Support NeoGAF

[eastwood333]

The guy is not a part of the Xbox team, hasn't been for some time.. and when he was he was in charge of the Backwards Compatability program.

So no this guy has not discussed next-gen I/O speeds with his Hololens project team lol

1.I'm talking about Thomas Mauler, head of Moon Studios. Not sure what you're talking about.

 

[IntentionalPun]

1.I'm talking about Thomas Mauler, head of Moon Studios. Not sure what you're talking about.

Uhh.. the thread topic?

Microsoft’s William Stillwell.

 

[eastwood333]

The guy is not a part of the Xbox team, hasn't been for some time.. and when he was he was in charge of the Backwards Compatability program.

So no this guy has not discussed next-gen I/O speeds with his Hololens project team lol

And no it is not common sense that devs will design around insanely fast I/O speeds 100 x's faster than last gen all the time. Very few games are likely to actually require that level of I/O for anything. We'll mostly be getting better load times on PS5 and the same damn games.. that's common sense for a number of reasons.

Beyond that games that do push that far are probably far more likely to lower detail on the lesser machine than design a fucking elevator into one version of a game lol

I'm actually not disagreeing with what you're saying.

I thought u guys were talking about Thomas Mauler.

 

[Tulipanzo]

Going to be interesting how things shape up, because we definitely didn't have much in terms of balancing systems with different i/o speeds AFAIK.
Probably not going to get extra elevators, but it'd be pretty funny ngl.

Squeezethroughs in Halo Infinite are 3x as long depending on your hardware.

 

[phil_t98]

Witcher 3 HD Reworked is 7GB in size.
Mostly 2x textures of the vanilla.

so in terms of that its hardly anything spread across the game. deffo gonna be interesting to see how this all plays out

 

[Mister Wolf]

So basically the Xbox Series X is holding back next-gen games.

More like PC because of the storage speeds of most people are SATA SSD with alot still using HDD. It's weird though because although you can say PC is holding it back because of storage those same PCs will still be running the games at better settings, framerates, and resolutions than PS5 if they have better GPUs and CPUs.

 

[PharaoTutAnchAmun]

Dunno about that. It comes down to the engine. UE5 seems to have made optimizations to attempt both leveraging PS5's SSD I/O in particular (but not at the expense of other systems), AND having targeted performance that can be scaled across different I/O implementations.

So I'd assume devs using that engine, even for multiplats, should be able to get very good use of PS5's SSD I/O without needing to do too much work or compromise. So they can still design the game as they want, and target strengths of the PS5 architecture (and for XSX, target strengths of XSX's architecture). Meaning we'll be able to look forward to more than just 1st-party games making fuller use of these systems...if they're using engines like UE5



Hate to break it to 'ya but we aren't getting anything near the shift from 2D to 3D gaming here. In fact almost all examples I've seen in terms of trying to state PS5's SSD I/O advantages have amount to basically higher polygon assets and texture streaming, but graphics-related.

I don't just think of prettier graphics when I think of fundamental game design innovations. They're a part of the equation, but not all there is to it. So I think a lot of people are underselling the benefits of PS5's SSD I/O in that respect, and they're also ignoring other features that play a big role in game design shifts like CPU innovations, GPU breakthroughs (ML, upscaling, AI neural network capabilities, asynchronous programming etc.) and more, some of those being things XSX happens to have the advantage in.

It'll take leveraging all of these things in tandem to get big shifts, plus the market demanding for things aside from just prettier graphics for a change.

You forgot the Tempest engine for 3D sounds

 

[DunDunDunpachi]

Why would they need to address the advantage if it isn't much of an advantage?

I think we will see more expressions of hardware differences in next-gen design, with third-party multiplatform games having to shoulder the burden of making their games look good on two different systems.

This past gen, PS4 was lead system out of the gate. By the time X1X showed up, the generation (and console sales) were too entrenched, so there was no genuine reason to push the X1X hardware further beyond upres, better framerate, etc.

This time, the systems will be unevenly matched in hardware respects, and I'm anticipating some fierce graphical battles early on. SSDs offer more than loading time improvements. Near-instant LoD scaling, faster geometry loading, etc. I guarantee that one of the hottest talking points will be "player experience" versus "visual fidelity".

