Current Generation Games Analysis Technical Discussion [2022] [XBSX|S, PS5, PC]

This is a hyperbole btw. Some people keep repeating over and over again how Spiderman is special and how it uses CPU to create BHVs and so on. Almost ALL ray tracing games so far utilize CPU heavily, most likely for BHV creation. There are certain factions

1) NVIDIA faction: "The game is not utilizing RTX GPU properly!!!" - No relation: Almost all ray tracing games so far utilized CPU heavily like Spiderman did, and most likely for similar reasons
2) PS5 faction: "PS5 is creating them with special hardware, you have to do it on CPU on PC, etc. etc." - It is quite possible PS5 may have some special trick up in its sleeve. I can't deny or agree with this statement.

1st argument is used by NVIDIA fans to speak badly of the game, despite not understanding or realizing that since the arrival of ray tracing to PC games, it has been super CPU bound, and almost always created an extra %15 to %30 CPU boundness. It is no different in Spiderman, it will simply take %25-30 more CPU power, just like how it happens in Cyberpunk, GOTG and many other Ray Tracing titles I've PERSONALLY tested the CPU impact of.

Some people see how CPU bound Spiderman is and naturally think it is evil caused by ray tracing and Sony/Imsomniac purposefully gimping PCs to not utilize RTX hardware. I really don't think it is the case. I'm sure HW accerelated Ray Tracng still plays a part in BHV structures, but it is quite evident that almost all ray tracing games so far have increased CPU demands. On top of BHV stuff, Alex also stated that extra objects drawn / rendered in reflections also pile up on CPU as well.


Again this is not some special trick and this is known since 2020 an Xbox Series presentation. Read Digitalfoundry presentation or last UE 5 presentation, on consoles this is possible to build BVH for static geometry offline and just stream it from the SSD.
 
All I know is that this game (Miles Morales) runs better for me than Spider-Man Remastered while swinging through the city... and DLSS3 seems REALLY good now. It feels a lot better than it did in SM:Remastered when I played it last. There's still some slight artifacting if you look hard enough.. but for me personally, in this game.. I'd call it a non issue. Before DLSS3 felt like some kind of weird acceleration was happening and there was a slight disconnect... but now it's just as responsive as you'd expect, and feels like the game is running at the framerate being displayed.

As for negatives, the texture bug still exists.. although I feel like the late textures resolve a bit faster than in Remastered.

But yea, quite happy with the game. And I've noticed with the past handful of releases, I've been quite happy overall.. which is very welcome since for a while there I was quite down on the state of PC gaming. I feel like things are slowly, but surely getting/going to get better. Sackboy was one of the recent ones I was really down on.. but they turned that around rather quickly which is nice. I'm mostly just worried about The Callisto Protocol now and although I'm pretty sure the game is going to have the compilation stuttering issue at launch, I'm hopeful that Alex covering it will get them to fix it quickly.

If only developers would utilize DF, and maybe have them do a QA pass before they launch their games... maybe that would cut down on the number of games which have this issue at launch. 🧐
 
Again this is not some special trick and this is known since 2020 an Xbox Series presentation. Read Digitalfoundry presentation or last UE 5 presentation, on consoles this is possible to build BVH for static geometry offline and just stream it from the SSD.
I should probably know this already, but I'm drawing a blank here. Why is this not possible on PC? DXR API limitation?
 
Crowd density on PS5 is below Very Low on PC - that OG Video had the settings on PC for crowd density not functuoning.

A Ryzen 5 3600 with RT on can get 60 FPS and above in tiems squared with my optimised settings. If your 9900K gets worse than that, I would be shocked.

For Spiderman, the 9th gen Core series had abnormally poor performance, and saw noticeable boosts with HT disabled. Just somewhat of an oddity with the 9th gen Core series.

They really have to fix that texture pop-in issue though, surprising it still hasn't been addressed. I can't see that it's a processing limitation as it makes no sense to stream in every high res textures in cutscenes but not have the processing power left for a cop's radio.

The PS5 in RT performance Mode Drops below 60 in this game when swinging

This is true. It's relatively rare but Miles is not quite the locked 60 that Spiderman is on my PS5, which made my concerned about Miles' PC potential performance. Good to hear outside of some demanding RT it's actually a bit better than Spiderman RM.

But people told me this port was amazing!! A 4090+5800X3D dropping to the 40's on an enhanced PS4 game with RT is lolworthy. Terrible port performance-wise.

