Support NeoGAF

[DaGwaphics]

Ah...I thought this was talking about the same tech I've seen demos of that upscaled and corrected a video from say 1904 with all the grain and noise and makes it 4k. Seems to be how they're describing it in the articles though to be fair that may be just tech-nerd journalism at work then. If It's not the same 'deep learning' type of tech I stand corrected. I have seen people claiming that this will in fact be the case (using a 40kb texture and somehow magically inflating it into a 4k texture) so that's where I was coming from.

For the record, I DO think those deep learning techniques will fundamentally change a LOT of thing for us. More so even than compression did initially, by a long ways. I just don't believe we're close yet to that kind of technology being feasible. Although I have to say..though it's not graphics related, they are doing some almost spooky stuff with deep learning in the cyber security space. There's a product called "Deep Instinct" for example that I've been looking at for my company. They were making what seemed to be outlandish claims about their product but from what I've seen so far....it appears to be realistic! Really amazing stuff. I still don't buy that we'll be able to generate TRUE AI either, but with deep learning, we really may not need to. We shall see.

This is one of the better videos of DLSS 2.0, a good variety of content.



Still might need some work, but things like this could help boost RT. With no tensor equivalents, not sure if it's feasible for the consoles.

 

[Ascend]

The Xbox Series X RAM Setup

How is it really set up?
After thinking about it, I think the XSX memory setup is a bit deceiving in its advertising... They are basically telling you it is like this (each number is a memory chip with the amount of GB it has);
1 1 1 1 1 1 1 1 1 1 + 1 1 1 1 1 1 (560GB/s + 336 GB/s)

While in reality the memory config is more like this;
2 2 2 2 2 2 + 1 1 1 1
That makes it look like 336 GB/s + 224 GB/s, but technically that's not true either... Because the lanes from the 2GB chips and the 1GB chips are not 'separate'. The RAM is not split, but one pool. So like this;

2 2 2 2 2 2 1 1 1 1
The question is, why don't they simply advertise with 560 GB/s? That looks like a perfectly viable 10 x 56GB/s setup... Right? Well... They are aware that if you do not allocate RAM efficiently, you'll run into problems. If you fill only the 2GB chips first, you get 336GB/s. If you fill the 1GB chips first, you get 224GB/s. If you fill them randomly, you'll get inconsistent performance and the effective bandwidth constantly changing on you. They want developers to use the RAM like they are advertising it, which is entirely possible. The more lanes you use for data, the better, obviously. Even though it is not configured like that in reality, by artificially 'splitting' the 2GB modules in two 1GB modules, you achieve the same result as what they are advertising.

Aaaand here's where the complexity starts...
There is one caveat though. Obviously the 2GB modules use the lanes that they have. So even if you artificially split them, there isn't magically additional lanes for data transfer. The lanes needs to be shared by the two sections of the 2GB chip... To put it another way, the 1GB chips get the full 56GB/s per chip and thus per GB (please stick with me here). The 2GB chips, if not used correctly, rather than getting the advertised 56GB/s to reach the total of 560 GB/s will get 28GB/s per GB in the worst case scenario . So you can't really advertise it as 560GB/s + 336 GB/s here. In the worst case scenario, you are talking about 280 GB/s + 336 GB/s. Now that is REALLY atrocious bandwidth.

Is the RAM split or not?
The reality is, that the RAM will work like a hybrid between a split and a unified RAM pool. What do I mean by that? It will work as a unified RAM pool in the sense that both the GPU and the CPU will have access to all the data on all the 16GB. However, it will work as a split RAM pool in terms of data allocation. There will have to be two priority levels in the 2GB RAM chips. Only when there are no 1st priority calls on the RAM, can the 2nd priority be executed. So whatever uses high bandwidth (like textures) will need to be given 1st priority, and whatever uses low bandwidth, can go into 2nd priority.

It's getting more complicated...
And sadly, once again it is not that simple either... Because if you need something right now on screen that is low bandwidth, and it is set to low priority, you will get pop-ins for example. Or if you allocate all sound to the low priority section, then you'll get weird sound delays etc... Yes. That is quite complicated... If it is like that, I can see why we are getting many developers saying are liking the PS5 more. It's simply much simpler. Despite the power of the XSX, it will require some creativity to learn and work with the RAM system of the XSX. If it really is like this, it's actually possible (if not inevitable) that initially we see PS5 games looking better than XSX games, unless they decide to keep all RAM usage under 10GB for both for ease of development. I don't think it will actually work that way. I certainly hope it doesn't...

