• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Graphics detail and SSD bandwidth.

psorcerer

Banned
I have seen a lot of confusion in different threads on how SSD bandwidth can impact graphics fidelity. So let's do it.

There are a lot of misconceptions like: SSD has nothing to do with better graphics, more bandwidth means more RAM per frame, it's useless, and so on.

TL;DR
Higher SSD bandwidth can give you much higher detail around the player while keeping RAM usage and RAM bandwidth exactly the same.


Now let's start.
I have a picture for you.

WLKWvlr.png


WTF is that?
It's a graphical approximation of two LoD systems: orange and blue.
The horizontal axis (x) is distance from the player (player being always at zero coordinates)
The vertical axis (y ) is asset quality (texture resolution and such) the closer we are to the player (at zero) the better the asset quality should become.

The blue LoD system has 2.66x worse assets close to the player but kind of a better ones farther from them.
The area below each graph represents the amount of memory that each system uses.
You can calculate it yourself if you want, but I will tell you that the area is exactly the same.
I.e. both systems will use exactly the same amount of RAM and RAM bandwidth,

On the bottom of the graph you can see two lines with dots at the end.
Each line represents the length of the curve for the same color graph.
As you can see it looks like the orange graph is 2x longer than blue one.
And it's indeed the case (1.9x longer) but what does it mean?

Let's think about it.
When player moves anywhere, what happens to the graph?
It moves right.
Why? Because when player moves anywhere (left, right, forward or backward) the LoD system needs to discard the closest data (to the player) and stream-in the newer one.
Player always needs to see the best LoD around.
In fact the closer the data is to the player the faster it becomes discarded. And long distance data stays almost the same for a pretty long time.
(when you pass a building in wilderness all its textures are not needed anymore, but the mountains on the horizon will stay the same)
So the length of the curve represents the amount of data that needs to be reloaded in each frame when the player moves somewhere.

In our blue vs orange systems we load 1.9x more data each time into the orange system and get 2.6x more details around the player.

Here we obviously sacrificed some "longer distance" details for that.
We can tweak the orange system and reduce details around the player to 2x, or even below 1.5x and then we will have enough RAM to fill the "farther" details whuch will be pretty close to the blue system.
Still overall picture will look much better than in the blue system.
Also, in real LoD systems the far details will change so slow that they will be probably never reloaded anyway. And the real curve length will be shorter for both systems thus increasing the difference even more.

Update: how things compare to no streaming
 
Last edited:

IntentionalPun

Ask me about my wife's perfect butthole
Why? Because when player moves anywhere (left, right, forward or backward) the LoD system needs to discard the closest data (to the player) and stream-in the newer one.

Isn't the closest data at risk of being seen again very quickly, and thus can't be discarded without risk of pop-in?

Or am I missing something here?
 

psorcerer

Banned
Isn't the closest data at risk of being seen again very quickly, and thus can't be discarded without risk of pop-in?

Or am I missing something here?

That's a LoD system, i.e. all of that happens with a delay.
If it was like you're describing, you could never discard the closest data ever.
Think of it like you're walking on a checked board, the moment you leave one rectangle you cannot safely discard it, but if you move even one rectangle away - you suddenly can.

To reduce pop-in usually a distance function with hysteresis is used.
But that doesn't really change things too much.
 
Last edited:

IntentionalPun

Ask me about my wife's perfect butthole
That's a LoD system, i.e. all of that happens with a delay.
If it was like you're describing, you could never discard the closest data ever.
Think of it like you're walking on a checked board, the moment you leave one rectangle you cannot safely discard it, but if you move even one rectangle away - you suddenly can.

To reduce pop-in usually a distance function with hysteresis is used.
But that doesn't really change things too much.
OK we understand each other; I just misunderstood what you were claiming. By "closest data" you mean that data right outside that moving rectangle (so not the CLOSEST data, but the nearest data which can be discarded.)
 
Last edited:

BeardGawd

Banned
It's much simpler to think of this as buffer size. Faster streaming of data from SSD means you can reduce the buffer in RAM from the next 30 seconds of gameplay to the next 5 seconds of gameplay. This saved space can now be used for other things if you have the compute resources available (higher quality textures, more variety, more graphics effects, etc..).

