Sony is still far far away from releasing day on games on PC
1 Like
I think the problem with your line of reasoning is you think they will continue to shoot for 4K or near 4K. We have already seen plenty of sub-4K games on both the PS5 and Series X, so there’s no reason to think they would push for 4K if it would create additional artifacts.
Regarding RT and DLSS support, RT was spotty at best before the new gen consoles came out and the number of titles that support DLSS is surprisingly still really low considering how long the feature has existed. The latter was especially surprising to me considering how much Nvidia has dominated the PC space and how much they invest in developers supporting their technology.
Last, I disagree that the difference between CBR and temporal reconstruction is irrelevant. CBR is more prone to produce artifacts, especially at low resolutions, while TI may produce a soft but still clean image. Games like Spider-Man on the PS4 Pro will drop below 1440p but still looks amazing on a 4K display using the temporal injection technique used in their engine. The more I think about it, the more I think a modified version of AMD’s Super Resolution with a temporal reconstruction addition has a much larger chance of catching on than anything specifically tied to Microsoft or Xbox.
With them recently acquiring Nixxes, I wouldn’t worry about that.
1 Like
I think you’re misunderstanding of what I am saying. I am saying: internal res–>CBR to higher res–>4k via bilinear upscale. You keep talking as if I said they would go: internal res–>CBR to 4k in a single step.
DLSS support was slow on the uptake bc few cards existed to use it. That is improving over time, just as the prominence of RT rises since that is the use case in question. Devs will have high asset fidelity PC games using RT and targeting 4k/60fps via using DirectML SR and many/most of these games will be built on the GDK. Why would you expect them to not use that tech on the XSX version once it’s available in the exact same dev environment? Again, the tech is not Xbox-only, it also works on Nvidia’s cars and likely will at some point work on AMD cards at some point (and maybe even a Switch successor too).
My point wrt CBR/alternatives wasn’t about artifacts. It was about the ceiling for the perf boost being the same 2x. With temporal injection you are injecting at most every other frame. With CBR you are reconstructing at most half the pixels.
1 Like
Isn’t this you saying internal rez of 720p and using CBR to get to 4K?
In the end, I don’t think it matters if we’re ultimately saying the same thing then. If we’re both talking about them shooting for a sub-4K reconstructed image, then I’m not sure why you’re disagreeing with me.
You’re making some assumptions here. We don’t know what resolution or performance they’d target with DML SR since we don’t know how performant it’ll be on consoles. I’ve also explained why I think DML SR may not be widely adopted. If it’s not possible to use some variant of the technology on the PS5, why would devs split their focus on two separate reconstruction techniques when possibly half of their audience won’t use it? Instead it would make more sense, in terms of investment, to focus on a more universally supported reconstruction technique. That’s why I wouldn’t be surprised to see AMD’s SR or some similar technique become more of the standard. A number of engines already supports some form of temporal reconstruction, it only make sense to further invest and refine what is already being done instead of switching gears to something one of the major platforms may not support. You keep coming up with scenarios where only playstation misses out or comes out last when the vast, vast majority of 3rd party studios would not think or design that way.
And my point is I think we should wait until this technology is actually present on consoles before jumping to conclusions or making bold predictions on the type of gap it may produce. For all we know, we could see a similar 2x boost with DML SR but the final image is a higher quality. I don’t think it’s unreasonable to have the opinion that we should wait until we have more info. We can agree to disagree and I’d be happy to be wrong here, but I’m extremely hesitant to assume we’re going to be seeing a 2x performance delta.
That maybe was poorly worded on my part then. I was saying in that comparison the CBR scenario ends up at 4k, not that it has to get there in one step. Instead, it seems more likely devs would forego being in that position in the first place by removing RT from a 60fps mode and keeping it only for 30fps ‘quality’ mode. In such scenarios, you’d have a 60/30fps split b/t the consoles, which is what we were discussing at the start with you being skeptical we would ever see that come about. You were saying they can just turn down the internal res to be able to use RT and still hit 60fps, but you’d have to turn it down the point of literally shading half the pixels and then scaling up to get to the display’s 4k target.
We can expect ML SR on XSX to take about 5ms to do a 4x upscale from 1080p–>4k. That estimate comes from DF (extrapolated from the 2.5ms it takes on a 2060 with twice the TOPs as XSX). And there is no additional investment on the dev side. They aren’t the ones training their SR model, MS is. This isn’t bespoke code written to target specific hw configurations like other graphics code is. There is no additional investment required beyond what is already gonna be commonplace on the PC space, which will be done in the GDK via the same api calls.