Here's a video with i7-13700k and a 4090, max settings 3440x1440. The res is largely irrelevant as the 4090 is generally under 70% usage here so it's still a CPU bottleneck, but in general it's 80-100 fps swinging through the city. The video I linked earlier was a 5800X, not the 3D cache version as well.

1668882763949.png

On the RT shadows as a whole: I question the value-add of the RT shadows myself considering the visual bugs, reduced quality even at max settings in some cases, and the massive performance impact - having an even higher setting than very high doesn't really seem like much of a 'solution' to me considering how many CPU's can barely handle it now.

They're not as egregious as Sackboy's Ultra RT which showed zero benefit mind you, but this veers towards the territory of the 'slapping-enhanced-RT-on-the-box' scenario that befalls PC ports where certain RT features weren't in the original game and there isn't enough time/budget to truly optimize them. Ultimately it just means most youtubers will crank all the sliders to the right and give the impression the port is performing far worse than it can with more reasonable settings, I think if you have a setting that's going to bring the most advanced CPU's/GPU's to their knees it should be prominently superior to the rasterized implementation in all aspects, or you tuck it away in an 'advanced' section of the config menu with a disclaimer of how demanding it is.

All indications are with maxxed RT but with RT shadows off, many modern CPU's will be able to achieve ~100+ fps. So it's not great like the original, but I wouldn't say 'terrible' performance-wise either.

This game also has ugly ass shit reflections on PC like the other game does.

They look dreadful and if they're genuinely that poor on PC I'm not using RT.

The RT reflections were indeed quite flickery at points. That, along with the texture issue that hasn't been fixed 3 months out, flickering tree shadows in some lighting conditions which the PS5 version doesn't exhibit, DLSS specular/DOF issues, culling performance bottleneck issue when leaping over buildings (has that been improved? It's not an RT issue, it affects non-RT modes just as much), I wouldn't put this into the 'great' port category myself.
 
Last edited:
This is a limitation of DXR but I think IHV prefer to keep the BVH optimization they do undocumented.

Ah yes, that's right. Nvidia locks things down pretty hard I'd imagine. With AMD though, and AMD sponsored titles, I still don't really understand why not.

I wonder what's next for DXR?

I think they have done a few of these already.

Yea, I believe they have in certain cases. We know they often contact developers to inform them when a game has a particularly egregious issue/s out of respect, to allow them a chance to address it before their videos go live.. but I'd love to see some kind of official co-operation throughout the later stages of optimization and leading up to release.

Obviously just day dreaming here, as realistically they shouldn't need DF to essentially post-op their game to find issues when they should be clearly visible to anyone with 2 eyes... but I digress. I'm sure their QA departments are informing them of all these issues... and it's the lack of action in a timely manner, or inaction of higher ups within the chain which causes these issues to go unresolved by launch.. but who knows. I just think having DF in the mix even a bit sooner, could help management of some of these studios/publishers see the importance of getting it right for launch, an not some weeks/months later.
 
Been like that since release and I have no idea why. My 2080 Ti paired with a 9900K is constantly sitting around 60-70% usage which shouldn't be the case.

They were low even for DDR4 systems, but Spiderman is also the poster child for DDR5 too for some reason. I don't think there's another game which shows such dramatic benefits with it over DDR4.
 
Is there a reason we ignore that consoles use APUs and unified memory?

The gist of the discussion seems to revolve around the ideal that the cpu on a PC is doing more work or working more inefficiently than it’s console brethren. Thereby the gpu is forced to wait on the cpu and is often underutilized.

Could it be that consoles do a better job of keeping their gpu fed because their memory systems are more performant? And that’s it’s no fault of the developers because it’s hard to optimize around that reality?

Not really as it's trivial on PC to prove that the CPU is the actual bottleneck. You can literally see the utilisation of the CPU being maxed out on one or more threads. You can turn up or down settings which are known to have an impact on CPU performance and see the game performance change accordingly, or you can simply swap the CPU out and see faster CPU's reducing the bottleneck.

Unified memory will certainly make things easier for devs - Nixxes said as much in their interview with Alex. And if the split memory pools of the PC aren't handled properly that that certainly could result in performance degradations that wouldn't happen on a console (e.g. excessive swapping of data between system memory and vram over the PCI-E bus), but if used skilfully there is more, not less raw performance to be extracted from the PC model thanks to the greater overall bandwidth afforded from separate pools, the lack of contention over that bandwidth between CPU and GPU, and in the CPU's case, lower latency memory access as well.