Textures take up the vast bulk of this memory. So it remains to be seen if SFS and AI-upscaling of textures which according to MS reduces the amount of space needed for textures by 2x to 3x will help bridge the gap between the two SSD speeds.
 

martino

Member
infinite more detail with zero process on them will look bad though.
process them will limit number of detail you can display.
with that in mind did ms see too low or ps too high ?
generation will tell.
 
This SSD and I/O combine with the Geometry Engine and Frustrum Culling will give us CGI looking graphics.

The GPU (10.3TF) doesn't need to render the geometry not within the view of the player. The SSD will provide the details as soon as it is needed.
 
Last edited:
giphy.gif


Don’t you still need CPU/GPU power to render the image/extra assets?

SSDs. Do. Not. Render. Images.


This SSD and I/O combine with the Geometry Engine and Frustrum Culling will give us CGI looking graphics.

The GPU (10.3TF) doesn't need to render the geometry not within the view of the player. The SSD will provide the details as soon as it is needed.

What? Is this serious?

edit

Also LOL at STILL pretending the XsX doesn’t also have an ultra fast SSD. Sure, Jan. Sure.

Sony studios are about to blow us away next generation

I am being one hundred percent honest when I say it seems like you post this in literally every thread.
 
Last edited:

Sota4077

Member
No. You don't. I think I have clearly explained why there is no extra load. Please read the OP.

What? That is ludicrous. Of course you still need a GPU and CPU to render any image or asset that is on your screen. Please tell me I am drastically misunderstanding you right now, but if I am not you are laughably wrong.

In incredibly simplistic terms a consoles GPU is essentially pulling data from the SSD. The SSD is not pushing it to the GPU. If the GPU of any console or PC has 6GB of VRAM that is the most it is ever going to show on screen at any point and time(Not exactly, but I am making it overly simplistic). An SSD is not doing any heavy lifting besides passing assets to the GPU it is working with. The analogy I used later in this thread is if you and friend are filling a 1 gallon bucket with water and that bucket represents the visible frame on your screen. If you have different diameter hoses one will fill up sooner than the other, but both of you never had more than a gallon of water in your bucket.

The speed of a hard drive does not unburden the GPU it is working with. It just doesn't. Assets can be loaded faster because they are passed to the back buffer faster certainly. But the GPU does not become more powerful as a result. There are never more assets being physically loaded to the GPU. It means exactly what most developers have stated already--that there will be fewer, shorter or even no loading screens in games. That is it. There is nothing that the SSD in any PC or console will do besides allow for developers to load in assets extremely fast so that games are no longer designed with loading screens masked as gameplay where you sneak down a narrow corridor or slowly open a door.

Another big thing is the level of detail of assets comments. SSD's will let developers load higher quality assets faster. It will not let them load more assets of higher quality. If they had access to 6GB of GPU RAM before they have the same amount now. The most a game will ever show on the screen is the maximum amount allowed given the available ram to the GPU. The GPU can read it, back buffer it, pass it to forward buffer and dump it extremely efficiently without having to wait for things to load in. That is all. Every frame still gets rendered in and passed to your screen. There is no secret to SSD's that has been newly discovered with new consoles aside from the efficiency of designing to an identical benchmark all the time.
 
Last edited:
What? That is ludicrous. Of course you still need a GPU and CPU to render any image or asset that is on your screen. Please tell me I am drastically misunderstanding you right now, but if I am not you are laughably wrong.

Utterly amazing how CPU and GPU stopped mattering as soon as Cerny spoke. Legit, very second. Absurdity

it’s getting stupid, this is embarrassing
 

psorcerer

Banned
What? That is ludicrous. Of course you still need a GPU and CPU to render any image or asset that is on your screen. Please tell me I am drastically misunderstanding you right now, but if I am not you are laughably wrong.

I mean that you don't need extra GPU power to render better LoD levels. And you don't need extra RAM space and bandwidth to hold that extra detail.
 
Don’t you still need CPU/GPU power to render the image/extra assets?

Yes, but the problem is that the system can render a much higher detail but the RAM space combined with slow streaming cannot provide the data in time.

FF7 Remake shows that perfectly. Their engine can render multiple characters with CGI like details (the party characters). But there are instances that even if cloud is alone, the characters around him, as well the details in the environment are too poor yet the game can obviously render them in detail. It's because the RAM cannot house all the details. The slow streaming doesn't help.