 

[VincentMatts]

I think ultimately, you wont see a huge difference between XSX and PS5 SSDs. Load times (or whats left of them) will probably be slightly better on ps5. But for most of us, waiting 4 seconds or 9 seconds wont change much.

I think youll see new techniques to "hide" the upcoming areas behind mountains or something and those hidden areas will ultimately load faster on next gen. Shit like that.

But the differences between next gen consoles will be extremely small in terms of what the user actually sees.

 

[Esppiral]

People talks as if rendering all the highest lod's on screen at once comes free on the GPU....

 

[Mobilemofo]

They will just not use the full speed of the PS5 SSD. That's the solution that costs the least and that's how businesses work.

The decision that costs the least is the one they will go with. And that's certainly the way to do it.

 

[Trueblakjedi]

It's not sequential.
You load and re-load the same assets multiple times.
Faster SSD means that working set will be either smaller or have 2x more data.
If a game for XBSX is 50GB then PS5 can run 100GB game at the same speed.
I.e. 2x bigger textures, and models.

If the xsx can compress the 100GB game to 50GB (it can) based on the texture compression rate and game package composition where textures ar 66%-75% of game data,
And only need to select the 1/3rd or 1/4th of the texture map at the resolution that will be displayed,
and decompress and display at the same or faster rate (much higher bandwidth to the GPU)
then your statement isnt quite as true as when you thought of it.

That is the XVA.

 

[psorcerer]

If the xsx can compress the 100GB game to 50GB (it can) based on the texture compression rate and game package composition where textures ar 66%-75% of game data,
And only need to select the 1/3rd or 1/4th of the texture map at the resolution that will be displayed,
and decompress and display at the same or faster rate (much higher bandwidth to the GPU)
then your statement isnt quite as true as when you thought of it.

Overall the PS5 data streaming rate is 2x of XBSX at least, after XVA and everything else applied.
There is no free lunch.

 

[Naddy]

or you will see more ps5 exclusives.

only if Sony pays for it.

 

[Trueblakjedi]

Overall the PS5 data streaming rate is 2x of XBSX at least, after XVA and everything else applied.
There is no free lunch.

If they are streaming the same exact assets sure, but if you are able to stream more various assets or asset of higher quality then no.

Lets say that it takes the ps5 1 second to send an 4k 8mb texture. Thus it will take the XSX 2 seconds to send that same texture.

Now if that 8 mb is compressed to 5.6 mb (30% kraken) that's great. In a race to send a 5.6 mb texture to VRAM PS5 would beat the xsx every time. In fact, it can send almost 1.3 textures in the time it takes the xsx to send one.

If that same 4k asset on XSX is 8 mb, and it gets compressed to 4mb with BCPACK, then the time to send to vram is cut in half. The same texture takes exactly the same amount of time to send to the vram on PS5.



Now lets say that in each case they only really needed to display 1/3 of the texture.

As far as we know, the PS5 still sends the whole 5.7mb and the gpu shows that 1/3.

The XSX on the other hand only sends 3.75MB (1/3 of the 8mb) and compresses that again by 50% for transport (these gains are stacked).

So its only sending 1.875 mb of that 8 mb texture... Now that within that same bandwidth cap, it can send TWO of those 4k 8mb textures in a single go (for a total of 3.75 mb).

 

[Naddy]

but if you are able to stream more various assets or asset of higher quality then no

why would a multiplatform dev create various assets or assets of higher quality that would not be used in 2/3 of the platforms they are going to sell the game?
Why spend so much time and resources, so that only a single platform will benefit from this and 2/3 of the platforms will not use it? Is this really cost-effective?

Instead, Devs will just create assets and use assets so that it will work on PC SSDs, so it will also work on PS5 and XSX. then PS5 and XSX will have shorter loading times, and that's it.

 

[Trueblakjedi]

why would a multiplatform dev create various assets or assets of higher quality that would not be used in 2/3 of the platforms they are going to sell the game?
Why spend so much time and resources, so that only a single platform will benefit from this and 2/3 of the platforms will not use it? Is this really cost-effective?

