Xbox Series S [XBSS] (Lockhart) General Rumors and Speculation *spawn*

Status
Not open for further replies.
It not being able to run 1X profile in BC would be a huge blow for me.
There's many games that are hard limited to 900p on XO for example, and don't have 60fps modes.

I guess if the resolution doubling works for Lockhart for XO games, then maybe it won't be so bad if it works for majority of games. May even be able to more than double it.

Also they could easily update the emulator for OG xbox and X360 so it still gets the enhancements.

If this is the case then all is ok with the world again.
 
Last edited:
Might not be there on launch, but the GPU grunt is there.

I suspect 6TF would have been HW BC out of the box. 4TF RDNA2 means some work needed by the BC team.
The gpu I don't see as the problem, it would be the memory.

Also been thinking, it wound be nice if the api's wasn't based on the console.
Be based on power i.e. 1x, 4x, 8x and then base resolution on that, then in the future it would get automatic resolution boost with mid gen of 16x etc. Also would map power to generation of feature sets.
Would especially work well with things like dynamic resolution.
 
I think that although Lockhart won't run X1 games in X1X enhanced profile (it actually doesn't need to since this is a machine designed for~1080p) I would wager that the BC team will devise methods to enhance X1 games on the system. Naturally, games with dynamic res will run at their max; framerates will be constant and I am sure they will increase texture filtering. I would also assume that it will be capable of the new Auto HDR since that is in the display controller. I won't be surprised if they are able to create a Lockhart-Enhanced mode for X1 games, maybe increasing the base res to 1080p for games that render at lower res.
I think BC enhancement will be an area of constant innovation for the BC team at Xbox.
 
It's powerful enough to do higher than 1080p when running current and previous gen games, it can then either down sample or look better on a higher resolution screen. 1440p or 4k.
Doesn't have to be limited to 1080p, the console can also be for people who have 4k screens that don't game enough to justify buying the higher end model. Or 1440p monitors like they added support for.

1X profile isn't just 4K, its 60fps modes also.
Dynamic resolution doesn't mean it tops out at the best the console can do, it has a max resolution, a lot of if the time its 900p for XO.

Now if they can double (or more resolution) in enough games, and double framerate then it's not an issue. 9 times out of 10 that's the only real benefit to 1X, the extra memory regarding higher textures didn't make a huge difference to me.

The thing is they've not given any indication of the amount of games their able to make those enhancements on. May just be a few.

I had my idea what Lockhart should be capable of a long time before they started talking about resolution and framerate doubling etc, so it's only now that I've considered that it may not need 1X profile to run current gen games decently.
As just running XO at locked framerate is a waste in my view when it has the equivalent of 1X performance for that gen games.
 
I don't think Lockhart is the intended upgrade step for X1 owners, so it won't really matter that it can't run X1 versions of games. I mean if you're going for a budget console that can't push 4K output, and is closer to 1080p, you probably weren't someone who was too concerned with X1 resolution bumps to begin with. It might be nice to have, but not so nice as a very low retail price.

MS's up-resing BC antics so far have always been based around integer multiples of the original, by the same number in both axis. None integer scaling would probably break things as you'd have to start making decisions about how you biased sampling, and how you blended to turn it into super sampling. As Jay points out, if they could do a BC mode that just doubled up one axis then used that for super sampling, that might be worthwhile in most cases.

I agree that there are BC tricks that could make even lower native resolutions (e.g. 900p) look a lot better. Machine learning HDR for example, or the idea they teased about doubling frame rates on some games (not sure how you'd do if but if anyone can it's probably MS's BC team). Maybe there's other stuff they could do, like using several free TF to force additional post process anti aliasing, or alternating between vertical and horizontal upres, and using all the extra memory to construct a pseudo 2x2 buffer. Plus there's the usual benefits (16x aniso for the win).

In terms of next gen, something that's occurred to me is that with smart delivery you might see Lockhart games deployed in smaller packages (drop those highest level 8K texture LODs) you might get more games on the SSD. That could suit those on a budget as there'd be less pressure to buy an expensive expansion SSD. Would be nice for Gamepass too.