Now FF7 Remake was made in a 1.8TF of GCN GPU. PS5 is 10.3TF of RDNA2 GPU, but only 16GB of RAM?

If this gen with 1.8TF GPU, we are seeing games that are held back by RAM capacity (5.5GB, more than half used as buffer) and held back by slow streaming, 16GB for 10.3TF of GPU wouldn't cut it.

Also, remember PS5 Geometry Engine. It will only render what is in the current view of the player. I suppose similar to frustum culling but next-gen.



 
Last edited:

VFXVeteran

Banned
I mean that you don't need extra GPU power to render better LoD levels. And you don't need extra RAM space and bandwidth to hold that extra detail.

That's not a clear statement at all. The game will be more dependent on GPU and RAM than the SSD. SSD doesn't operate at the per-pixel level. The GPU does. If you remove the LOD algorithm completely up close, you'll need the GPU and RAM for storage of large textures and shader performance for computing the pixels. Take a very up close and highly detailed character with the most complex shaders (i.e. RTX) and textures at say 16k for example.
 
Last edited:
This is a post from back in MARCH



Video is time stamped right to the good part

I’ll subscribe to this, thanks for posting

for those that can’t watch Karak Karak is basically saying

“You CANNOT draw more than the cpu and gpu can put in the screen.....saying SSDs will allow for more polygons on the screen is bullshit....YOU CANNOT CHEAT A FILLRATE WITH AN SSD.”

Case closed.

I’d rather have more flops than a faster SSD.


Karak Karak handled this on his ACG podcast from day one

The video in the post is timestamped right to the quotes about SSDs

"You CANNOT draw more than the fuckin CPU and GPU can put on the screen."
"You're full of shit...you cannot cheat a fillrate with a SSD."
 
Last edited:

Sota4077

Member
I mean that you don't need extra GPU power to render better LoD levels. And you don't need extra RAM space and bandwidth to hold that extra detail.


OK. I think I understand you better now that I went back and read your post and all your replies. I think most folks are looking at this through the lens of console warrior bullshit since any talk of an SSD is obviously focused on Sony leaning their messaging so heavily on their SSD because it is the spec they beat Microsoft handedly on.

A developer using an SSD still needs a beast of a GPU and CPU to do what they want to do. I presume you were not implying otherwise. What an SSD allows developers to do, from my understanding, is load assets faster. Nothing more. By having the ability to load assets faster they can have higher quality assets closer to the player and load them in faster. That I agree with. But what you seem to be implying with your posts and where I divert with what you are saying is that it unburdens the GPU and CPU. Everything the player sees still needs to be loaded through the GPU of the console. The SSD does not allow the console to perform above spec. So the assets closer to a player may be higher quality, but that does not mean that the assets distant from the player stay the same. Something has to give. If I am designing a scene and I have a GPU memory pool of 1GB I can use it however I like, but there is nothing I can do to 1) Make the GPU load more than that. 2) Make the GPU load assets faster than it is capable of. So If I use 70MB of the memory on foreground assets that means I have 30MB to work with in the background. If I improve those assets in the foreground and use 85MB of the memory for foreground assets I do not still have 30MB of background. I have 15MB. That is a brutal oversimplification, but it still demonstrates my point.

Again unless I am misunderstanding you it really appears that you are implying an SSD lets you load assets so fast that it effectively acts as GPU ram and giving better performance and that is not the case.
 
"You CANNOT draw more than the fuckin CPU and GPU can put on the screen."
"You're full of shit...you cannot cheat a fillrate with a SSD."

Of course, but is the problem the lack of GPU grunt to draw more?

1.8TF in the PS5 can already draw a lot of detailed objects and show high resolution textures. Final Fantasy 7 Remake, when everything is loaded in the RAM, the game is so beautiful it's almost CGI. That' 1.8TF of GCN.

"Even with current gen hardware we can render most single objects at lifelike detail. Every hair in Drakes stubble was a triangle. We can render vast spaces by using smart LoD systems. But we can't store all of the super detailed high res versions for all objs in memory at once. " - Andrew Maximov

That's why we have games in the PS4 like Order 1886 that render very detailed environment, characters and objects but you cannot wander around the environment because the details cannot be streamed inside the RAM in time.