Free cache lesson for you. It's relevant, I promise
The best way to really do it is that you use the 2GB RAM chips as a sort of L1 and L2 cache. I don't know if people here know how cache works... I'll try and explain it shortly...
Say you have a processor, and the processor has two levels of cache. The first level L1 can store two letters, and the second level can store four letters (typically, the L2 cache is larger, but for the XSX RAM it would be smaller). Cache basically saves the most frequently used data in it for data access. The 'closer' the cache to the CPU, the faster it is.
In the beginning, the cache is empty, and as the processor does jobs, it fills the caches and changes the data in the caches accordingly.
Now imagine I am typing a long word, like pneumonoultramicroscopicsilicovolcanoconiosis (yeah that's the longest word in the english language lol). The CPU has nothing in cache in the beginning, but it doesn't know that. It checks L1, no data. Then checks L2, no data. Then Checks RAM, no data. Ultimately it arrives at the storage device, and copies all the used letters of the program into the caches and the RAM, in order. The letter that is the most common will be saved in L1. The second most common letter will also be saved in L1. Now L1 is full. The 3rd, 4th, 5th, and 6th most common letters will be saved in L2. The rest are in RAM. Next time if I type that word, the CPU can read much of the data from L1, then L2, then the RAM. It will do it much faster than before.

So if we go letter for letter, first the L1 and L2 caches will look like this;
L1 [p,n]
L2 [e,u,m,o]

Now as we type further, things start to change... As we type pneumono, the o and the n become the most common, so p is 'downgraded' to L2, and o is added to L1. n stays;
L1 [n, o]
L2 [e,u,m,p]

As we type ultra, u has been used as often as the other letters, but L1 is full, so it stays in L2, and everything stays the same. And so on and so on. When a letter has been used 3 times it will shift down one of the ones in L1 towards L2, and since L2 is full, the least common letter will be shifted to RAM if it's not already there.

Enough caches. I want some more RAM sweetness!
So that was the short lesson on caches... Going back to the XSX RAM... If they let it work like cache, then everything is allocated to the 10GB first as data is accessed. When there's a data lookup, it will always look in the 10GB first. So, that means the 1GB ram chips and the first 'tier' of the 2GB chips will have priority. Only when the required data is not found there, the lookup will take place in the 2nd tier of the 2GB chips. If done this way, the bandwidth will not interfere with each other, and realistically give you 56GB/s per GB in the 2GB RAM chips also, for both tiers. Now, the 560 GB/s is practically guaranteed, and so is the 360 GB/s.
If it works that way, it's actually a really smart design... And the XSX will have practically zero issues with RAM allocation. Then the XSX will truly have a great bandwidth advantage over the PS5. This is more likely the solution that MS came up with. Having developers manually tune it would be a nightmare. You might have to be a bit more careful with RAM than the PS5, but it would not be a huge issue.

 

[Bo_Hazem]

Cant find the CyberPunk thread

The worst thing about the game is being 1st PV, would've loved to be 3rd PV.

 

[James Sawyer Ford]

After thinking about it, I think the XSX memory setup is a bit deceiving in its advertising... They are basically telling you it is like this (each number is a memory chip with the amount of GB it has);
1 1 1 1 1 1 1 1 1 1 + 1 1 1 1 1 1 (560GB/s + 336 GB/s)

While in reality the memory config is more like this;
2 2 2 2 2 2 + 1 1 1 1
That makes it look like 336 GB/s + 224 GB/s, but technically that's not true either... Because the lanes from the 2GB chips and the 1GB chips are not 'separate'. The RAM is not split, but one pool. So like this;

2 2 2 2 2 2 1 1 1 1
The question is, why don't they simply advertise with 560 GB/s? They are aware that if you do not allocate RAM efficiently, you'll run into problems. If you fill only the 2GB chips first, you get 336GB/s. If you fill the 1GB chips first, you get 224GB/s. If you fill them randomly, you'll get inconsistent performance and the effective bandwidth constantly changing on you. They want developers to use the RAM like they are advertising it, which is entirely possible. The more lanes you use for data, the better, obviously. Even though it is not configured like that in reality, by artificially 'splitting' the 2GB modules in two 1GB modules, you achieve the same result as what they are advertising.