Instead, Devs will just create assets and use assets so that it will work on PC SSDs, so it will also work on PS5 and XSX. then PS5 and XSX will have shorter loading times, and that's it.

I'm thinking next gen assets versus other less gen assets. I think all developers start with a base texture asset of 16, 8 or 4k and then devlove from there per platfom.

 

[DeepEnigma]

why would a multiplatform dev create various assets or assets of higher quality that would not be used in 2/3 of the platforms they are going to sell the game?
Why spend so much time and resources, so that only a single platform will benefit from this and 2/3 of the platforms will not use it? Is this really cost-effective?

Instead, Devs will just create assets and use assets so that it will work on PC SSDs, so it will also work on PS5 and XSX. then PS5 and XSX will have shorter loading times, and that's it.

But I thought they develop for the highest, then scale it down.

That was the Lockhart (gross) excuse by many on here.

 

[psorcerer]

If they are streaming the same exact assets sure, but if you are able to stream more various assets or asset of higher quality then no.

Lets say that it takes the ps5 1 second to send an 4k 8mb texture. Thus it will take the XSX 2 seconds to send that same texture.

Now if that 8 mb is compressed to 5.6 mb (30% kraken) that's great. In a race to send a 5.6 mb texture to VRAM PS5 would beat the xsx every time. In fact, it can send almost 1.3 textures in the time it takes the xsx to send one.

If that same 4k asset on XSX is 8 mb, and it gets compressed to 4mb with BCPACK, then the time to send to vram is cut in half. The same texture takes exactly the same amount of time to send to the vram.

Now lets say that in each case they only really needed to display 1/3 of the texture.

As far as we know, the PS5 still sends the whole 5.7mb and the gpu shows that 1/3.

The XSX on the other hand only sends 3.75MB (1/3 of the 8mb) and compresses that again by 50% for transport (these gains are stacked). So its only sending 1.875 mb of that 8 mb texture... so now it can send TWO of those 4k 8mb textures in a single go (for a total of 3.75 mb).

Again. 4.8GB/sec on XBSX is including BCPack and all other tricks.
SFS only helps serializing the access, etc.

 

[onQ123]

People talks as if rendering all the highest lod's on screen at once comes free on the GPU....

And you talk like it's going to need a lot more GPU power than rendering multiple levels of detail & normal maps

 

[Esppiral]

And you talk like it's going to need a lot more GPU power than rendering multiple levels of detail & normal maps

Sorry for breaking your sweet, sweet dream but that's how it works.

 

[Trueblakjedi]

Again. 4.8GB/sec on XBSX is including BCPack and all other tricks.
SFS only helps serializing the access, etc.

Clearly you did not read what I wrote. It matters what you can fit into 4.8Gb/s.

If I can fit more or the same amount of textures of the same quality into my throughput, then it doesnt matter the speed of the other solution.

XSX is bandwidth by efficiency.

PS5 is bandwidth by architecture.

the best of all worlds will include both.

 

[Ascend]

Again. 4.8GB/sec on XBSX is including BCPack and all other tricks.
SFS only helps serializing the access, etc.

4.8 GB/s is a conservative value which takes only compression into account. This is the way these values are typically given. A raw value of raw throughput, and a compressed value. No other efficiencies or improvements are accounted for.

 

[Trueblakjedi]

Again. 4.8GB/sec on XBSX is including BCPack and all other tricks.
SFS only helps serializing the access, etc.

not true. 4.8 only takes into consideration only the compression and its conservative.

With SFS you reduce the texture request to only what you need to transport rather than the entire map, or even the mip. You only transport the part of the mip that you will see.

Its like culling and discard for textures BEFORE transport.

 

[onQ123]

Sorry for breaking your sweet, sweet dream but that's how it works.

Nope it's harder for the GPU to make a lower detail model look good on screen than it is to make a higher detail model look good on screen.

 

[psorcerer]

XSX is bandwidth by efficiency.

Efficiency where?
Software solutions?

With SFS you reduce the texture request to only what you need to transport rather than the entire map, or even the mip. You only transport the part of the mip that you will see.

