Digital Foundry Article Technical Discussion [2021]

Status
Not open for further replies.
Xess doesn't need to use Nvidia tensor cores, it just needs to be performant enough to compete well enough to make it the default go to as it runs on broader range of cards.

Using SM6.4(think it is) dp4a should be ok so doesn't matter if its not using direct ml.

I would expect Intel reference dp4a implementation to also just work on XS consoles which may also help take up.

If xess overheard is only determined by output resolution like dlss then on XSS would be better to aim for 1080p-1440p(preferably upper end)
It does need to use tensor cores, Rich explains why in the video. XeSS should use every hardware it runs on to its full potential, otherwise a ton of performance is wasted and DLSS stays the much better option for Nvidia cards. This is not the one does it all software we hoped for.
 
I get it. But given your position as a public face of a notable outlet, you *really* should learn to be able to defend yourself in a more....subdued manner. Even if you're not wrong about your judgements of them, lashing out at viewers looks bad.

Let us handle that sort of thing. :p
Even Peter Hines had to step in.

PF3NnWg.jpg
 
While this might be controversial to say, I believe many of these features were added with S in mind. The Series X can pretty much brute force it but the Series S is where these features are supposed to shine and in the end help the Series X down the road. Currently this generation I find the series S to be the most interesting console because I think Microsoft gambled that all of the efficiency features would help it hold it's own throughout the entirety this generation.
Does it hold though? It is interesting that even the One X outperforms it in situations.
Also how well is it doing in terms of sales? It is the one console from this generation you can find on shelves.
 
Does it hold though? It is interesting that even the One X outperforms it in situations.
Also how well is it doing in terms of sales? It is the one console from this generation you can find on shelves.

I think it does, the recent games like The Ascent and MFS shows that series s offers much better experience and this will only improve when new games will finally take advantage of features like SFS and so on.
The sales argument is valid thou, the majority of consumers right now preferes XSX over XSS and rather wait for X than buy S now.
The S is still very good option for families where more than one console is needed (ie when you have kids etc), kids dont care about lower res. And when finally last gen will get left behind S will be even more attractive.
 
While this might be controversial to say, I believe many of these features were added with S in mind. The Series X can pretty much brute force it but the Series S is where these features are supposed to shine and in the end help the Series X down the road. Currently this generation I find the series S to be the most interesting console because I think Microsoft gambled that all of the efficiency features would help it hold it's own throughout the entirety this generation.
16GB isn't remotely a 'brute force' RAM solution for a new generation of games. If they wanted to brute force it, they would have needed at least 24, maybe 32GB of RAM. SFS is going to be critical in allowing developers to really get the most out of the Series X.

Equally, if Series X is just brute forcing resolution, then devs are not gonna be able to really push the system as it will have much less overhead than if they used AI upsampling.

If it's only used to 'save' the Series S, then the Series X will never be fully utilized.

There is no future in which I think this will be the case. SFS and AI upsampling is too much of a game changer to be wasted on trying to prop up a weak system meant for light/casual gamers. Especially when they've already just spent a whole generation getting beat up by Playstation on a technical front and clearly put in a bunch of effort to push the specs of the Series X. I mean think about it, why would they have plans for these features and then 'brute force' things for Series X? They could have built a somewhat less powerful system(but still a fair bit more powerful than Series S) and saved money and then used these same techniques to make up the difference. Makes no sense.
 
Last edited:
16GB isn't remotely a 'brute force' RAM solution for a new generation of games. If they wanted to brute force it, they would have needed at least 24, maybe 32GB of RAM. SFS is going to be critical in allowing developers to really get the most out of the Series X.

Equally, if Series X is just brute forcing resolution, then devs are not gonna be able to really push the system as it will have much less overhead than if they used AI upsampling.

If it's only used to 'save' the Series S, then the Series X will never be fully utilized.