There is one caveat though. Obviously the 2GB modules use the lanes that they have. So even if you artificially split them, there isn't magically additional lanes for data transfer. The lanes needs to be shared by the two sections of the 2GB chip... To put it another way, the 1GB chips get the full 56GB/s per chip and thus per GB (please stick with me here). The 2GB chips, if not used correctly, rather than getting the advertised 56GB/s to reach the total of 560 GB/s will get 28GB/s per GB in the worst case scenario . So you can't really advertise it as 560GB/s + 336 GB/s here. In the worst case scenario, you are talking about 280 GB/s + 336 GB/s. Now that is REALLY atrocious bandwidth.

The reality is, that the RAM will work like a hybrid between a split and a unified RAM pool. What do I mean by that? It will work as a unified RAM pool in the sense that both the GPU and the CPU will have access to all the data on all the 16GB. However, it will work as a split RAM pool in terms of data allocation. There will have to be two priority levels in the 2GB RAM chips. Only when there are no 1st priority calls on the RAM, can the 2nd priority be executed. So whatever uses high bandwidth (like textures) will need to be given 1st priority, and whatever uses low bandwidth, can go into 2nd priority.

And sadly, once again it is not that simple either... Because if you need something right now on screen that is low bandwidth, and it is set to low priority, you will get pop-ins for example. Or if you allocate all sound to the low priority section, then you'll get weird sound delays etc... Yes. That is quite complicated... If it is like that, I can see why we are getting many developers saying are liking the PS5 more. It's simply much simpler. Despite the power of the XSX, it will require some creativity to learn and work with the RAM system of the XSX. If it really is like this, it's actually possible (if not inevitable) that initially we see PS5 games looking better than XSX games, unless they decide to keep all RAM usage under 10GB for both for ease of development. I don't think it will actually work that way. I certainly hope it doesn't...

The best way to really do it is that you use the 2GB RAM chips as a sort of L1 and L2 cache. I don't know if people here know how cache works... I'll try and explain it shortly...
Say you have a processor, and the processor has two levels of cache. The first level L1 can store two letters, and the second level can store four letters. Cache basically saves the most frequently used data in it for data access. The 'closer' the cache to the CPU, the faster it is.
In the beginning, the cache is empty, and as the processor does jobs, it fills the caches and changes the data in the caches accordingly.
Now imagine I am typing a long word, like pneumonoultramicroscopicsilicovolcanoconiosis (yeah that's the longest word in the english language lol). The CPU has nothing in cache in the beginning, but it doesn't know that. It checks L1, no data. Then checks L2, no data. Then Checks RAM, no data. Ultimately it arrives at the storage device, and copies all the used letters of the program into the caches and the RAM, in order. The letter that is the most common will be saved in L1. The second most common letter will also be saved in L1. Now L1 is full. The 3rd, 4th, 5th, and 6th most common letters will be saved in L2. The rest are in RAM. Next time if I type that word, the CPU can read much of the data from L1, then L2, then the RAM. It will do it much faster than before.

So if we go letter for letter, first the L1 and L2 caches will look like this;
L1 [p,n]
L2 [e,u,m,o]

Now as we type further, things start to change... As we type pneumono, the o and the n become the most common, so p is 'downgraded' to L2, and o is added to L1. n stays;
L1 [n, o]
L2 [e,u,m,p]

As we type ultra, u has been used as often as the other letters, but L1 is full, so it stays in L2, and everything stays the same. And so on and so on. When a letter has been used 3 times it will shift down one of the ones in L1 towards L2, and since L2 is full, the least common letter will be shifted to RAM if it's not already there.


So that was the short lesson on caches... Going back to the XSX RAM... If they let it work like cache, as data is accessed, then everything is allocated to the 10GB first. When there's a data lookup, it will always look in the 10GB first as well. So, that means the 1GB ram chips and the first 'tier' of the 2GB chips will have priority. Only when the required data is not found there, the lookup will take place in the 2nd tier of the 2GB chips. If done this way, the bandwidth will not interfere with each other, and realistically give you 56GB/s per GB in the 2GB RAM chips also. Now, the 560 GB/s is practically guaranteed, and so is the 360 GB/s.
If it works that way, it's actually a really smart design... And the XSX will have practically zero issues with RAM allocation. Then the XSX will truly have a great bandwidth advantage over the PS5.