No. You can read how SFS works here
Key points:
1. You need to sample, Enabling SFS for more than a small percentage of all the textures will tank performance (think around 1%).
2. HW only helps you to get a nice list of blocks to fetch, in fact just serializing the access. You can reorder the queue and so on.

 

[Ascend]

1. You need to sample, Enabling SFS for more than a small percentage of all the textures will tank performance (think around 1%).

What is actually written;

Performance and Stochastic Update of Feedback Maps
Recall how a feedback map entry represents a mip region of texels in the paired resource, not a single texel. In typical scene rendering, when samples come from closely-spaced-together UV coodinates, the written feedback can be redundant. Redundancy is exacerbated as the mip region size increases.

Given the possibility of wasteful redundancy, it is critical that the operation to update the feedback map does as much coalescing of redundant values as possible before going through with the feedback map write. With this in mind, reducing the sheer quantity of feedback map operations by stochastic methods (e.g., random discard) may be worth pursuing by applications in their shader code.

Remark
Experimental results have shown that discarding over 99% of feedback map updates can still provide enough reliable data to the app. A discard heuristic approach is not proposed as a runtime/compiler feature in this document but would be a possible implementation idea for the app developer.

Sampler Feedback

Engineering specs for DirectX features.

microsoft.github.io

 

[psorcerer]

What is actually written;

Performance and Stochastic Update of Feedback Maps
Recall how a feedback map entry represents a mip region of texels in the paired resource, not a single texel. In typical scene rendering, when samples come from closely-spaced-together UV coodinates, the written feedback can be redundant. Redundancy is exacerbated as the mip region size increases.

Given the possibility of wasteful redundancy, it is critical that the operation to update the feedback map does as much coalescing of redundant values as possible before going through with the feedback map write. With this in mind, reducing the sheer quantity of feedback map operations by stochastic methods (e.g., random discard) may be worth pursuing by applications in their shader code.

Remark
Experimental results have shown that discarding over 99% of feedback map updates can still provide enough reliable data to the app. A discard heuristic approach is not proposed as a runtime/compiler feature in this document but would be a possible implementation idea for the app developer.

Sampler Feedback

Engineering specs for DirectX features.

microsoft.github.io

Yup. All of that is not free.
You need to write a value into the feedback map, and it's not cacheable, because if you can predict writes you don't need the map.
I.e. feedback writes are random memory writes, pretty heavy performance hit.

 

[Lethal01]

well I keep asking the question but seem to not get any answers, if games are similar to this gen in size as being predicted and games are around about 100gb plus for a 12 hour game will we need such a fast speed SSD. not trolling PS5 SSD its a genuine question. how much data will be needed any one time? I mean to get full use of the 6GB speed of PS5 how big with the game files be?

It's not just whether you can get 10GB/s by the next second it's about the being able to design a game differently when you know that you can get the specfic 100mbs you need in the next frame instead of only being able to get 50 or something like that.

Having more available Vram due to having to keep less assets in ram since you can stream in so much per frame. Allowing you to have higher quality or more diverse assets, And allowing for more memory intensive graphical effects.

Allowing for new assets streaming technology like Nanite

Being able to make the world look more seamless for the same reason. for example you may be able to create a cityscape that has a bunch of openings showing an underground portion of the city that you couldn't have done at the quality if you had to keep it all in ram.

Being able to seamlessly transition into giant buildings that house their own levels at any time. Some may think this is already possible but it usually has to be very carefully orchestrated OR the buildings have to be low enough quality that you can have it ready in ram for whenever the player wants to go inside.

Less restrictions on dynamic events in the city due to being able to pull any enemy, vehicles, npc, animation, etc into ram instantly.

better cutscenes, once again you can now make cuts that happen anywhere in the world without having to switch to a video file.
You could have a phone conversations between two characters doing their own thing in two totally different locations that fill the ram to the brim.

here, I'm going to post this shit for the thousandth time.

Once again it's not about streaming things to the ssd nonstop it's about not having to keep things in ram because when it is needed you can load a lot more data in the next frame. That's just one aspect of it though.

 

[Ascend]

Yup. All of that is not free.
You need to write a value into the feedback map, and it's not cacheable, because if you can predict writes you don't need the map.
I.e. feedback writes are random memory writes, pretty heavy performance hit.