There is no future in which I think this will be the case. SFS and AI upsampling is too much of a game changer to be wasted on trying to prop up a weak system meant for light/casual gamers. Especially when they've already just spent a whole generation getting beat up by Playstation on a technical front and clearly put in a bunch of effort to push the specs of the Series X. I mean think about it, why would they have plans for these features and then 'brute force' things for Series X? They could have built a somewhat less powerful system(but still a fair bit more powerful than Series S) and saved money and then used these same techniques to make up the difference. Makes no sense.
But don't forget if SF(S) is also supported on PCs, the PCs have the big advantage of having just much more memory for caching (and other) stuff.
 
But don't forget if SF(S) is also supported on PCs, the PCs have the big advantage of having just much more memory for caching (and other) stuff.
Well sure, but I'm talking about Series S vs Series X. And the idea that techniques like AI upsampling and SFS will be what allows the Series S to get away with its on-paper lackluster memory setup as many have argued.

My point is that you cant have it both ways. These techniques cannot 'save' the Series S unless it's at the detriment of what developers do with Series X. Alternatively, developers can use these techniques to push the Series X harder, which will require more significant sacrifices for any Series S version. Which makes a lot more sense to me, given the target audience for Series S. And the fact that Microsoft didn't go through the effort of making a console as powerful as the Series X just to limit what devs can do with it thanks to the existence of the Series S. That would be shooting themselves in the foot.

Basically, my prediction is that Series S will suffer in the long run and these techniques will not save it. The more devs want to be able to do with the Series X, the more that Series S will require notable compromises and cutbacks. Something MS might have alleviated if they'd just gone with 12GB of RAM and a 192-bit bus. But since they wanted to hit that magic low price point, they've created a system that is hamstrung and will likely age poorly. I really think people underestimate the inevitability of developers and their desire(and perceived necessity) to push their ambitions and create the most impressive games possible. They will not let the Series S hold them back.
 
Something MS might have alleviated if they'd just gone with 12GB of RAM and a 192-bit bus.
As games mature into the technology, it's likely this won't be a factor. A general lack of compute power will be the overall bottleneck here for Series S if you want to speak to scaling issues vs Series X in the long term.

You are unlikely to cap out the memory on Series S when it's working in 1080p with 8GB of memory in today's titles. Factoring in things like SFS and upscaling where your buffer sizes are reduced further, it's unlikely to see any 'footprint' issues on Series S. Bandwidth and compute will be the main limiting factors for it, but reducing resolution there will help immensely.

SFS won't be needed for Series S from a footprint perspective (as any regular texture streaming would likely suffice, and games just moving to relying on JIT SSD texture retrieval will be a large impact already) and if games take on upscaling algorithms in which it can create a fairly good 1080p image with reconstruction, than that system should be sufficient in keeping up with Series X.

The reality is that both PS5 and XSX will be targeting ~4Kish with or without reconstruction for the whole generation. You will never see either console releasing something at 1080p reconstruction, so this idea that developers will continually scale down to Series S level is fairly unlikely and not a reasonable edge case to discuss.

Series S will be a fairly good stop-gap for gamers who do not feel the need to move to 4K screens (and this is likely the last generation to do so), but PS5 and XSX are meant to support the mainstream movement into 4K. These consoles are not likely to drop out of their 1440p-4K range, and Series S is not likely to drop out of it's 720p-1440p range.

All the challenges you see with Series S today really just stems from engines that are still built around older base technologies.

You are welcome to blast Series S to the sky, but if history is any teacher here; Xbox One started with all games running at 720p. CoD Ghost and Battlefield 4. Many predicted the worst, but with Halo Infinite and Forza 5 releasing this year on XBO, the game looks stunningly better and the hardware hasn't changed at all frankly speaking. Both of these titles (and Gears 5) will likely be the best looking titles for the Xbox One era.
 
Last edited:
I don't believe XSS needs 'saving'.
Not saying it's a walk in the park, just not earth ending.