There is no great bandwidth advantage over PS5

 

[LemmingWithaParachute]

The worst thing about the game is being 1st PV, would've loved be 3rd PV.

Completely disagree, but theres no reason why they couldnt include a camera choice. If Bethesda's developers can figure it out, Im sure CDPR can.

They show the player character standing in front of the mirror in that video, but whats the point of dressing them up if you cant see them? Its a bit like a game that lets you create your face, but then you spend the entire game with a helmet on for max armor.

 

[Three Jackdaws]

I sure HOPE so, though I'm tempering my expectations. Bottom line, I want to see demos and games from BOTH platforms even though I'm almost certainly getting a PS5. I just think they're going to be close enough to where I can at least see where things will be this generation by viewing demos of either platform. Or not. We'll see that too! Bottom line, I'm so hyped it's not even funny! Still just trying to keep my feet on the ground though so I'm not crushingly disappointed.

Yeah I have a bias towards Playstation because I’ve grown up playing it, almost all of my favourite games are Playstation exclusives such as God of War and the Uncharted series, which says a lot because it explains why everyone plays the “exclusives” card when it comes to console wars. Truth be told, I’ll definitely buy the Series X if it has some killer exclusives (I’m really hoping it does). If we zoom out a little and look at the bigger picture, this only benefits us the consumers, competition between PS5 and Series X means we’ll get better prices and deals, along with more amazing exclusives, so it’s a win/win from my perspective

 

[THEAP99]

The worst thing about the game is being 1st PV, would've loved be 3rd PV.

Honestly, having gone on an immersive simulator bing these last few months with BioShock and Dishonored, there isn't much wrong with first person. But the issue I see with cyberpunk is it's all about the world and stuff and it is an RPG, doesn't make sense you can't see your character that you create.
It also doesn't help that the gameplay they've shown so far really does just seem like standard first-person shooter stuff. Hopefully, it's improved from how janky it looked before.

 

[Ascend]

There is no great bandwidth advantage over PS5

Oh. Ok... Just to be clear.... Data goes like this (simplified);

SSD -> RAM -> GPU -> your screen

9 - 4.8 = 4.2GB/s. That is the difference between the console SSDs, for the data travel between the SSD and the RAM. That is considered humongous...

But then...
560 - 448 = 112 GB/s. That is the difference between the console RAM, for the data to travel between the RAM and the GPU. This is 'no great bandwidth advantage'.

I wanna have what you're smoking.

 

[James Sawyer Ford]

Oh. Ok... Just to be clear.... Data goes like this (simplified);

SSD -> RAM -> GPU -> your screen

9 - 4.8 = 4.2GB/s. That is the difference between the console SSDs, for the data travel between the SSD and the RAM. That is considered humongous...

But then...
560 - 448 = 112 GB/s. That is the difference between the console RAM, for the data to travel between the RAM and the GPU. This is 'no great bandwidth advantage'.

I wanna have what you're smoking.

1. 560 GB/s is only for 10 GB.

2. The effective bandwidth for all 16GB is lower

3. The bandwidth/TF for VRAM is basically identical for PS5 and XSX

4. The SSD has access to the entire game data, not just the VRAM

 

[HawarMiran]

The worst thing about the game is being 1st PV, would've loved be 3rd PV.

early on it was rumored to be FPS and TPS. That would be awesome . Like in Skyrim. I am the type of guy who needs to see his char with the aquired armor and shit

 

[Ascend]

1. 560 GB/s is only for 10 GB.

2. The effective bandwidth for all 16GB is lower

3. The bandwidth/TF for VRAM is basically identical for PS5 and XSX

4. The SSD has access to the entire game data, not just the VRAM

1) Yes? So?
2) Thank you for not understanding anything I wrote on the last page
3) Which means the 12TF XSX GPU will not be bottlenecked and thus will easily be able to show its 2TF advantage over the PS5 GPU.
4) Uh... I showed in which direction data goes, not what has access to what. But thank you for not understanding.