If it's cheap, and there's an optical version (doubtful) I might make one my first console in a decade just for casual comfy couch gaming and BC. I still haven't got a 4K display in my house, and if I'm dumping a load on hardware it's probably going to be on my PC, which I pretend to use for productivity.
 
In terms of next gen, something that's occurred to me is that with smart delivery you might see Lockhart games deployed in smaller packages (drop those highest level 8K texture LODs) you might get more games on the SSD. That could suit those on a budget as there'd be less pressure to buy an expensive expansion SSD. Would be nice for Gamepass too.

SeriesX hardware hits xcloud early next year. It'd be interesting if this changes the download habits of Lockhart customers (and SeriesX user as well really). What's the weekly playtime for the 10 most played games on people hard drives at the moment? Bet no.4 downwards barely gets a look in for most players. Knowing you can play any game in your library in a reasonable fashion instantly may mean that those stored locally end up as few favourites + a current flavour of the month rotating in.
 
Here's how Micrososft describes how BC works on Xbox Series X:

https://news.xbox.com/en-us/2020/05/28/xbox-series-x-next-generation-backward-compatibility/

Not only should gamers be able to play all of these games from the past, but they should play better than ever before. Backwards compatible games run natively on the Xbox Series X hardware, running with the full power of the CPU, GPU and the SSD. No boost mode, no downclocking, the full power of the Xbox Series X for each and every backward compatible game. This means that all titles run at the peak performance that they were originally designed for, many times even higher performance than the games saw on their original launch platform, resulting in higher and more steady framerates and rendering at their maximum resolution and visual quality.


Interepeting that is kind of confusing but the debate over whether Lockhart has a X1X BC mode may not even be relevant to how BC even works. It sounds more generalized than that.
 
I don't think Lockhart is the intended upgrade step for X1 owners, so it won't really matter that it can't run X1 versions of games. I mean if you're going for a budget console that can't push 4K output, and is closer to 1080p, you probably weren't someone who was too concerned with X1 resolution bumps to begin with. It might be nice to have, but not so nice as a very low retail price.
From what we know the expected resolution is <=1440p. Even if its just 1080p playing current gen at 900p that would be a shame given the power to do a lot more.
MS's up-resing BC antics so far have always been based around integer multiples of the original, by the same number in both axis. None integer scaling would probably break things as you'd have to start making decisions about how you biased sampling, and how you blended to turn it into super sampling. As Jay points out, if they could do a BC mode that just doubled up one axis then used that for super sampling, that might be worthwhile in most cases.
i agree and I reckon it has the power to double the resolution, double of XO would be more than adequate. And is something that I had not factored in before, and makes it an interesting product to me again. Although we don't currently know how many and which games yet though.

And talking from a general perspective and not just my own now, I think it would be a huge selling point to people, even current XO owners if they can play current XO games at a lot better fidelity. XO games can look very bad even on 1080p sets.

I agree that there are BC tricks that could make even lower native resolutions (e.g. 900p) look a lot better. Machine learning HDR for example, or the idea they teased about doubling frame rates on some games (not sure how you'd do if but if anyone can it's probably MS's BC team). Maybe there's other stuff they could do, like using several free TF to force additional post process anti aliasing, or alternating between vertical and horizontal upres, and using all the extra memory to construct a pseudo 2x2 buffer. Plus there's the usual benefits (16x aniso for the win).
Yea, I'm hopeful that running XO profile with enough work will look and play pretty well on Lockhart, even without running 1X profile.
What their saying about BC gives me hope about that, as I said I never took that into account before.
Hoping the BC team is working magic on the whole Xbox Series not just the X.

In terms of next gen, something that's occurred to me is that with smart delivery you might see Lockhart games deployed in smaller packages (drop those highest level 8K texture LODs) you might get more games on the SSD. That could suit those on a budget as there'd be less pressure to buy an expensive expansion SSD. Would be nice for Gamepass too.
Yea I always expected that the assets/textures would be smaller on Lockhart, and so having a smaller ssd may only store a couple less games than on xsx.
Unsure how much ssd space that would actually save though to be honest, it's more beneficial for the lower gddr amount.
If it's cheap, and there's an optical version (doubtful) I might make one my first console in a decade just for casual comfy couch gaming and BC. I still haven't got a 4K display in my house, and if I'm dumping a load on hardware it's probably going to be on my PC, which I pretend to use for productivity.
what would be cheap enough for you to do that and would gamepass factor in?