Some features benefit one machine more than the other.
VRS benefits and will have bigger affect on XSX more than XSS.
SFS benefits XSS more than XSX due to memory capacities and targets.
ML Upsampling benefits XSS more than XSX as scaling from 720p to 1080p will look worse than 1080p to 4K.
Just because something has a relatively bigger benefit on one machine doesn't negate its benefits on the other though.

I don't see XSS needing to be a stop gap either, there's plenty of reasons people may choose the XSS and be more than happy.
The only complaint I've heard is storage space, but if I'm right and before ever thinking of doing price cuts they add value first, by making it 756GB or even 1TB then that will solve that.
XSX moving to 1.25TB or 1.5TB respectively.

Edit: Forgot it wasn't soldered on. In that case may need to go straight to 1TB.
 
Last edited:
And yet you have Valve SteamDeck releasing which everyone is excited for that imposes it's own limitations.
Steam Deck isn't a 'next gen' gaming console. It's not its own unique platform that will direct the ambitions of AAA developers. These consoles are and will.

Anybody who thinks they're gonna be playing the latest games from 2026 on the Steam Deck will be in for a rude awakening. But the Series S very much *needs* to be able to do this.

Entirely different situations.
 
You will never see either console releasing something at 1080p reconstruction
We already have! lol

That's exactly what Ratchet and Clank Rift Apart is doing!

Even as reconstruction techniques are now, this will be quite viable in order to offer the max amount of overhead to push visuals/scope, but reconstruction techniques will only get better. UE5's temporal upscaling is looking amazing and I'm sure Microsoft will have something good themselves. There will be little need to push anything notably beyond 1080p as a baseline resolution with great reconstruction. Especially with the fact that most people using consoles are doing so in a living room with a TV from a distance.

This is the future. So much so that in five years, I'd bet most every major game will be built with the expectation that reconstruction is used. We already have something like that now in the case of Minecraft RTX, and this will only be the beginning.

Also, Xbox One did not start out with ALL games running at 720p. This is wildly false. And the expected problems with base XB1 versions of games absolutely did come to fruition in the past few years. Hardware matters.
 
VRS benefits and will have bigger affect on XSX more than XSS.
SFS benefits XSS more than XSX due to memory capacities and targets.
ML Upsampling benefits XSS more than XSX as scaling from 720p to 1080p will look worse than 1080p to 4K.
Just because something has a relatively bigger benefit on one machine doesn't negate its benefits on the other though.
Why would SFS benefit the Series S more than the Series X? I've literally just made several long posts detailing why this isn't the case. At least explain why I'm wrong instead of just saying so.

As for 720p->1080p looking worse than 1080p->4k, you missed my point completely. If Series X is doing 1080p->4k, then you will NOT be able to just do 720p->1080p on a Series S. The hardware gap is much bigger than that. We'd be talking more like 576p->1080p or something. The sacrifices are gonna be bigger than people think. And for memory concerns, will probably need noticeable downgrades beyond just easily scalable aspects like resolution.

The whole point is that the more devs push Series X, the more Series S is gonna suffer. And this means utilizing techniques like SFS and AI upsampling. Series S doesn't 'benefit more' from these if both are using them. It's largely the same discrepancy as if both systems weren't using either. The RELATIVE differences between them are critical and exist no matter what. The only way these techniques help Series S is if devs dont actually use them to push the Series X.
 
16GB isn't remotely a 'brute force' RAM solution for a new generation of games.
If the maximum amount of RAM that consoles are going to have this gen is 16 gigs of RAM, I stand by my brute force comment.
Well sure, but I'm talking about Series S vs Series X. And the idea that techniques like AI upsampling and SFS will be what allows the Series S to get away with its on-paper lackluster memory setup as many have argued.