 

[James Sawyer Ford]

I understood it, the effective bandwidth is lower by having split pools. You tried to explain how it’s somehow not.

yes, the XSX will in theory be able to push more pixels. But not much, and the relative bandwidth demanded to push those pixels is the same for XSX as PS5, no advantage per pixel

Yes the 120% difference in speeds that the VRAM will soak up constantly from the SSD is a much bigger deal

 

[Ascend]

I understood it, the effective bandwidth is lower by having split pools. You tried to explain how it’s somehow not.

I explained how it definitely can be but why it doesn't have to be lower.

yes, the XSX will in theory be able to push more pixels. But not much, and the relative bandwidth demanded to push those pixels is the same for XSX as PS5, no advantage per pixel

A useless metric you just made up? No advantage per pixel + more pixels = still more pixels

Yes the 120% difference in speeds that the VRAM will soak up constantly from the SSD is a much bigger deal

How and why?

 

[Metroiddarks]

The Xbox Series X RAM Setup

How is it really set up?
After thinking about it, I think the XSX memory setup is a bit deceiving in its advertising... They are basically telling you it is like this (each number is a memory chip with the amount of GB it has);
1 1 1 1 1 1 1 1 1 1 + 1 1 1 1 1 1 (560GB/s + 336 GB/s)

While in reality the memory config is more like this;
2 2 2 2 2 2 + 1 1 1 1
That makes it look like 336 GB/s + 224 GB/s, but technically that's not true either... Because the lanes from the 2GB chips and the 1GB chips are not 'separate'. The RAM is not split, but one pool. So like this;

2 2 2 2 2 2 1 1 1 1
The question is, why don't they simply advertise with 560 GB/s? That looks like a perfectly viable 10 x 56GB/s setup... Right? Well... They are aware that if you do not allocate RAM efficiently, you'll run into problems. If you fill only the 2GB chips first, you get 336GB/s. If you fill the 1GB chips first, you get 224GB/s. If you fill them randomly, you'll get inconsistent performance and the effective bandwidth constantly changing on you. They want developers to use the RAM like they are advertising it, which is entirely possible. The more lanes you use for data, the better, obviously. Even though it is not configured like that in reality, by artificially 'splitting' the 2GB modules in two 1GB modules, you achieve the same result as what they are advertising.

Aaaand here's where the complexity starts...
There is one caveat though. Obviously the 2GB modules use the lanes that they have. So even if you artificially split them, there isn't magically additional lanes for data transfer. The lanes needs to be shared by the two sections of the 2GB chip... To put it another way, the 1GB chips get the full 56GB/s per chip and thus per GB (please stick with me here). The 2GB chips, if not used correctly, rather than getting the advertised 56GB/s to reach the total of 560 GB/s will get 28GB/s per GB in the worst case scenario . So you can't really advertise it as 560GB/s + 336 GB/s here. In the worst case scenario, you are talking about 280 GB/s + 336 GB/s. Now that is REALLY atrocious bandwidth.

Is the RAM split or not?
The reality is, that the RAM will work like a hybrid between a split and a unified RAM pool. What do I mean by that? It will work as a unified RAM pool in the sense that both the GPU and the CPU will have access to all the data on all the 16GB. However, it will work as a split RAM pool in terms of data allocation. There will have to be two priority levels in the 2GB RAM chips. Only when there are no 1st priority calls on the RAM, can the 2nd priority be executed. So whatever uses high bandwidth (like textures) will need to be given 1st priority, and whatever uses low bandwidth, can go into 2nd priority.

It's getting more complicated...
And sadly, once again it is not that simple either... Because if you need something right now on screen that is low bandwidth, and it is set to low priority, you will get pop-ins for example. Or if you allocate all sound to the low priority section, then you'll get weird sound delays etc... Yes. That is quite complicated... If it is like that, I can see why we are getting many developers saying are liking the PS5 more. It's simply much simpler. Despite the power of the XSX, it will require some creativity to learn and work with the RAM system of the XSX. If it really is like this, it's actually possible (if not inevitable) that initially we see PS5 games looking better than XSX games, unless they decide to keep all RAM usage under 10GB for both for ease of development. I don't think it will actually work that way. I certainly hope it doesn't...