That's why they mention the coalescing of the redundant values. That avoids memory writes altogether. The issue is making sure what you're keeping is accurate. If MS advertising is correct, they can achieve a 2x-3x bandwidth boost with this.

 

[eastwood333]

It's not just whether you can get 10GB/s by the next second it's about the being able to design a game differently when you know that you can get the specfic 100mbs you need in the next frame instead of only being able to get 50 or something like that.

Having more available Vram due to having to keep less assets in ram since you can stream in so much per frame. Allowing you to have higher quality or more diverse assets, And allowing for more memory intensive graphical effects.

Allowing for new assets streaming technology like Nanite

Being able to make the world look more seamless for the same reason. for example you may be able to create a cityscape that has a bunch of openings showing an underground portion of the city that you couldn't have done at the quality if you had to keep it all in ram.

Being able to seamlessly transition into giant buildings that house their own levels at any time. Some may think this is already possible but it usually has to be very carefully orchestrated OR the buildings have to be low enough quality that you can have it ready in ram for whenever the player wants to go inside.

Less restrictions on dynamic events in the city due to being able to pull any enemy, vehicles, npc, animation, etc into ram instantly.

better cutscenes, once again you can now make cuts that happen anywhere in the world without having to switch to a video file.
You could have a phone conversations between two characters doing their own thing in two totally different locations that fill the ram to the brim.

here, I'm going to post this shit for the thousandth time.

Once again it's not about streaming things to the ssd nonstop it's about not having to keep things in ram because when it is needed you can load a lot more data in the next frame. That's just one aspect of it though.

And XSX games will be able to do data management improvements via SSD as well

 

[Lethal01]

And XSX games will be able to do data management improvements via SSD as well

Sure, just (reportedly) not nearly as good as devs are boasting about in on ps5 which mean some things made possible with it won't be easy to do on XBSX and thus they will need to be scaled back or changed in some areas which leads us to.

 

[eastwood333]

Sure, just (reportedly) not nearly as good as devs are boasting about in on ps5 which mean some things made possible with it won't be easy to do on XBSX and thus they will need to be scaled back or changed in some areas which leads us to.

Maybe for 1st party, but for 3rd party you're going to be looking at XSX winning in resolution,framerate, ray tracing etc... with PS5 winning by a few seconds in loading times

Can you elaborate on these revolutions in game design that's enabled by a 5GB/s SSD that isn't possible on 2.5 GB/s SSD without tales from my ass speculation?

Did you not see where even Matt said XSX is more powerful in many ways and can do things that PS5 can't?

Have seen that article 1. it's from a guy that literally knows nothing about tech and admits he's just guessing. 2.Its only talking about loading times

Did you see Thomas Mauler's (head of Moon Studios) take on this subject? He says to not expect anything more than a few second faster loading times for multiplats on PS5 with XSX having the better resolution/framerate etc?

 

[Lethal01]

Maybe for 1st party, but for 3rd party you're going to be looking at XSX winning in resolution,framerate, ray tracing etc... with PS5 winning by a few seconds in loading times

And you will see PS5 winning in asset quality, which is more important than resolution to me personally, but raytracing is nice..

Did you not see where even Matt said XSX is more powerful in many ways and can do things that PS5 can't?

Ofcourse, he did so after stating how amazing the SSD will be for development, they will both have their strengths.

Can you elaborate on these revolutions in game design that's enabled by a 5GB/s SSD that isn't possible on 2.5 GB/s SSD without tales from my ass speculation?

Elaborate how? you want me to show you a game from 2 years in the future? Or do you just need a little demo in unreal engine to show how a faster ssd correlates to increase asset quality? The latter I could actually throw together a little demo to show people how memory utilization goes down the faster you unload unneeded assets from memory. but honestly at this point I'm ready to just wait until the ps5 event.

Have seen that article and it's from a guy that literally knows nothing about tech and admits he's just guessing.
Did you see Thomas Mauler's (head of Moon Studios) take on this subject? He says to not expect anything more than a few second faster loading times for multiplats on PS5 with XSX having the better resolution/framerate etc?

