Current Generation Games Analysis Technical Discussion [2024] [XBSX|S, PS5, PC]

Maybe that's because the issue with ExecuteIndirect and such APIs is that they are not making the GPU's life easier, quite the contrary. You're supposed to use them when performance is limited by either the CPU or backreads from it (so that it's beneficial to do the job on the GPU for overall performance rather than wait for the CPU). As there is a Cortex-A15 class CPU (performance-wise) on the PS4, I can easily see why they tried to offload the weak CPU there by abusing indirect commands, but for a game of such complexity on PC, porting over such a design seems to be a bad choice.
I wouldn't be so sure of that. GPU performance for the past decade has advanced much more than single-threaded CPU performance. A game that pushes flagship GPUs to the limit at release will run easily on midrange GPUs four years later. The same can't be said for CPUs - a game that is bottlenecked by single-thread CPU performance at release will still be so years later. Also, PC gamers upgrade their GPU more often than their CPU, and there are many more of them with flagship GPUs and midrange CPUs or midrange GPUs and budget CPUs than the other way around.
 
I wouldn't be so sure of that. GPU performance for the past decade has advanced much more than single-threaded CPU performance
CPU single-thread performance, as judged by SPECint 2006, has increased by an order of magnitude since the Jaguar in the PS4. However, I am not so sure that the command processor performance in GPUs has increased by the same amount. Indirect execution stresses out the command processor, which is probably the most serial part of the GPU. That's why people perform compaction and other optimizations to achieve the best performance. As far as I remember, AMD has recently optimized the RDNA 3's command processor and increased its throughput. Maybe they foresaw that some games, such as Starfield, would abuse the serial command processor with ExecuteIndirect instead of using more parallel features like mesh shaders.
 
So tried the Ghost of Tsushima PC port, quick impressions:

The Good/Decent:

  • Performance with the High preset, at least in the opening gameplay scenes, isn't actually that bad for my 3060. High, very high textures, 16X AF, 4k with DLSS performance is just below a locked 60. Something like a 3060 ti could likely get a very stable 60, at least in these areas. The Ultra shadows+LOD give a big hit to performance so on high the game isn't as exorbitantly demanding as it first appeared.
  • Consistency of framerate when not GPU limited is excellent. No huge spikes at all.
  • DLSS performance looks very good, better than 4K CBR. However...
The bad:

  • Outside of shadows/draw distance, scaling with other graphical options is basically zero, at least on my rig. Dropping down reflections, volumetrics etc has very little uplift to performance.
  • DLSS looks very good in most instances, but like DF showed, it messes up with DOF - hard. Considerable specular aliasing/flickering. The particle effects are fine when dof isn't engaged, but as dof can be active in gameplay for distant objects, you can have extended gameplay sequences with lots of flickering pixels in the distance.
  • DLSS also doesn't take motion blur into account for certain elements - like a waterfall early on has significant shimmering with motion blur. Also once again, no separation for camera and object blur.
  • Ground textures without grass, like mud, are brutal. This isn't really the fault of the port as this is how the original looks and I would never expect a texture overhaul for the PC version, but I gotta think something could have been done to help this single asset look less like a trilinear filtered texture from 10 years ago. A detail map? Dunno, just anything.

The Ugly:

  • Frame/camera pacing. It is indeed fucked, at least at 60hz. Constant microjudder as Alex showed. Rivatuner/Special K cannot fix it.
  • Like Forbidden West, Dynamic Res (at least on my system), is broken. It's constantly overshooting by using a higher DLSS preset in scenes that results in more of a 'lock' to ~57 fps instead of 60. For example, a scene where I would get a solid 60 with DLSS performance gives me 57-58 with dynamic. It will never go to 60.

So overall, a little better performance than I expected, but it's annoying that this it the second Nixxes game in a row for me that has problems with post-process reconstruction artifacts (forbidden west also has issues with motion blur + DLSS and some vegetation), poor camera/frame pacing and a faulty (?) dynamic res. Like, reconstruction fucking up with post process effects + stuttering are my two big bugaboos!