I think maybe in your head you see ML SR as being in the same sorta boat as the other upscaling techniques in the sense of requiring devs code to the metal to make good use of them for a given target platform, but that isn’t the case with ML algorithms. There’s no reason at all, for instance, that XSX can’t run literally the exact algorithm DLSS2.0 uses. The algorithm is platform agnostic.
Anyhow, you seem really focused on this idea that if PS5 doesn’t have access to a hw feature, devs will ignore it. That is a really outdated thought process. By that logic, we should never expect to see much use of SFS or mesh shaders either, as if you imagine MS only included these things for 1P teams exclusively despite them telling us these solns came about from talking to 3rd party teams. Xbox games are made in the GDK alongside the PC version, so PS5 is already the odd man out here wrt how api’s are being leveraged.
And no, I am not ‘coming up with scenarios’ to make PS5 look bad. I am simply pointing out that the same type of algorithms work across PC/XSX and are implemented in literally the same dev environment/GDK so why would a dev decide not to leverage an ML soln when it is already there in the PC sku? Just to make PS5 owners happier by gimping the XSX version? I don’t see that happening at all. Nor does it sound feasible to just ‘turn own the resolution’ to erase a huge gap in performance. Much more likely imho is devs add a display mode on PS5 and split RT and 60fps into separate modes, as they have been doing for a while now.
2 Likes
Conversely, I think some are far too dismissive of technology that we have seen implemented in an existing engine that does run on Xbox (ForzaTech and DirectML). Equally, some are dismissing the role the additional hardware on the Series consoles plays in alleviating any additional load required to achieve these results. On a limited timeframe but I wanted to address the point; some of you forget how much Xbox engineers achieved with pisspoor hardware last generation (the BC alone is incredible given the actual gains seen on native hardware).
3 Likes
No one is bullish on anything. Someone saying tech that MS has invested heavily in will be used for performance gains in the future is not “bullish”. ML and SFS WILL be used in the future.
Where have we seen DirectML stuff running on Series S|X? I don’t remember anything.
But we have seen nothing on this front from Microsoft. This will take a lot of time to arrive in games. And the performance gains on this front are unknown. At least for SFS and VRS we have some demos or results from games.
Remember last year we saw pathtraced Minecraft and 2x resolution boost in Gears. And nothing has arrived a year later on consoles in our living rooms. Stuff is hard. What some here expect on the other hand (60fps on Xbox vs 30fps PS5) i would call very bullish.
Never seen this in this thread lol. And yeah, things do take time, but that doesn’t mean they aren’t coming. Regarding the 2x resolution boost in Gears, we do know they are still working on the backward compatible titles. Also, keep in mind these consoles have been out less than a year.
EDIT: also, not to be petty but i find it hilarious that Sony fans can get worked up over an SSD and secret sauce features (that don’t actually exist) yet we’re being bullish for being excited for tech features that actually exist and have been invested in by MS.
(not saying the tech specs of the ps5 ssd don’t exist, referring to the “secret sauce” stuff that was going on a while back)
VRR is a game changer. I think more studios should unlock the framerate.
1 Like
We haven’t, but again, the algorithm it will be running isn’t bespoke to the hw like other graphics code is. It’s just a bunch of numbers (the weights/biases) that use motion vectors to act upon pixel values with given settings in place for the res target. There is nothing, for instance, that would prevent Nvidia’s model from running on XSX (the inference model, not the training process).
Also worth noting that MS confirmed explicitly that they have teams using ML SR on XSX games for texture upscale at runtime and reportedly it works “scarily well”, to the point they are changing how they deliver the games (i.e. shipping low res textures an letting the console upscale them as you play).
If they are able to upscale textures at runtime as of a year ago, there is reason to be bullish. That is a tougher task than what is being discussed here. This is not nearly as ‘unknown’ as you guys are pretending it is. The reason it is taking time is just down to MS wanting to be able to have it be as easy to implement as possible as part of the GDK tool set. That’s also probably why we have not seen Minecraft pathtraced on XSX yet. It was 1080p on XSX and they’d want it to use ML to get to 4k instead (since XSS version would be super low res with path tracing so devs would prefer just holding off). I’d wager Minecraft path tracing edition hits XSS/XSX once they have ML SR implemented well on consoles.
Wrt 60/30, read my posts. If a game on PS5 has a RT fidelity mode and a non-RT perf mode, there is no reason to presume devs making the PC/XSX versions on GDK using DirectX12U are going to implement DirectML SR on PC but not on XSX. The result of that implementation would be an XSX version running 60fps with RT vs 30 fps with RT on PS5.