I think its a good device for more than just the XO owner. But also for:
  • Second console if PS is primary format.
  • Second room
  • Casual gamer
  • 1X owner who wants to play next gen, but given world economy may not be able to justify xsx
  • PC gamer secondary supplementary device.
  • How did I forget parent buying kids a console
 
Last edited:
What if you use MS DirectML on Lockhart to increase the 768p display image to 1440p, or even the 1080p to true 4K image quality? Achievable?
 
If it's cheap, and there's an optical version (doubtful) I might make one my first console in a decade just for casual comfy couch gaming and BC. I still haven't got a 4K display in my house, and if I'm dumping a load on hardware it's probably going to be on my PC, which I pretend to use for productivity.
Yeah, I'm more keen on investing in a high-end procrastination rig than a console.
 
Interepeting that is kind of confusing but the debate over whether Lockhart has a X1X BC mode may not even be relevant to how BC even works. It sounds more generalized than that.
Guess it is open to interpretation so I'll not say mine is right or wrong.
BC games format:
  • OG Xbox
  • X360
  • XO
  • 1X
So in that context XO and 1X profile matters.
I expect the xsx to be able to play the 1X version of the game and any further enhancements to be based on that version.
If memory reports are correct then Lockhart downloads XO version.
When you currently download a game it is based on your console/profile.
 
Last edited:
Actually, if Lockhart and Anaconda really can treat 100GB of SSD space as virtual memory, then the 7.5GGB might not be a big deal. If you've already worked out what can be treated as pageable (like most assets, audio, menus) then the same will be true for both Anaconda and Lockhart.

You create some rules and let the system page data in, on the fly, hundreds of times a frame as you move around the environment.

Lockhart might have a lot less memory but I'm guessing what matters is that 7.5 GB is enough for "core" data like game code and gameplay influencing physics, and also for immediate minimum level LODS etc. And as Lockhart is proportionately more likely to be drawing less detailed frames (lower lods, less texture data etc) they can possibly lean on the virtual memory as an extension of regular memory even more heavily than Anaconda. E.g., for an area of the screen that might need four 64 x 64 texture tiles to be transferred in at 4K, 1080p might need only 1, meaning you have more available virtual memory transfers left to support something else.

I'm really starting to think MS have very heavily profiled the bejesus out of games to arrive at their dram / SSD bw balance for these machines. Handing off a lot of the juggling work to a virtual memory / paging system would strike me as a good way to mitigate the increased dependency on very frequent, small transfers into a relatively small pool of dram.

It might also allow more aspects of developing for two tiers of next gen system to be automated. I.e. for *some* aspects of the game you can just forget about the physical dram differences.
 
What if you use MS DirectML on Lockhart to increase the 768p display image to 1440p, or even the 1080p to true 4K image quality? Achievable?
Not enough power for DLSS. There’s not enough for Anaconda either I don’t think. I mean, you can do it using compute, nvidia did do this for a little while IIRC. But the results pale to their DLSS 2.0 solution running on their tensor hardware.

If you want better quality you’re going to need to crush a larger network faster
 
Actually, if Lockhart and Anaconda really can treat 100GB of SSD space as virtual memory, then the 7.5GGB might not be a big deal. If you've already worked out what can be treated as pageable (like most assets, audio, menus) then the same will be true for both Anaconda and Lockhart.

You create some rules and let the system page data in, on the fly, hundreds of times a frame as you move around the environment.

Lockhart might have a lot less memory but I'm guessing what matters is that 7.5 GB is enough for "core" data like game code and gameplay influencing physics, and also for immediate minimum level LODS etc. And as Lockhart is proportionately more likely to be drawing less detailed frames (lower lods, less texture data etc) they can possibly lean on the virtual memory as an extension of regular memory even more heavily than Anaconda. E.g., for an area of the screen that might need four 64 x 64 texture tiles to be transferred in at 4K, 1080p might need only 1, meaning you have more available virtual memory transfers left to support something else.