As mentioned Nixxes is fantastic at consistent framerates, but it's largely for naught if the viewport movement doesn't reflect that consistency - a constant small microjudder is better than large shader stutters I guess, but it's still an anomaly compared to 95% of other games, including most of Nixxes past work (albeit come to think of it, even Ratchet and Clank had some pacing anomalies but they could be largely fixed with a Rivatuner cap). So perhaps something in their workflow/engine, or how they're evaluating actually presented frames has changed since, it's not a good sign they're not picking these up before release.

Hopefully these can be addressed, as at least the camera pacing is a deal-breaker for me - but as it's likely not something you're going to notice unless you're using a gamepad and slowly pan the scene so it may not be highlighted by the majority of PC users, thankfully at least DF shone a light on it so we'll see.

Edit: Hmm, maybe at least on my system this isn't just a camera problem! The actual framerate is staying locked at 60, but whoa - look at this mess:


1716845525178.png

This is what I would expect if I was running the game at PCIe 3.0 8X (which btw - would still be uh, bad!) , but I've confirmed my slot is at least identifying as 4.0 16X.

Is there a throughput utility I can use to see if my PCIe bandwidth is borked?
 
Last edited:

The biggest differences between the two I can see are the shadows most of the time, props to Andrew et al. on the virtualized shadow maps, they make a big difference between the crawly old cascaded ones.

Otherwise, sure nanite has "higher" poly counts and less pop in, but it's not so much that's it's an obviously huge jump, though that's versus the previous best in class for geometry density. And Forbidden West still has better water, at least during the day. Great work from the Guerilla programmers on the breaking waves and the arists on the daylight look.
 
  • Like
Reactions: snc
I think nanite is a huge improvement over other current geometry systems, especially when combined with virtual shadow maps. Alan Wake 2 is up there, but I haven't really placed anything else in the same ballpark as Hellblade 2. Truth is high resolution textures with displacement/normal mapping can be pretty good at faking things at the right distances, but you'll always notice shape limitations that nanite can handle without issue(almost). On top of that nanite just handles continuous lod better than anything else I've seen, and by a pretty big margin. The main issue with the quality of nanite is just being able to create assets that are high enough quality to show it off and to be able to fit them in memory and on disk. You'll see games have a mix of quality of objects depending on how close you can get to things during gameplay, or how much attention the player is likely to give something in a scene. Nanite has its own displacement system now which is partially to address storage and memory, I think, but also to allow for animating geometry and stuff.

Nanite basically offering per-pixel geometry works really well with vsms, because you basically get per-pixel accurate shadowing and it seems to greatly reduce shadow pop-in over other games. You just get a much more stable image than other games. A smooth curve will stay smooth as you get close to things, and tree lods and their shadows just seem to stay stable in the distance. So far I've heavily played two nanite enabled games: Fortnite and Remnant 2, and in both cases I think the games can be a great showcase for the tech. Remnant 2 has a lot more rough edges being a "AA" title of sorts that started as a UE4 game, but it still has moments where it shines, but in both cases the stability of the geometry is something I really appreciate.

Edit: I'm going to add that I think the consistency is actually more important than geometric complexity. You don't need super high-poly meshes. What I want is a stable image. I don't want lods and shadows to pop in. I just want them consistent. The actual meshes and world can be outdated and not cutting edge.
 
Last edited:
I think with Ghosts of Tsushima PC there's something going on with the particle effects that is outside the problem with DLSS and DOF. In addition, one thing I noticed from Digital Foundry's video and definitely myself is how blurry the PC version can look - there's no sharpening control as there usually is with Nixxes ports - there seems to be none at all. This is usually the opposite when comparing a PS5 game with CBR vs DLSS.

