The need for sustained high throughput loading of data in games? *spawn*

Of course they will be used.You only invest in silicon if you meant to use it.
Slight correction. You only invest in silicon if you intend it to be used. It may not be used in the end, as per XB360 Tessellation units, dynamic branching on RSX (far too slow), and the ID buffer on PS4 it would seem, and I'm sure many other examples across the generations. Hardware has a long history of speculative technologies that fell by the wayside for various reasons.
 
8GB/s consistently would represent less than 100 seconds of unique data before you've spunked the contents of the entire SSD. Unless you're running a gameplay loop that involves loading and dumping the same assets repeatedly every few seconds (which would once again be back to taking a lot of work on the designers end), you're not going to be seeing 8GB/s consistently.

If the way games worlds are generated doesn't change, yes.

I'm not saying you there are no use cases for the ability to stream 8GB/s into VRAM on a constant basis. Of course there are, they are practically infinite. What I'm saying is: where are you pulling all this new data from? 16GB VRAM is a good 10% of your entire game content. If you're refreshing it every 2 seconds then you've used up your entire game content in 20 seconds. If we had 20TB game installs then this would surely make for some amazing looking games. But we have 100GB game installs so the super fast IO system is limited by the amount of content available to load.

I agree but, again, assuming the way game worlds are generated doesn't change. Games have been using mostly fixed geometry and large textures for decades because this is the most efficient way to store and load data. You don't have a piece of brick geometry and 500 different brick textures and build a wall from them, you have your wall geometry and slap a texture over it. Job done. But now, maybe you don't have to do that, with a lot of I/O you could build your worlds from smaller pieces that are slotted together that is naturally more destructive (if you need that) because you fundamentally have small pieces for the whole.

Now I'm not suggesting that every wall in every game should be modelled from 2,000 individual bricks, what I'm saying is that couldn't have been done on current gen and will be possible next generation and that it's pretty much impossible to predict how the removal of this technical barrier will impact game design. But also, when we're talking 'unique' data/assets, that dismisses the many possible combinations in which geometry, textures and shaders can be combined which has to be almost limitless.

Landing? How about flying over an entire city (with road traffic) of it accurately mapped with photogrammetry? And of course, nothing beats low altitude bush flying in FS 2020. If ever there was a game that screams PCMR, its FS 2020.

Flighting over a city means all of the objects are far away and that you're never close enough to see the details. Can you land the aircraft in a street, get out of the aircraft and get up close the vehicles and pedestrians going about their business - the detail and density we've seen in Night City? Because if not, how is this relevant? :???:

edit: typos
 
Last edited by a moderator:
If the way games worlds are generated doesn't change, yes.



I agree but, again, assuming the way game works are generated doesn't change. Games have been using mostly fixed geometry and large textures for decades because this is the most efficient way to store and load data. You don't have a piece of brick geometry and 500 different brick textures and build a wall from them, you have your wall geometry and slap a texture over it. Job done. But now, maybe you don't have to do that, with a lot of I/O you could build your worlds from smaller pieces that are slotted together that is naturally more destructive (if you need that) because you fundamentally have small pieces for the whole to dis are more easily and naturally destructible.

Now I'm not suggesting that every wall in every game should be modelled from 2,000 individual bricks, what I'm saying is that couldn't have been done on current gen and will be possible next generation and that it's pretty much impossible to predict how the removal of this technical barrier will impact game design. But also, when we're talking 'unique' data/assets, that dismisses the many possible combinations in which geometry, textures and shaders can be combined which has to be almost limitless.



Flighting over a city means all of the objects are far away and that you're never close enough to see the details. Can you land the aircraft in a street, get out of the aircraft and get up close the vehicles and pedestrians going about their business - the detail and density we've seen in Night City? Because if not, how is this relevant? :???:

In FS 2020, you definitely can with fully animated ground crews and all that. Only at select airports though. Also, in a low altitude fly-by of New York, the amount of data being thrown at you is just mind boggling with individual buildings being recognisable. No current games come even close.
 
Highway Traffic, bottom left corner. Ace ;) shadows are spot on as well.
lol, i have no words, I honestly didn't think they were doing crowds and cars either.

and animals ;)

lol. I guess will flyby Africa

 
In FS 2020, you definitely can with fully animated ground crews and all that.
This is literally orders of magnitude less complex than the density, detail and complexity of Cyberpunk rendering Night City. This is the point I'm making. You need to be able to transition from flying, which likely has many, many more assets (because you can inherently see further) but at much, much lower detail (because they're further away) to the absolute opposite in a few seconds - depending on how quickly the vehicles can land, you can get out and start talking to people and going into buildings.
 
A hybrid of the two would be interesting. And possible for the first time this generation thanks to SSD's being able to provide the bandwidth, and streaming able to overcome storage limitations.
 
But next gen consoles don’t actually have 16GB of VRAM. PCs can have that much VRAM plus substantial amounts of system RAM.
So the usable pool of VRAM in consoles is quite a bit smaller than that, and since when do we scoff at fast things? We’re getting them, they will he used!

Yeah I was just round up for convenience. I think it's 13.5 GB usable on the XSX (mentioned in one of my earlier posts). I'm certainly not scoffing at the new IO solutions though, I think they're awesome! We just need to be careful about setting expectations around how they're going to be used.

But now, maybe you don't have to do that, with a lot of I/O you could build your worlds from smaller pieces that are slotted together that is naturally more destructive (if you need that) because you fundamentally have small pieces for the whole.

