sure some ppl may play on monitors, But I expect most use TVs right? Like I said above its typically monitors & PCs vs TVs & consoles
WRT supersampling, its free if its really free (but nothing really is as unless its eg tetris, you could always add more, eg post process, higher res shadowmaps, textures, better filtering whatever )
If you are downscaling from 1440p to 1080p then that is a waste for a marginally better image, For the optimal use of the resources you should always target the final resolution

I just rendered my game in 2160p , 1440p, 1080p with all 3 being downscaled to 1080p

better quality where you can see the differences, yes 2160 is better than 1440 which in turn is better than 1080p but you've lost half your framerate in the process, worth it? Sure if you're playing on a 2160p screen but if its a 1080p screen then no its not

http://zedzeek.com/junk/ 1080p.jpg 1440p.jpg 2160p.jpg

 

[snc]

 

[rabbit]

I wonder if anyone has already compared Nanite and Mesh Shaders?
What are the differences? Advantages and disadvantages, etc.

 

[techuse]

I wonder if anyone has already compared Nanite and Mesh Shaders?
What are the differences? Advantages and disadvantages, etc.

Andrew stated Nanite already uses mesh shaders for triangles that use the fixed function hardware. For micro polygons Nanite presumably beats mesh shaders.

 

[snc]

PS5 and Xbox Series X in the Ray Tracing Mode both use a dynamic resolution with the highest resolution found being 3840x2160 and the lowest resolution found being 2560x1440. Pixel counts were often similar for PS5 and Xbox Series X in the Ray Tracing Mode with one exception found being the opening cutscene where pixel counts came in at approximately 3200x1800 on Series X and 2560x1440 on PS5.

Stats: https://bit.ly/3EbGbwX

 

[rabbit]

Andrew stated Nanite already uses mesh shaders for triangles that use the fixed function hardware. For micro polygons Nanite presumably beats mesh shaders.

Is Mesh Shader ineffective for deformation/hair/vegetation/fluid simulation?

 

[iroboto]

Is Mesh Shader ineffective for deformation/hair/vegetation/fluid simulation?

its perfectly suitable for that.
The only reason you wouldn't use mesh shaders is for micro geometry. ie, sub 8pixel triangles.

 

[techuse]

Is Mesh Shader ineffective for deformation/hair/vegetation/fluid simulation?

No, but if those triangle get smaller then 4 or 8 pixels performance plummets unlike with Nanite.

 

[rabbit]

its perfectly suitable for that.

And Nanite is the opposite.
Such details are interesting to me.
The Matrix Awakens does not use Mesh Shader. And in fact, I didn't find any mention anywhere that UE5 natively supports Mesh Shader.
I would be grateful if you share the link.

 

[Jay]

And in fact, I didn't find any mention anywhere that UE5 natively supports Mesh Shader.
I would be grateful if you share the link.

Nanite already uses mesh/prim shaders (check out the shader source) for hardware rasterization. As with any other features, I'm sure we'll continue to evaluate and use those that make sense.

 

[iroboto]

And Nanite is the opposite.
Such details are interesting to me.
The Matrix Awakens does not use Mesh Shader. And in fact, I didn't find any mention anywhere that UE5 natively supports Mesh Shader.
I would be grateful if you share the link.

https://forum.beyond3d.com/posts/2125567/
^^ basic high level premise as to why nanite must use compute shaders vs mesh shaders.
Mesh Shaders are similar to compute shaders on the 3D Pipeline, it redoes the geometry front end and dispatches work very similar to the way compute shaders would do it's job, which means it needs to use the hardware pipeline to draw a triangle. As per the graphs in that post, you can see as it approaches pixel sized triangles, the hardware basically crashes.

Pure compute shaders however write back to memory, and they have full control over the result, so if they detect that the triangle is sub pixel they can remove it without penalty. Unlike the hardware rasterizers which basically break down trying to map a triangle into an increasingly smaller size.

If in the scenario there are lots of triangles that are larger than 4/8 pixels, 16 pixel is still ideal, then the fixed function rasterizers are probably more performant here.

 

[snc]

 

[snc]

suprising ps5 perf advantage and defenitly bigger than 1fps suggested by df

 

[Clukos]

suprising ps5 perf advantage and defenitly bigger than 1fps suggested by df

Nit-picking but nanite doesn't use tessellation as NXG is describing in this video, if anything nanite is the exact opposite of tess as it is used in real-time rendering

 

[Globalisateur]

suprising ps5 perf advantage and defenitly bigger than 1fps suggested by df

It's not surprising. I read plenty of people (who own both versions) on forums saying the demo performed noticeably better on PS5. But I think it's only visible when going fast on a big road (as the game is capped at 30fps).

Nit-picking but nanite doesn't use tessellation as NXG is describing in this video, if anything nanite is the exact opposite of tess as it is used in real-time rendering

I heard it as the general definition of tesselation (adding more tiles ie. polygons). Not the one done by dedicated hardware.

 

[Phantom88]

He's mismatching footage, he doesnt keep the same line on both consoles and they run at different time of day it seems. Also, isnt that demo random, with the cars and pedestrians ? Can you mirror match two runs ?

 

[Shortbread]

It's not surprising. I read plenty of people (who own both versions) on forums saying the demo performed noticeably better on PS5. But I think it's only visible when going fast on a big road (as the game is capped at 30fps).

I mentioned it was a bit smoother on PS5, especially, during rapid speeds. At certain points (mostly surrounding heavy traffic roots with parked cars nearby) the X version felt like it was dropping down in the mid-teens more frequently (like an auto-save had occurred). Regardless, neither systems is hitting 30fps consistently during rapid speeds/motion.

 

[Remij]

Yea I noticed the footage doesn't seem very well matched.

Still likely a tiny advantage to the PS5.. but quite clearly Epic are more invested in optimizing for PS5, so it's not entirely unexpected.

It's not going to matter much, because multiplat devs will work to make the games perform close enough... and then you'll have exclusives on both sides which utilize the engine and not have any direct comparison between them.

 

[PSman1700]

He's mismatching footage, he doesnt keep the same line on both consoles and they run at different time of day it seems.

Yea I noticed the footage doesn't seem very well matched.

I dont put any energy in watching NXG's analysis anymore, theres too much bias involved (against the xbox). The PS5 very well might have an advantage in this techdemo, but whenever i see an NXG video shared, theres following posts like these of how well the benchmarks/analysis where done.
Sharing a video with such content is going to backfire in the comments, im sure the posters are well aware of that (but post them anyway). I generally try to refrain from posting such content as it doesnt only paint a different picture but also tends to de-rail discussions.

 

[snc]

It's not surprising. I read plenty of people (who own both versions) on forums saying the demo performed noticeably better on PS5. But I think it's only visible when going fast on a big road (as the game is capped at 30fps).


I heard it as the general definition of tesselation (adding more tiles ie. polygons). Not the one done by dedicated hardware.

I heard that also from polish youtube creator making comparisions, I checked both versions but quite briefly, not feeling capable of benchmarking it and wiritng results on forums.