Nobody in this thread has talked up PS5’s SSD to the level you are suggesting. There are plenty of PS5 fanboys parading around as ‘tech gurus’ who just spout bullshit on the internet to chase clout, but they aren’t posting here.
Was referring to online gaming spaces in general, not this particular thread.
1 Like
How does the Steam Deck pack in Zen2 CPU, RDNA 2 GPU, NVMe SSD in a small mobile form factor? I‘m really confused. Can somebody explain? Whereas the Switch is still using an old Android chipset.
Does this mean we can see smartphones running on Zen 2 and RDNA 2 rather than the mobile SoCs?
It’s not really “small”, portable yes, but not small.
1 Like
I hope the next Samsung devices use this desktop grade chipsets and beat Apple SoCs.
To compare it to the Series S, the SteamDeck is 1/2 the CPU and 2/5ths the GPU. Theres also power saving from memory too. They’ve also had 12 to 15 months of process improvements to leverage.
So scale power usage of a SeriesS down to 40% as a rough idea. Series S peak draw of 82.5 Watts in Gears 5, scaled to 40% and you’re at 33 Watts.
The SteamDeck has 2 to 8 hour battery life while packing a 40Whr battery.
1 Like
It looks like it also throttles (not sure on the conditions) quite a bit:
AMD APU
CPU: Zen 2 4c/8t, 2.4-3.5GHz (up to 448 GFlops FP32)
GPU: 8 RDNA 2 CUs, 1.0-1.6GHz (up to 1.6 TFlops FP32)
APU power: 4-15W
Still claiming only up to 15W for just the APU.
1 Like
I still think there are issues with the scenario you’re describing. What happens to the teams and engines where RT is a core focus going forward? I also can’t believe I haven’t thought of this yet, but what you’re saying here would also require games to run twice as fast on the CPU. While the CPU is more capable on the Series X, it’s not 2x more capable. As games grow more complex over this generation, there will be even less headroom on the CPU side. So I guess IF we did see this scenario, maybe it would be with an indie game that is less taxing on the CPU side? To be fair, we did have that one indie developer say that the PS5 version of their game wouldn’t have a 120fps mode. So I guess in a way you are already right. 
I understand where you got the figures from but I think you’re underestimating the work necessary to get DLSS into a game. You can’t just take the algorithm and plop it in the code and call it a day. To get DLSS into a game, it takes a collaborative effort between the developers and NVIDIA. Aspects of the engine needs to be considered and tuned to make sure it works well with something like DLSS.
No, I do not think devs will ignore Xbox specific features. You and I have had conversations in the past on how SFS will almost be a necessity on Xbox due to it having a slower SSD. Before anyone jumps on me about that, this isn’t meant to be a dig on Xbox or their design. SFS is a core part of taking advantage of both memory management and SSD utilization, as we’ve discussed in the past. Same goes for mesh shaders. That will most definitely be taken advantage of. I’m not taking things to extremes where it’s an all or nothing approach and neither of those features are proposed to give a 2x performance boost over the competition. Plus there are features on the PS5 that developers can use to at least help offset those advantages, even if it may be just a little bit.
I’m also not sure it’s safe to assume that every PC developer will opt to use the GDK for the PC SKU. On the outside does that make sense? Yeah, I definitely think it does, but we have we have no information on how widely it’s being used so far and honestly I’d love to figure out a way to get that information. I’d also prefer not to go down the rabbit hole of whether or not the PS5 is going to hold this generation back. Neither of us are developers nor do we have access to the code to even make that assessment.
Again all I’m saying is I think we should wait and see how things play out. You want to assume we’re going to see 2x performance on the Series X, great, again I hope I’m wrong, but I prefer to take a more cautious approach to such predictions, especially when the GPU is only one piece of the equation.
Assuming you’re talking about me since I’m the only one conversing with Tavish, there’s a difference between not leaning into predictions of a 2x delta and dismissing the tech in the system. I’ve already said that the additional features are going to increase the gap, multiple times. It shouldn’t be controversial for someone to be a bit more reserved in how large that gap will be. I’m not being negative, dismissive, or downplaying what the engineers at Xbox have accomplished just because I find it hard to believe we’re going to see a 30/60fps split.
Edit: To add I do agree with you 1000% on the engineering side. I firmly think Xbox has the best hardware engineers of the three hardware manufacturers (ironic considering how often Microsoft is solely described as a software company) and especially agree about BC. I’ve always been hugely impressed with that accomplishment and why I’ve made two videos covering it. Unfortunately, I’m talking about 3rd party developers and what they may or may not do and not Microsoft’s engineers directly.
1 Like