Free cache lesson for you. It's relevant, I promise
The best way to really do it is that you use the 2GB RAM chips as a sort of L1 and L2 cache. I don't know if people here know how cache works... I'll try and explain it shortly...
Say you have a processor, and the processor has two levels of cache. The first level L1 can store two letters, and the second level can store four letters (typically, the L2 cache is larger, but for the XSX RAM it would be smaller). Cache basically saves the most frequently used data in it for data access. The 'closer' the cache to the CPU, the faster it is.
In the beginning, the cache is empty, and as the processor does jobs, it fills the caches and changes the data in the caches accordingly.
Now imagine I am typing a long word, like pneumonoultramicroscopicsilicovolcanoconiosis (yeah that's the longest word in the english language lol). The CPU has nothing in cache in the beginning, but it doesn't know that. It checks L1, no data. Then checks L2, no data. Then Checks RAM, no data. Ultimately it arrives at the storage device, and copies all the used letters of the program into the caches and the RAM, in order. The letter that is the most common will be saved in L1. The second most common letter will also be saved in L1. Now L1 is full. The 3rd, 4th, 5th, and 6th most common letters will be saved in L2. The rest are in RAM. Next time if I type that word, the CPU can read much of the data from L1, then L2, then the RAM. It will do it much faster than before.

So if we go letter for letter, first the L1 and L2 caches will look like this;
L1 [p,n]
L2 [e,u,m,o]

Now as we type further, things start to change... As we type pneumono, the o and the n become the most common, so p is 'downgraded' to L2, and o is added to L1. n stays;
L1 [n, o]
L2 [e,u,m,p]

As we type ultra, u has been used as often as the other letters, but L1 is full, so it stays in L2, and everything stays the same. And so on and so on. When a letter has been used 3 times it will shift down one of the ones in L1 towards L2, and since L2 is full, the least common letter will be shifted to RAM if it's not already there.

Enough caches. I want some more RAM sweetness!
So that was the short lesson on caches... Going back to the XSX RAM... If they let it work like cache, then everything is allocated to the 10GB first as data is accessed. When there's a data lookup, it will always look in the 10GB first. So, that means the 1GB ram chips and the first 'tier' of the 2GB chips will have priority. Only when the required data is not found there, the lookup will take place in the 2nd tier of the 2GB chips. If done this way, the bandwidth will not interfere with each other, and realistically give you 56GB/s per GB in the 2GB RAM chips also, for both tiers. Now, the 560 GB/s is practically guaranteed, and so is the 360 GB/s.
If it works that way, it's actually a really smart design... And the XSX will have practically zero issues with RAM allocation. Then the XSX will truly have a great bandwidth advantage over the PS5. This is more likely the solution that MS came up with. Having developers manually tune it would be a nightmare. You might have to be a bit more careful with RAM than the PS5, but it would not be a huge issue.

Hi Ascend

Sorry If don't understand you but as I got it you mean the Memory for 560 GB/s will work similar to cache L1 and the slower memory is a L2.

So:

1.- The ram actually can works in this way or is just a theory ?
2.- As I know usually the L2 should be bigger than L1 or not because doesn't make sense fill data in a faster memory from a smaller pool?
3.-Move the data in this way will no increment the latency which is worse than decrement in bandwidth?
4.-If you theory was right why just don't say 13.5 Gb/s are from the dev with the part faster and slow, instead to say the slow part is for OS of the game?

 

[saintjules]

The first three sentences are commonly used in 'leaks'..

 

[James Sawyer Ford]

I explained how it definitely can be but why it doesn't have to be lower.


A useless metric you just made up? No advantage per pixel + more pixels = still more pixels


How and why?

its not a useless metric. It means whatever extra bandwidth XSX has, it has the same limitations on a relative basis as PS5. It would be more meaningful if the bandwidth per TF were higher

the PS5 with its faster SSD and GPU cache scrubbers will not need to have as much idle data sitting in the GPU pr VRAM, allowing for better asset quality

if the 120% speed increase doesn’t matter, why did they pursue it?

 

[Neo_game]

Xbox BW advantage is only if game uses 10gb of RAM and that too probably is much less than most people think. According to a person called Lady Gaia. This was posted in other thread by geordiemp

https://www.neogaf.com/threads/the-...epth-analysis-suggests.1534248/post-257618553

 

[Ascend]

Hi Ascend

Sorry If don't understand you but as I got it you mean the Memory for 560 GB/s will work similar to cache L1 and the slower memory is a L2.