If the issue were simply an inability to keep the GPU fed because of limitations of the PC's memory setup then increasing PCIe bandwidth would result in a big difference in performance, but in virtually every test of that nature ever performed, it makes almost no difference as long as the baseline speed is sufficient, e.g. going from PCI-e 3.0 16x to PCI-e 4.0 16x.

PC CPU's doing more work than console CPU's isn't just an idea, it's a fact. Both logically and confirmed by devs. For example we know that currently, IO decompression needs to be performed by the CPU on PC's where-as consoles have a dedicated hardware unit for this. And Nixxes have said in their interview with Alex that that uses around 1 full CPU core on the PC for Spiderman. There's also the thicker API's on PC which will suck up a little extra performance - again something Nixxes confirmed.

It was actually recorded at 1440p even with the black bars but it seems Geforce Experience or Youtube just puts the video in the aspect ratio of the monitor or more than likely the aspect ratio of the OS resolution at the time of the upload. Maybe the NVIDIA capture tool just ignores the OS settings. Not sure. Whatever the case, as I play, it’s definitely 16:9. I made sure to also change the monitor resolution to 2560x1440 and I even set the aspect ratio to 16:9 here:


Identical performance. Also, changing those settings usually has almost no impact in other games. Usually narrowing or widening the FOV has virtually no impact in my games, nor does manually setting the display resolution to 16:9 rather than just in game.

You're running at Object Detail 8. That's two steps higher than Alex's optimized settings that he used on the 3600x and will account for a decent amount of CPU performance. It's also higher than what the PS5 is using in performance mode, albeit marginally. I must admit I would expect to see more performance from a 9900K than that though even knowing the additional overheads the PC CPU needs to deal with in this game. Are you sure there isn't anything else that could be impacting things? e.g. background tasks or suboptimal system memory config? Unless the Coffee Lake architecture is particularly poorly suited to this game... perhaps that is indeed the case looking here (check out the minimum fps):

1440p-High-p.webp

This is normal. PS5 CPU does less stuff for example in RT mode the static part of the BVH is streamed. I don't expect Spiderman Miles Morales to push the SSD but in future title it will cost more on a PC CPU than PS5 CPU to decompress CPU data and manage memory to check in and other stuff.

It will cost less in the future because games will offload this to the GPU via DIrect Storage.
 
This game also has ugly ass shit reflections on PC like the other game does.

They look dreadful and if they're genuinely that poor on PC I'm not using RT.

Yeah in the case of the streets the PC reflections are clearly sharper and more detailed, but it seems as though they are missing the snow/melt water texture that is present on the PS5 which adds a lot IMO - at least from what I can see in the screens. On the other hand the foliage reflections are clearly better on the PC and I'd certainly say RT is a much better option than SSR.
 
This game also has ugly ass shit reflections on PC like the other game does.

They look dreadful and if they're genuinely that poor on PC I'm not using RT.
Yes, it's strangely inconsistent. Sometimes, I'm actually quite impressed by the reflections, but most of the time, it's very clear somethings off and they look terrible. A lot of it has to do with the lighting conditions. When the sun is at a low angle, you get decent shadows in the RT reflections, but at certain times, it almost completely lacks visible shadows and looks very flat and low quality.
 
Not really as it's trivial on PC to prove that the CPU is the actual bottleneck. You can literally see the utilisation of the CPU being maxed out on one or more threads. You can turn up or down settings which are known to have an impact on CPU performance and see the game performance change accordingly, or you can simply swap the CPU out and see faster CPU's reducing the bottleneck.

Unified memory will certainly make things easier for devs - Nixxes said as much in their interview with Alex. And if the split memory pools of the PC aren't handled properly that that certainly could result in performance degradations that wouldn't happen on a console (e.g. excessive swapping of data between system memory and vram over the PCI-E bus), but if used skilfully there is more, not less raw performance to be extracted from the PC model thanks to the greater overall bandwidth afforded from separate pools, the lack of contention over that bandwidth between CPU and GPU, and in the CPU's case, lower latency memory access as well.

If the issue were simply an inability to keep the GPU fed because of limitations of the PC's memory setup then increasing PCIe bandwidth would result in a big difference in performance, but in virtually every test of that nature ever performed, it makes almost no difference as long as the baseline speed is sufficient, e.g. going from PCI-e 3.0 16x to PCI-e 4.0 16x.