Now imagine 10.3TF of RDNA2 with only 16GB of RAM. Of course, developers can always cut details here and there as they have always done. With ultra-fast SSD and I/O, developers don't need to cut those details now. It will be streamed just as when they are needed. But why stop there? Developers can now add more and more details as they want because the SSD can stream them 'just in time'.
 
Last edited:

Shmunter

Member
This is a post from back in MARCH



Karak Karak handled this on his ACG podcast from day one

The video in the post is timestamped right to the quotes about SSDs

"You CANNOT draw more than the fuckin CPU and GPU can put on the screen."
"You're full of shit...you cannot cheat a fillrate with a SSD."

I’m sure you’d agree 10 different cars on screen looks better than 5 cars duplicated twice. Still 10 cars on screen in both scenarios, still the same load. You still following, or the math is escaping you?

PS5 = 10 different cars, XsX = 5 x 2 cars

The above is bought to you by Cerney SSD spicy sauce. Crying is optional.
 
Last edited:
This is a post from back in MARCH



Karak Karak handled this on his ACG podcast from day one

The video in the post is timestamped right to the quotes about SSDs

"You CANNOT draw more than the fuckin CPU and GPU can put on the screen."
"You're full of shit...you cannot cheat a fillrate with a SSD."
Wow, I didn't see this originally posted but this is great shit from Karak and it's exactly why I hate the console warring bullshit and the FUD FROM BOTH SIDES. Equally. Both sides are spreading so much fud it is truly hilarious. And both sides (for the most part) think they're actually being very fair in their analysis and are above reproach. It just too fucking annoying.
 

IntentionalPun

Ask me about my wife's perfect butthole
Developers can now add more and more details as they want because the SSD can stream them 'just in time'.
Game size becomes the bottleneck.

I know some things help like not needing to duplicate data based on what Cerny said, but I still wonder how big these games are gonna be lol
 
Last edited:

Aceofspades

Banned
This is a post from back in MARCH



Karak Karak handled this on his ACG podcast from day one

The video in the post is timestamped right to the quotes about SSDs

"You CANNOT draw more than the fuckin CPU and GPU can put on the screen."
"You're full of shit...you cannot cheat a fillrate with a SSD."

I have some news for you 🤣

PS5 has higher graphics fillrate than XsX 😅
 

MCplayer

Member
These people seem to forget that CPU and GPU are the ones that usually bottleneck when gaming (on PCs with nvme or SSD), not the storage device, yes ps5 AND XBOX SERIES X have great nvme speeds, still those speeds wont make cpu or gpu faster, they will still bottleneck the CPU and GPU just like before.

Do you know what you will get if you have a slow storage device?
Get Star Wars jedi fallen order and the xbox one | S or PS4 and you will see, game will stop mid gameplay for some seconds to load assets, and then resume, and load textures slower, a faster drive will JUST LOAD ASSETS AND TEXTURES IN TIME so the GPU can continue render and therefore the game continue. PS5 will load faster but wont really make a diference, just in loading screens WHICH WILL STILL EXIST.

 
Last edited:
Game size becomes the bottleneck.

I know some things help like not needing to duplicate data based on what Cerny said, but I still wonder how big these games are gonna be lol

I understand that. But it still depends on the game.

Imagine in the open world game with a setting in a suburban neighborhood. The developers can make each and every house explorable with 10GB of details inside of the house. The SSD can stream all the details as soon as the player opens the door.

Let's say the neighborhood has 30 houses all explorable, does that mean that there will now be 300GB of data needed (10GB*30)? Of course not. Each house can be unique and distinct from each other but a lot of data, geometry and textures can be shared between all of them. It can have 20GB of data shared for the interior of the houses. Different arrangements and combination can create a distinction on each and every house. Imagine the same set-up in the game set in a building with 20 floors and 200 rooms.

Now that is only example in my mind right now. I'm sure there are more examples and better examples that will show that taking advantage of the ultra-fast SSD and I/O would not necessarily result in the games ballooning in size.
 

Shmunter

Member
I understand that. But it still depends on the game.

Imagine in the open world game with a setting in a suburban neighborhood. The developers can make each and every house explorable with 10GB of details inside of the house. The SSD can stream all the details as soon as the player opens the door.