I posted it because I know his guesses are accurate regardless of his background. I can agree with partially agree with Thomas.
I think there will be several third part games that just have better loading times but I think it's most likely we also see many devs finding ways to make use of the ssd that are easily scalable from PS5 down to pc level, just like the Unreal Engine is.

But I can agree that for the first year you will see games that do nothing more than load in half a second on ps5 while they load in 2 seconds on XBSX. Then the forums will get flooded by fanboys screaming that the SSD means nothing until more developers make smarter use of it which I think we will actually see very quickly.

 

[Lethal01]

How will devs address the Sony shortcomings?

Why is everyone trying to make this a one way street?

Because addressing systems having slightly less GPU power is something that we are far more used to doing, On the other hand we aren't sure what will be the best answer to scaling a game down to use on weaker SSD's since we are very used to just making sure everything works on the worst one.

 

[eastwood333]

I agree with you on one point. Lets let the games do the talking. I want to see what Sony's 1st party can do and I want to see what Ninja Theory, The Initiative (full of lead/senior naughty dog and Sony Santa Monica devs)Playgrounds new Fable etc can do with XSX's power. All of which are using UE5. Nanite is going be huge

 

[phil_t98]

And you will see PS5 winning in asset quality, which is more important than resolution to me personally, but raytracing is nice..



Ofcourse, he did so after stating how amazing the SSD will be for development, they will both have their strengths.




Elaborate how? you want me to show you a game from 2 years in the future? Or do you just need a little demo in unreal engine to show how a faster ssd correlates to increase asset quality? The latter I could actually throw together a little demo to show people how memory utilization goes down the faster you unload unneeded assets from memory. but honestly at this point I'm ready to just wait until the ps5 event.



I posted it because I know his guesses are accurate regardless of his background. I can agree with partially agree with Thomas.
I think there will be several third part games that just have better loading times but I think it's most likely we also see many devs finding ways to make use of the ssd that are easily scalable from PS5 down to pc level, just like the Unreal Engine is.

But I can agree that for the first year you will see games that do nothing more than load in half a second on ps5 while they load in 2 seconds on XBSX. Then the forums will get flooded by fanboys screaming that the SSD means nothing until more developers make smarter use of it which I think we will actually see very quickly.

Won’t the assists be the same on 3rd party games?

 

[Handy Fake]

But elevators will be higher resolution and with raytracing on XBox though. You will marvel at their shiny glory.

 

[eastwood333]

Because addressing systems having slightly less GPU power is something that we are far more used to doing, On the other hand we aren't sure what will be the best answer to scaling a game down to use on weaker SSD's since we are very used to just making sure everything works on the worst one.

XSX has nearly a 2tf advantage at sustained performance. Far more CU"s that will benefit ray tracing. Better GPU,CPU, and a lot more bandwidth. Thats nothing to scoff about

 

[Lethal01]

XSX has nearly a 2tf advantage at sustained performance. Far more CU"s that will benefit ray tracing. Better GPU,CPU, and a lot more bandwidth. Thats nothing to scoff about

Okay let's pretend we agree on that. Right now we can scale a game that runs on Xbox One X down to working on the nintendo switch. We know what kind of methods will be done to do it so there isn't much to discuss. but for me going from 4k to something like 1600p is the kind of very slight difference I would expect.

Won’t the assists be the same on 3rd party games?

Assuming you mean assets. They may come from the same high quality source but you could see lower res textures or more simplified versions of models on series X.

 

[phil_t98]

Okay let's pretend we agree on that. Right now we can scale a game that runs on Xbox One X down to working on the nintendo switch. We know what kind of methods will be done to do it so there isn't much to discuss. but for me going from 4k to something like 1600p is the kind of very slight difference I would expect.



Assuming you mean assets. They may come from the same high quality source but you could see lower res textures or more simplified versions of models on series X.

So
How big do you think next gen game sizes will be?

 

[Lethal01]

So
How big do you think next gen game sizes will be?

I just want to say it depends but.
100~300gbs

 

[VCL]

with its respective graphics downgrade

 

[phil_t98]

I just want to say it depends but.
100~300gbs

Will only for 2 games on the standard SSD that comes with the console if that big