So:

1.- The ram actually can works in this way or is just a theory ?

It can, depending on what MS put in the controller. If they do it software based, it's going to be a lot slower. It wouldn't be the wisest thing. I don't think doing it any other way than using the RAM in the concept of caches will be efficient or easy for developers.

2.- As I know usually the L2 should be bigger than L1 or not because doesn't make sense fill data in a faster memory from a smaller pool?

Normally, L2 is indeed larger than L1. But that's for speed and cost reasons. Normally, L1 is the fastest and the smallest. L2 is slower than L1 but larger etc. In this case, the L2 will be smaller, but equal in speed to L1. It's not literally cache, but, it will likely work as one.

3.-Move the data in this way will no increment the latency which is worse than decrement in bandwidth?

Depends if it's a hardware or a software solution. Technically, data in the 336GB/s portion will have a longer latency, since the CPU needs to work through the whole 10GB portion first. But I doubt it will be a software solution. They have hardware accelerators for everything, basically. But we'll have to wait and see.

4.-If you theory was right why just don't say 13.5 Gb/s are from the dev with the part faster and slow, instead to say the slow part is for OS of the game?

That's actually how it is. The OS will use 2.5GB of the 360GB/s portion. So that means the developer gets 3.5GB of the 'slow' L2 RAM. And that is again yet another deep dive we can go into... Because they can split the OS allocation however they want to. They can divide the 2.5GB over all the six 2GB modules, reserving around 417MB per RAM chip. Or they can use two 2GB chips, or anything in between... Most likely they'll split them over the six modules to maximize the free bandwidth of the L2 portion.

its not a useless metric. It means whatever extra bandwidth XSX has, it has the same limitations on a relative basis as PS5. It would be more meaningful if the bandwidth per TF were higher

You can look at it that way. Or you can see it as the XSX being able to stretch its legs just as well as the PS5 on a relative basis.

the PS5 with its faster SSD and GPU cache scrubbers will not need to have as much idle data sitting in the GPU pr VRAM, allowing for better asset quality

Although true, the amount of idle data will still be a fraction compared to HDDs in both cases. And this would be assuming that all else is equal. If we take SFS into account, things may flip completely. And I say may, because it may not; We simply don't have enough details on both to be able to know for sure.

if the 120% speed increase doesn’t matter, why did they pursue it?

If the 120% speed increase matters, why didn't MS pursue it?

 

[Gamernyc78]

Completely disagree, but theres no reason why they couldnt include a camera choice. If Bethesda's developers can figure it out, Im sure CDPR can.

They show the player character standing in front of the mirror in that video, but whats the point of dressing them up if you cant see them? Its a bit like a game that lets you create your face, but then you spend the entire game with a helmet on for max armor.

I love COD first person but it annoys me I can't really see the cool outfits I get or buy so I know the feeling.

 

[James Sawyer Ford]

It can, depending on what MS put in the controller. If they do it software based, it's going to be a lot slower. It would be the wisest thing. I don't think doing it any other way will be efficient or easy for developers.


Normally, L2 is indeed larger than L1. But that's for speed and cost reasons. Normally, L1 is the fastest and the smallest. L2 is slower than L1 but larger etc. In this case, the L2 will be smaller, but equal in speed to L1. It's not literally cache, but, it will likely work as one.


Depends if it's a hardware or a software solution. Technically, data in the 336GB/s portion will have a longer latency, since the CPU needs to work through the whole 10GB portion first. But I doubt it will be a software solution. They have hardware accelerators for everything, basically. But we'll have to wait and see.


That's actually how it is. The OS will use 2.5GB of the 360GB/s portion. So that means the developer gets 3.5GB of the 'slow' L2 RAM. And that is again yet another deep dive we can go into... Because they can split the OS allocation however they want to. They can divide the 2.5GB over all the six 2GB modules, reserving around 417MB per RAM chip. Or they can use two 2GB chips, or anything in between... Most likely they'll split them over the six modules to maximize the free bandwidth of the L2 portion.



You can look at it that way. Or you can see it as the XSX being able to stretch its legs just as well as the PS5 on a relative basis.