Let's say the neighborhood has 30 houses all explorable, does that mean that there will now be 300GB of data needed (10GB*30)? Of course not. Each house can be unique and distinct from each other but a lot of data, geometry and textures can be shared between all of them. It can have 20GB of data shared for the interior of the houses. Different arrangements and combination can create a distinction on each and every house. Imagine the same set-up in the game set in a building with 20 floors and 200 rooms.

Now that is only example in my mind right now. I'm sure there are more examples and better examples that will show that taking advantage of the ultra-fast SSD and I/O would not necessarily result in the games ballooning in size.
Yes that’s right, you can still refresh gigs of data in ram for optimum scene detail, but it doesn’t mean it has to be constantly unique. Rearranging, flipping, warping. Anything goes.
 
These people seem to forget that CPU and GPU are the ones that usually bottleneck when gaming (on PCs with nvme or SSD), not the storage device,

That's true because the CPU has to decompress the data from the SSD. That takes too much toll on the CPU. Basically, PC CPU cannot stream a lot of data during RUNTIME. (During loadtime it doesn't matter).

PS5 I/O has a decompressor that is equivalent of 9 ZEN2 CORES. That will allow it to stream 9GB/s-22GB/s or data during runtime.

With the geometry engine not rendering what is not visible on-screen as well as the back-facing details, CPU and GPU will be starved with data to render if not for the SSD. Unless of course, it's a fighting game like Tekken and Streetfighter.

You just have to see what the 1.8TF GCN GPU inside the PS4 was already capable of rendering. Order 1886 and FF7 Remake has CGI looking moments, at the cost of not being able to go anywhere because the data can't be streamed 'just in time'.
 

IntentionalPun

Ask me about my wife's perfect butthole
Let's say the neighborhood has 30 houses all explorable, does that mean that there will now be 300GB of data needed (10GB*30)? Of course not. Each house can be unique and distinct from each other but a lot of data, geometry and textures can be shared between all of them. It can have 20GB of data shared for the interior of the houses. Different arrangements and combination can create a distinction on each and every house. Imagine the same set-up in the game set in a building with 20 floors and 200 rooms.

For sure; brought that up before. What we should see is more variety of textures in any given scene. Rather than a single "pattern" like 'this is the area where everything is brown wood plus some stone' they can mix together things far more to create the appearance of uniqueness.
 
Last edited:

MCplayer

Member
That's true because the CPU has to decompress the data from the SSD. That takes too much toll on the CPU. Basically, PC CPU cannot stream a lot of data during RUNTIME. (During loadtime it doesn't matter).

PS5 I/O has a decompressor that is equivalent of 9 ZEN2 CORES. That will allow it to stream 9GB/s-22GB/s or data during runtime.

With the geometry engine not rendering what is not visible on-screen as well as the back-facing details, CPU and GPU will be starved with data to render if not for the SSD. Unless of course, it's a fighting game like Tekken and Streetfighter.

You just have to see what the 1.8TF GCN GPU inside the PS4 was already capable of rendering. Order 1886 and FF7 Remake has CGI looking moments, at the cost of not being able to go anywhere because the data can't be streamed 'just in time'.
yeah just like xbox series x has BCPack, still I see the GPU and CPU being the bottleneck in next gen too.
 
Nice, been a couple weeks since I've seen an SSD secret sauce thread.

The part where most people get confused is that people aren't implying that an SSD can't change how games are made or played, it's just that this benefit extends to the XSX as well. People around here talk as if PS5 is the only console with an SSD or custom designs to improve development.
 

VFXVeteran

Banned
Of course, but is the problem the lack of GPU grunt to draw more?

1.8TF in the PS5 can already draw a lot of detailed objects and show high resolution textures. Final Fantasy 7 Remake, when everything is loaded in the RAM, the game is so beautiful it's almost CGI. That' 1.8TF of GCN.

"Even with current gen hardware we can render most single objects at lifelike detail. Every hair in Drakes stubble was a triangle. We can render vast spaces by using smart LoD systems. But we can't store all of the super detailed high res versions for all objs in memory at once. " - Andrew Maximov

That's why we have games in the PS4 like Order 1886 that render very detailed environment, characters and objects but you cannot wander around the environment because the details cannot be streamed inside the RAM in time.