I'm really starting to think MS have very heavily profiled the bejesus out of games to arrive at their dram / SSD bw balance for these machines. Handing off a lot of the juggling work to a virtual memory / paging system would strike me as a good way to mitigate the increased dependency on very frequent, small transfers into a relatively small pool of dram.

It might also allow more aspects of developing for two tiers of next gen system to be automated. I.e. for *some* aspects of the game you can just forget about the physical dram differences.
Profiled and simulated. There is no doubt they should know exactly how well it will perform without needing to burn like they did with Scorpio.

If Lockhart is green lighted, I expect it to perform within expectations of what they will market it as, as will Series X.

so I guess it’s just a question of what Lockhart Performance targets, features and price points are.
 
Wouldn't be suprised if Lockhart development mostly consists of an auto-generating a derivative of the XSX build with some tool Microsoft developed. So yeah heavily automated.
 
And talking from a general perspective and not just my own now, I think it would be a huge selling point to people, even current XO owners if they can play current XO games at a lot better fidelity. XO games can look very bad even on 1080p sets.

Yeah, low res X1 games can look a bit rough especially on large sets. I think there more to it that just resolutions sometimes - shadows, texture filtering etc. For games that were really hurt by the small amount of esram and lost a lot of efficiency when using ddr3 for buffers, perhaps the unified memory will be enough to allow a 2x2 increase in everything (hopefully shadow buffers too!).

Perhaps the best option is to allow users to enable some BC enhanced features on a game by game basis, and at their own risk in terms of performance.

Hoping the BC team is working magic on the whole Xbox Series not just the X.

I think if they can get them to run okay, they'll be keen to bring some version of all their enhancement features to both levels of device.

what would be cheap enough for you to do that and would gamepass factor in?

With an optical drive, I'd probably pay up to £180 ish I guess? I do think Gamepass is great value. On PC everything's digital now, and you don't really own those purchases in the same way that you do physical console games. It's made it easier for me to accept a subscription model I think. Damn. I keep forgetting about GoG. You actually "own" those, need to stop being a lazy Steam browsing bum. :(

I think its a good device for more than just the XO owner. But also for:
  • Second console if PS is primary format.
  • Second room
  • Casual gamer
  • 1X owner who wants to play next gen, but given world economy may not be able to justify xsx
  • PC gamer secondary supplementary device.
  • How did I forget parent buying kids a console

Yeah, I think it'd be a good fit for all those. Also a nice little streaming box in addition to its primary uses. My 360 is mostly used for streaming services now, but MS aren't going to support that old thing forever ....

What if you use MS DirectML on Lockhart to increase the 768p display image to 1440p, or even the 1080p to true 4K image quality? Achievable?

Hard to say, Nvidia use very powerful Tensor cores to do that, and I have no idea how it would fare on regular compute shaders. Tensor cores seem to support a range of int and float formats, and while the XSX does support accelerated int8 and int4 rates I don't know if these could be used or if they'd be fast even if they could.

On the one hand in a basic BC mode the next gen consoles should have a lot of power left over, but on the other hand if regular compute shaders could do this fast enough, I don't see why Nvidia would have stuffed their GPUs with Tensors. Though over time techniques do tend to become better, faster and less expensive.

So ... basically ... I have no idea. :runaway:
 
Hard to say, Nvidia use very powerful Tensor cores to do that, and I have no idea how it would fare on regular compute shaders

I hold out hope for a cheap compute solution. I'm basing this hope purely on Nvidia's history of needlessly resource hogging features. Hairworks, grass works etc. :D
 
I hold out hope for a cheap compute solution. I'm basing this hope purely on Nvidia's history of needlessly resource hogging features. Hairworks, grass works etc. :D
There is compute upres. There are simple upres demos using DirectML on a intel iGPU. As long as you have very high FP16 or lower the performance should be OK. But that’s not to say you are getting DLSS2.0 which is more than just up resolution.

tensors just have a lot more to power to give. From a pure silicon perspective, I think they went tensors because it’s going to provide the biggest bang for buck without needing to further increase bandwidth requirements even more.

things may change moving to HBM. But at this point we are locked.

someone ie MS still has to make the models for it. And the developers need to implement it
 
Last edited:
Status
Not open for further replies.
Back
Top