Now I'm not suggesting that every wall in every game should be modelled from 2,000 individual bricks, what I'm saying is that couldn't have been done on current gen and will be possible next generation and that it's pretty much impossible to predict how the removal of this technical barrier will impact game design. But also, when we're talking 'unique' data/assets, that dismisses the many possible combinations in which geometry, textures and shaders can be combined which has to be almost limitless.

I totally agree that increased re-use of very high quality assets may be one way to make increased use of the IO capabilities. The UE5 demo may have been doing something along those lines as I understand they only used a handful of megascans for the whole demo which can obviously put together in near infinite different ways to create that kind of rocky environment.

Something like Spiderman may also be a good candidate given that most of the roads and buildings are likely to re-use assets.
 
But next gen consoles don’t actually have 16GB of VRAM.

Its 13.5 GB game usable on SeriesX compared to 5 GB on base console. So it's 2.7 times more than they had available to use before.
 
IAlso, in a low altitude fly-by of New York, the amount of data being thrown at you is just mind boggling with individual buildings being recognisable. No current games come even close.
Any videos of that?

The high altitude stuff is very simple models (textured cuboids on the whole) slowly being swapped out for more. Once you get low enough, travelling at high speed, you don't need lots of geometry detail because it's not present on screen for very long, and even less so if there's motion blur applied.

So I'm not convinced the detail is anything special. Contrast that with an open world game, a GTA or Spider-Man, with that level of detail on the ground for being able to walk into buildings and look at people, only where you can transition from one end to another in an instant. I dunno, a catapult that'll shoot you 10 blocks in one second. That'll need to change from city-level quality to fast, streamed, low level objects, to city-level quality, in that one second, which is far more challenging than the linear fight of a plane.

What FS2020 seems to be doing more impressively is amount of data, from the cloud. Lots and lots of data changing fairly slowly.
 
I'm wondering if we'll start to see procedurally generated texture data?

Edit: some examples:

http://www.ctrl-alt-test.fr/2018/texturing-in-a-64kb-intro/

The attached image is a procedurally generated forest floor texture.

Okay, so maybe I'm being a lunatic, but I'm curious. Would it be possible to create half a dozen texture spheres - each unique, as per the example in the quote - and then blend them together like SDF's?

I'm imagining this in reference to ground textures, as it's the most straightforward example I can think of. To my mind, that would take less storage than an enormous, bespoke terrain texture, but allow for greater variety than e.g. a generic grassland texture leading to generic forest floor texture.
 
This is literally orders of magnitude less complex than the density, detail and complexity of Cyberpunk rendering Night City. This is the point I'm making. You need to be able to transition from flying, which likely has many, many more assets (because you can inherently see further) but at much, much lower detail (because they're further away) to the absolute opposite in a few seconds - depending on how quickly the vehicles can land, you can get out and start talking to people and going into buildings.

Nope. Cyberpunk's night city is simply not more complex to render than New York at close range in full 4K glory. Every single building is being rendered with lot of details as per photogrammetry.

191015160910-ny-full-169.jpg
 
Any videos of that?

The high altitude stuff is very simple models (textured cuboids on the whole) slowly being swapped out for more. Once you get low enough, travelling at high speed, you don't need lots of geometry detail because it's not present on screen for very long, and even less so if there's motion blur applied.

So I'm not convinced the detail is anything special. Contrast that with an open world game, a GTA or Spider-Man, with that level of detail on the ground for being able to walk into buildings and look at people, only where you can transition from one end to another in an instant. I dunno, a catapult that'll shoot you 10 blocks in one second. That'll need to change from city-level quality to fast, streamed, low level objects, to city-level quality, in that one second, which is far more challenging than the linear fight of a plane.

What FS2020 seems to be doing more impressively is amount of data, from the cloud. Lots and lots of data changing fairly slowly.

May be in Flight Simulator X.
 
Nope. Cyberpunk's night city is simply not more complex to render than New York at close range in full 4K glory. Every single building is being rendered with lot of details as per photogrammetry.
Everything is FAR AWAY. Can you link to any FS2020 pictures of city streets close up. Like a first person game at ground level in the city. Because that is what Cyberpunk does. Does FS2020 look good depicting things that are hundreds of feet away? Absolutely. Does it look good showing the city at 5 feet?
 
Everything is FAR AWAY. Can you link to any FS2020 pictures of city streets close up. Like a first person game at ground level in the city. Because that is what Cyberpunk does. Does FS2020 look good depicting things that are hundreds of feet away? Absolutely. Does it look good showing the city at 5 feet?
ooof 5 ft?

this is about 5 ft.


FF to 2:08

when the games come out, we can see what the city is like.
 
Everything is FAR AWAY. Can you link to any FS2020 pictures of city streets close up. Like a first person game at ground level in the city. Because that is what Cyberpunk does. Does FS2020 look good depicting things that are hundreds of feet away? Absolutely. Does it look good showing the city at 5 feet?

What does being far away actually mean in this case? Density of geometry is still absolutely insane. An yes the game does look good at 5 feet.

procedurally-generated_grass_normandie-1.jpeg


wecutthegrassbecausitwastoolong-scaled.jpg
 
ooof 5 ft?

this is about 5 ft.


FF to 2:08

when the games come out, we can see what the city is like.

Flight enthusiasts are definitely a bit of a weird bunch but one of our defining characteristics is being a graphics whore. Slight imperfections are discussed vehemently for weeks over at AVSIM.
And that New York scene (along with Barcelona for some reason) easily brings a 1080ti to its knees.
 
juste look at those cubic houses in the background, and all the NY building are just cubes with hires textures, but that's logical.
it's a super high res google earth, no way near the ground detail of a Spider man.
 
Back
Top