Now imagine 10.3TF of RDNA2 with only 16GB of RAM. Of course, developers can always cut details here and there as they have always done. With ultra-fast SSD and I/O, developers don't need to cut those details now. It will be streamed just as when they are needed. But why stop there? Developers can now add more and more details as they want because the SSD can stream them 'just in time'.

This is the main reason why several people on these boards have fights and arguments over the console vs. PC vs. everything. That statement is so far from the truth it's not ever funny.

You are taking your subjective opinion and then comparing it to CG which uses infinitely more complex algorithms and can NOT be rendered in realtime on ANY hardware platform - including the highest end graphics boards. And yet, this is what causes so much dissension. Subjective opinion != objective fact.
 

Sota4077

Member
This is a really dumb post SMH


Everything isn't CPU or Compute bound


No, but what this entire original post is stating is that an SSD effectively unburdens a GPU which is just patently false. And SSD allows assets to be loaded faster. That is all. It does nothing more than that. The system is still constrained by the specs of its hardware. The GPU and the CPU. The SSD does not allow MORE to be rendered on the screen at one time. It allows assets to be loaded to the screen faster when they are needed. If a GPU has 1GB of available memory it cannot and will not ever load more than 1GB of assets. Meaning there will never be more than 1GB of assets on screen at any time. The OP of this post is implying otherwise. He is implying that the SSD allows for the system to load more and that is false.
 
Last edited:

jakinov

Member
Don’t you still need CPU/GPU power to render the image/extra assets?

SSDs. Do. Not. Render. Images.




What? Is this serious?

edit

Also LOL at STILL pretending the XsX doesn’t also have an ultra fast SSD. Sure, Jan. Sure.



I am being one hundred percent honest when I say it seems like you post this in literally every thread.
I mean they both have powerful GPU/CPUs . So both can still render a lot of things. The idea behind a fast SSD is low latency when you want an asset from secondary storage and when you want to display it. The concept is about giving the developers the ability to instantaneously pull assets and display it to the users without technical/visual hiccups.

Also, nobody is saying SSDs render images. But SSD secondary storage play a role in the final output/experience. Secondary storage don't process data either but there's a reason why secondary storage important for software like databases. It's again not because it processes data but it's that secondary storage is a bottle neck. It's called secondary storage because it's not your main fast good source of storage but like games it's usually where the bulk of your data resides (due to costs). For databases, you can't process data quickly if it's bottle necked by slow secondary storage; similarly, in games you can't show game assets quickly to render while it's in slow storage. Your processing units can spend less time waiting on storage to do things and just start doing what it wants to do. The other thing with games is that it's an experience and you don't detract from that experience by spending extra second(s) to load things in. If your goal was to hypothetically dynamically load in a brand new environment every time the character does a 180, a developer might load less on one system over the other because they can't get enough details in time for it to not detract from the experience.

Whether or not Sony investment in SSD pays off we'll have to wait and see. I think that the benefit of the SSD would vary by game (or event part of games). As all games (or applications) have different requirements.

I think that if anyone is going to try to argue against the importance of PS5's fast GPU, it's better to argue of any diminishing returns rather than make strawman arguments about SSDs not being able to render or downplaying the benefit of ensuring your CPU (and GPU) have as fast as possible access to all your data (something that even Microsoft touts about their Xbox).
 

psorcerer

Banned
A developer using an SSD still needs a beast of a GPU and CPU to do what they want to do. I presume you were not implying otherwise. What an SSD allows developers to do, from my understanding, is load assets faster. Nothing more. By having the ability to load assets faster they can have higher quality assets closer to the player and load them in faster.

Made another image for you:

eFZ7cth.png


What we see here is: what happens if player moves.
So the player position (vertical black dashed line) changed.
The new asset state is in violet.
The left upper blue part is unloaded from RAM. The right upper violet part is loaded into RAM from SSD.
The common part stays the same.

Now on the bottom you have a rectangle. that's the same RAM when there is no streaming.
I.e. level was loaded once and no new data is coming, until we unload the whole level and load a new one.
The height of that red rectangle shows the asset quality for that case.
As you can see it's drastically (4x) lower than the blue LoD.
But it still uses the same amount of RAM (areas of blue and red graphs are the same).
And you don't need to load anything when player moves.

Does it make things clearer?
 
Top Bottom