Games Analysis |OT| Time To Argue About Pixels And Frames!

https://videocardz.com/newz/ul-releases-3dmark-mesh-shaders-feature-test-first-results-of-nvidia-ampere-and-amd-rdna2-gpus

I saw this posted on OtherEra and thought it would interest some of you here.

To start there is a caveat that this is just a benchmark of a feature and games have a lot more bottlenecks and components than just geometry processing, but it is an insight to how much more efficient mesh shaders are compared to the current method.

The Xbox Series has the hardware to do this along with VRS and sampler feedback and is a good reminder that devs have barely touched the capabilities of the Series X/S and we need to keep this in mind with the current slew of games being released.

@euV2 you beat me to it by seconds :sweat_smile:.

5 Likes

To say nothing of ML as well. Does the 547% figure include the updated drivers?

So what’s the technical TL:DR on this?

Updated drivers are showing 1750% uplift in performance lol.

https://twitter.com/FlorinMusetoiu/status/1359978463831224320?s=20

6 Likes

wat. lol 0.0

2 Likes

Yeah, the performance uplift in a game in TBC. When a game does finally end up using mesh shaders, I’m sure MS will write a blog like they did for VRS, but we might have to wait for Hellblade 2 maybe to get there.

2 Likes

Oh yeah, completely forgot about that lol. It definitely feels like there is a lot more exciting technical possibilities open to developers this gen compared to last gen.

I remember watching a video from Microsoft about mesh shaders and the employee mentioned that developers have to do a bit of work to take advantage of it, so I do expect it to be a while before we see developers using it and even then it will be first party developers who will likely use it before 3rd party. However, going by these numbers the performance uplift is dramatic.

2 Likes

Mesh shaders delivers huge performance boosts when processing geometry as expected. To the point where we can actually have a scene super ultra dense and have the gpu to parse all that data and render only what’s needed at a 1:1 pixel density.

Which is exactly what the UE5 demo promised btw, though with mesh shaders you can do the same for more types of inputs (the UE5 used a very specific implementation where the geometry was saved as a texture, but that has some limitations (no deformation for example, so not applicable to the characters)

Hopefully something like this

Probably a little inferior lighting + effects but better textures + polycount due to software + tool advancements.

One thing that I dont understand is NXganer + redgamingtech from youtube have said a few times that the PS5 hardware is different because its better at moving data then the seriesX. Does anyone know where they are getting this from? Like yeah the PS5s SSD should theoretically be able to move data to the ram faster, but I dont see this as a big win for the PS5 because how the game looks is still going to be dictated by what the GPU can render. The two youtubers I mentioned seem to think its more then just ssd+io transfer speed and something fundimentaly different with the PS5s architecture. I hope a clear lead happens sooner rather then later so these secret sauce style narritive can be retired.

Aren’t they getting it from Sony’s GDC presentation? It’s well known the PS5 is faster at moving data, and this is achieved through hardware and software, but I wouldn’t call it meaning the PS5 is fundamentally different compared to PC and the Xbox - it’s certainly not up there with the Cell processor for example which was a fundamental rethink of what a CPU is. In fact with features like SFS and mesh shaders, it could be argued the Xbox is in some ways fundamentally different compared to the PS5.

In any case moving data is just one part of the story, and as you say the GPU and CPU still need to render the scene. Where the faster i/o may come in is the ability to do something cool with it in terms of game design, but it is not going to make a PS5 game necessarily better looking.

Don’t depend on engineering information from youtubers. All they are doing is just regurgitating information from whitepapers or presentations. They don’t have engineering backgrounds and wouldn’t be able to survive at a tech company.

3 Likes

Is it though?

Some random youtuber saying it doesn’t make it true. If a workload is faster on PS5 or Xbox Series X depends always on the use case and the data. We have seen Series S been the fastest in Division 2, so what now?

6 Likes

NXgamer is full of crap and 100% biased, he only exists doing what he does cause he found a niche appealing to Sony fans since DF face offs favor the better hardware which is usually xbox

1 Like

So I guess Xbox Series X version of The Medium is the best place to play it going by the DF Video on the PC Version?

1 Like

Seems like The Medium Engine is heavy as fuck and dual rendering is a massive resource hog, needs a lot of optimization.

2 Likes

Seems like it.

At 15.16 he says “what we are looking at here is a significant shift in what the machines are focused to do”

This is just such bulloni, he is just saying this stuff to make the PS5 seem not inferior but just different. The seriesX and PS5 are not that different Microsoft went with 52cus @ locked 1825mhz and sony went with 36cus with smart shift and a max clock of 2230mhz, which does not mean much as we have seen in the PC space for ages. Here is an example of higher clocks actually getting less performance then more cus

Go to 13.15

The CPUs are the same, the ram setups are very similar both use GDDR6 and even though the seriesX has one pool but with different allocations having different speeds, the xsx ram will still be more performant because ram is allocated for different speeds anyway, not all ram usage uses its full bandwidth.

SSD do see the biggest difference but not as much as some people might think. While the raw 5.5gb/s vs 2.5gb/s is the biggest difference the decompression and io systems actually get similar results, they both have hardware decompression but the series consoles use part of a zen2 core and software optimization to achieve better io performance and instead sony instead of better optimising software being computed on a zen2 core they put dedicated hardware for io processing. So while we dont know which is the better approach, both approaches have the same goals.

This isn’t true. Both systems use a combination of CPU performance and custom compression block. Sony’s IO system also has some additional benefits over what MS have created, like the GPU scrubbers and co-processors that help manage data.

2 Likes

Raw speed is faster but actual apples to apples comparisons favor XSX wrt I/O and RAM utilization since SFS brings a 2.5 fold scale factor to streaming capabilities (it’s 3+ for textures alone but overall balances to a reliable 2.5 factor). Everyone ignores that for whatever reason. 4.8*2.5 = 12 GB/s (> 8-9 GB/s on PS5). The files being transfered into RAM are 2-3 times smaller on XSX if SFS is being used, so there are lots more room for assets there and less need to stream in the first place.

2 Likes