My point is that you cant have it both ways. These techniques cannot 'save' the Series S unless it's at the detriment of what developers do with Series X.
Well if the developer gets memory savings on the Series S using SFS and AL Upscaling and just decides to call it a day without using it on the X or PC then you do have a point. Though SFS does work on both the X, S, and I assume on the PC. So yes, the developer can use SFS and AI upscaling on all 3 save some memory and call it a day... or they can take those memory savings and use them to make a better looking and playing game. SFS in particular, if it is something that can be implemented in the real world, will not only help the S it will help the X and PC. Not all devs will take advantage of it and very few if any have taken advantage of it in the games that have been release so far but the bet is that a studio like Id or the Coalition would. The easier SFS is to implement the greater the number of studios that will likely make use of it.

Same thing with the savings from AI upscaling. Just to be clear, if the scaler is really good, I'm not talking about gamers turning the feature on and off themselves. I'm talking about using it like games use checkerboard rendering. The dev decides if it looks good enough. Though that does not stop them from letting gamers turn the feature off or on if they would like to but I think there might be cases that if you want high quality raytraced reflections or a rock solid 120 fps it might be turned on by default across both the X and the S. PC gamers I'm sure would still get their toggle one way or another.
 
Last edited:
And yet you have Valve SteamDeck releasing which everyone is excited for that imposes it's own limitations.
No new game must support the Steam Deck at launch for the next few years. I haven't heard of Valve imposing support on publishers and devs who want to sell new games on Steam.

The Steam Deck's excitement derives from the limitless number of current PC games it supports. Valve mentions some future proofing at <= 1280×800p30, but the fact that they're already commenting on future hardware helps keeping its PC DNA (i.e. its short life) in people's minds.
 
Why would SFS benefit the Series S more than the Series X? I've literally just made several long posts detailing why this isn't the case. At least explain why I'm wrong instead of just saying so.
I said I disagree with the whole premise that the XSS needs saving.
If both never had SFS I also think XSS would be fine.
So I never at any point presented it as a way for it to save it as it doesn't need saving. I was discussing the relative merits for each feature for each device as I found that interesting.

I could repeat and explain why SFS has bigger relative benefit for XSS same for why VRS has a bigger one for XSX, but I think you read my post out of context or as a response to your view, hopefully this helped.
But I will say in response to you that the XSS will benefit more from ML Upscaling as the quality of the final image will be better compared to other methods due to the initial lower amount of data to work with, this is where the benefit of ML comes in. XSX even using other upscaling methods will have a better final image, therefore more beneficial for XSS comparatively.

Probably my fault, I should've been more clear.
As I personally don't find your line of discussion or the rebuttal of it intresting, due to faulty initial premise imo.
 
Last edited:
We already have! lol

That's exactly what Ratchet and Clank Rift Apart is doing!
It reconstructs from 1080p is not an issue as reconstructing to 1080p.

As for 720p->1080p looking worse than 1080p->4k, you missed my point completely. If Series X is doing 1080p->4k, then you will NOT be able to just do 720p->1080p on a Series S. The hardware gap is much bigger than that. We'd be talking more like 576p->1080p or something.
Why? This seems completely false. You're just throwing some numbers in the air saying it can't. It's not linear power requirements to move from 1080p to 4K. It takes approximately 3x+ the compute to move from 1080p to 4K even though the pixel density has increased 4x. We have tons of benchmarks on the PC side that showcase that you don't need 4x the power to obtain 4x the resolution.

720p to 1080p is exactly 2x the pixel density and 1.5x along the axes. Hovering around 720p for reconstruction is a reasonable benchmark for it. And ultimately what does it really matter here what the base resolution is if the upscaling algorithm is doing a good job?

I just need to be clear here, your argument is that 10GB of VRAM is incapable of rendering 720p if a 16GB card is rendering from 1080p to 4K? And you are claiming that for Series S to keep up with Series X that it will need to downgrade to 576p because it doesn't have 12GB of VRAM? Seriously? Everyone should start throwing away all their 3000 series nvidia cards today.
 
Status
Not open for further replies.
Back
Top