Look at this opening scene here. While there is some slight additional aliasing on the wheel spoke in the PS5 CBR image, the fire particle effects are all but invisible on the PC version. Even with DOF off, and even with a native 4K, they're still far less prominent than the PS5. This is not a question of graphical settings or it being obscured by more volumetrics, from very high to low there is no difference, and I moved closer to the source of the particles with no change - the PC is just not visibly rendering a lot of those fire particle effects. The only time they really come into view is when you move the camera back and forth and the motion blur issue with DLSS causes them to blink in and out of existence.


GOT PC - DLSS 4k Perf, High Preset:

RDny6cA.jpeg


GOT PC - DLSS 4k Perf, High Preset, NO dof

kAYKSgD.jpeg


GOT PC - DLSS 4K, TAA

YL5DcHO.jpeg


GOT PS5

e4i2ZWN.jpeg
 
I think with Ghosts of Tsushima PC there's something going on with the particle effects that is outside the problem with DLSS and DOF. In addition, one thing I noticed from Digital Foundry's video and definitely myself is how blurry the PC version can look - there's no sharpening control as there usually is with Nixxes ports - there seems to be none at all. This is usually the opposite when comparing a PS5 game with CBR vs DLSS.

With Nvidia drivers now you can Alt+F3 to bring up their filter options and add a sharpening filter.

DLSS has always been on the blurrier side compared to other upscale methods, is FSR just as blurry?
 
Not really the best video title, since that's not really the point.
But an interesting attempt to go as low as possible on hardware while still maintaining 60fps.

 
The biggest differences between the two I can see are the shadows most of the time, props to Andrew et al. on the virtualized shadow maps, they make a big difference between the crawly old cascaded ones.

Otherwise, sure nanite has "higher" poly counts and less pop in, but it's not so much that's it's an obviously huge jump, though that's versus the previous best in class for geometry density. And Forbidden West still has better water, at least during the day. Great work from the Guerilla programmers on the breaking waves and the arists on the daylight look.
Also density of vegetation (+ there is interaction with it not sure why its missing in hb2) in Foribden West is in own league, textures quality upclose also better in Horizon
 
Last edited:
It's certainly not Capcom-levels of AAA presentation, and I fully expect this game to get a lot of hate for that, but I still think it's decent enough looking, and captures the atmosphere really well which is really the most important part of Silent Hill games. And definitely still a huge step up from the originals in visuals and gameplay.
 
  • Like
Reactions: snc
It's certainly not Capcom-levels of AAA presentation, and I fully expect this game to get a lot of hate for that, but I still think it's decent enough looking, and captures the atmosphere really well which is really the most important part of Silent Hill games. And definitely still a huge step up from the originals in visuals and gameplay.

But would you say it's Konami levels of AAA presentation ?
 
Also density of vegetation (+ there is interaction with it not sure why its missing in hb2) in Foribden West is in own league, textures quality upclose also better in Horizon
Of course it has more vegetation density. One game takes place in the western United States and the other one in Iceand.

The lack of dynamic foliage in Hellblade 2 is unacceptable though.
 
Of course it has more vegetation density. One game takes place in the western United States and the other one in Iceand.

The lack of dynamic foliage in Hellblade 2 is unacceptable though.
Yeah, breath of the wild has really good looking dynamic foliage on a 2014 tablet. Meanwhile 2024 games on 2020 consoles struggle to replicate it. Horizon also has the same problem, it moves but it's barely noticeable.
 
horizon FW grass is a lot more detailed, varied, and much more is drawn than BoTW, and it reacts to player, NPCs and wild life too, there is a unrealistic way it reacts in BoTW as it looks like there is some invisible bubble around link pushing the grass before his body even hits it.
 
horizon FW grass is a lot more detailed, varied, and much more is drawn than BoTW, and it reacts to player, NPCs and wild life too, there is a unrealistic way it reacts in BoTW as it looks like there is some invisible bubble around link pushing the grass before his body even hits it.
It would be much better if it was a more exaggerated animation, you have to squint looking at the tv to notice it. If they are going for realism than that's hurting the visuals, without mentioning that so many plants actually don't react, it's very selective.
 
Back
Top