PC CPU's doing more work than console CPU's isn't just an idea, it's a fact. Both logically and confirmed by devs. For example we know that currently, IO decompression needs to be performed by the CPU on PC's where-as consoles have a dedicated hardware unit for this. And Nixxes have said in their interview with Alex that that uses around 1 full CPU core on the PC for Spiderman. There's also the thicker API's on PC which will suck up a little extra performance - again something Nixxes confirmed.



You're running at Object Detail 8. That's two steps higher than Alex's optimized settings that he used on the 3600x and will account for a decent amount of CPU performance. It's also higher than what the PS5 is using in performance mode, albeit marginally. I must admit I would expect to see more performance from a 9900K than that though even knowing the additional overheads the PC CPU needs to deal with in this game. Are you sure there isn't anything else that could be impacting things? e.g. background tasks or suboptimal system memory config? Unless the Coffee Lake architecture is particularly poorly suited to this game... perhaps that is indeed the case looking here (check out the minimum fps):

1440p-High-p.webp



It will cost less in the future because games will offload this to the GPU via DIrect Storage.

Not all decompressed data is for the GPU, this is why I talk about CPU data and all memory management is done on the CPU. This is partially offloaded of the CPU for Xbox Series console and fully offloaded of the CPU for PS5.
 
Not all decompressed data is for the GPU, this is why I talk about CPU data and all memory management is done on the CPU. This is partially offloaded of the CPU for Xbox Series console and fully offloaded of the CPU for PS5.

Nevertheless, compared to today where all data, both CPU and GPU destined is decompressed on the CPU, in future with Direct Storage, only the CPU destined data will be decompressed on the CPU which should generally account for no more than 20% of the overall data set. So certainly in future the CPU's decompression work should reduce.

Also I don;t think you're correct about the series consoles there. The Series consoles hardware decompression block can handle zlib as well as BCPACK as far as I'm aware, meaning it will decompress everything coming from the SSD just like the PS5 will. The CPU shouldn't have to get involved at all.
 
Nevertheless, compared to today where all data, both CPU and GPU destined is decompressed on the CPU, in future with Direct Storage, only the CPU destined data will be decompressed on the CPU which should generally account for no more than 20% of the overall data set. So certainly in future the CPU's decompression work should reduce.

Also I don;t think you're correct about the series consoles there. The Series consoles hardware decompression block can handle zlib as well as BCPACK as far as I'm aware, meaning it will decompress everything coming from the SSD just like the PS5 will. The CPU shouldn't have to get involved at all.

I don't talk about decompression. I talk about memory management like data check in. This is done on the complex I/O(2 coprocessors, one DMAC, some SRAM and the hardware decompressor) on PS5 and partially done on one of the coprocessor of Xbox I/O on Xbox Series. There is more to do than decompression when you load data from a fast storage.

On PS5, this is fully transparent. Using Sony I/O API, developer just load a batch uncompressed files and they can use the option to give some priority level for data loaded. All the process is done by the Sony software and the hardware complex I/O, custom controller...
 
Last edited:
I don't talk about decompression. I talk about memory management like data check in. This is done on the complex I/O on PS5 and partially done on one of the coprocessor of Xbox I/O on Xbox Series. There is more to do than decompression when you load data from a fast storage.

This is insignificant. We already have DirectStorage demos which show near 0% CPU usage in extreme data loading scenarios (GPU only data).
 
No this is coming from Xbox Series presentation itself. They only have one I/O coprocessor against two on PS5 and they use a little bit of CPU power to do some of the operations. Same on PS5 side, everything was detailed from spec sheet to road to PS5 presentation, the 2015 and 2016 patent and some very interesting tweet from Fabian Giesen.

All of this is done on the CPU and GPU on PC side.

This is insignificant. We already have DirectStorage demos which show near 0% CPU usage in extreme data loading scenarios (GPU only data).

Again I wait game where we will have real data with CPU and GPU data. I have no interest in demo...
 
No this is coming from Xbox Series presentation itself. They only have one I/O coprocessor against two on PS5 and they use a little bit of CPU power to do some of the operations. Same on PS5 side, everything was detailed from spec sheet to road to PS5 presentation, the 2015 and 2016 patent and some very interesting tweet from Fabian Giesen.

All of this is done on the CPU and GPU on PC side.

It was Fabian Giesen in one of his blogs that specifically said the IO helper processors in the PS5 are nothing special and that every SSD controller has them. They are just DMA engines running custom firmware. That's the whole point of DMA - to offload the work of data transfers from the CPU.
 
Back
Top