Although true, the amount of idle data will still be a fraction compared to HDDs in both cases. And this would be assuming that all else is equal. If we take SFS into account, things may flip completely. And I say may, because it may not; We simply don't have enough details on both to be able to know fore sure.


If the 120% speed increase matters, why didn't MS pursue it?

I think MS would have pursued it if they thought it was technically feasible

 

[Ascend]

Xbox BW advantage is only if game uses 10gb of RAM and that too probably is much less than most people think. According to a person called Lady Gaia. This was posted in other thread by geordiemp

https://www.neogaf.com/threads/the-...epth-analysis-suggests.1534248/post-257618553

That person is not wrong, if things are done in the traditional way.

 

[Neo_game]

So essentially the XSX will play Lockhart games at 4K?

Thats one way to handle this.

I think all multiplats next gen will use S version as the base console sadly. It will be unfair if a person on S is playing at 30fps against someone online on X at 60fps. There will be parity for console I think. Only resolution will be different. If they have some headroom RT can make up for it accordingly.

 

[Foamy]

I think all multiplats next gen will use S version as the base console sadly. It will be unfair if a person on S is playing at 30fps against someone online on X at 60fps. There will be parity for console I think. Only resolution will be different. If they have some headroom RT can make up for it accordingly.

I don't think this is correct. They promised to allow everyone to play together, they never promised identical experiences or a level playing field.

 

[kensama]

If the 120% speed increase matters, why didn't MS pursue it?

maybe, because on one side the architect of the PS5 is also a an industry veteran developer and know which part of a hardware will give struggle a developer.

On the other side as incredible as the Surface team be, they maybe have no personn (as far as we know) as cerny that had the two knowledge of hardware (since Sega era) and Software (since VCS2600).
And it's reflecting into the conception of both hardware. One more classical (XsX) and the other more balanced and exotic.

 

[Neo_game]

That person is not wrong, if things are done in the traditional way.

For PS5 they need to hold data on RAM for next 1sec of gameplay only thanks to the SSD. The more RAM the games uses the better it will be for PS5.

 

[Corndog]

maybe, because on one side the architect of the PS5 is also a an industry veteran developer and know which part of a hardware will give struggle a developer.

On the other side as incredible as the Surface team be, they maybe have no personn (as far as we know) as cerny that had the two knowledge of hardware (since Sega era) and Software (since VCS2600).
And it's reflecting into the conception of both hardware. One more classical (XsX) and the other more balanced and exotic.

Please tell me about these awesome games Cerny has made.

 

[pasterpl]

Remember what ms said about their architecture, it can apparently multiply the effective available memory pool x2-3 times

The components of the Xbox Velocity Architecture all combine to create an effective multiplier on physical memory that is, quite literally, a game changer.

 

[Gamernyc78]

Please tell me about these awesome games Cerny has made.

He can tell you about the awesome hardware he's been involved in

 

[LiquidRex]

Is it possible all 16GBs of PS5 RAM is available for developers?

 

[SgtCaffran]

Please tell me about these awesome games Cerny has made.

Mark Cerny - Wikipedia

en.m.wikipedia.org

 

[husomc]

Thanks a lot, dear brother! Well, it's Saturday for us in Oman.

I didn't realize countries still follow the moon sighting. At my place it's strictly by calendar. And Ramazan Mubarak to all members who keep fast during this month.

 

[FranXico]

Is it possible all 16GBs of PS5 RAM is available for developers?

Not likely. There always are some system functions running in the background.

 

[SgtCaffran]

Is it possible all 16GBs of PS5 RAM is available for developers?

Maybe not all but I expect two things:
- the PS5 probably needs less idle ram due to the fast SSD and highly optimised I/O
- the PS5 probably needs less reserved system RAM due to the same reason and might have more RAM available for games

This can lead to higher quality assets or greater draw distance in some games. There might be other benefits but I can't tell you.

 

[LiquidRex]

Cerny focused on bottlenecks in his presentation because he wanted to emphasize the SSD, but just because he said that bottlenecks are a priority doesn't make the PS5 bottlenecks-less console. According to every leak we've had it seems as if Sony had experimented with higher than 500GB/s memory bandwidth for months if not years and back then the PS5 GPU profile was just 9.2TF, now it's over 2200Mhz and they only have 448GB/s, probably because GDDR6 prices spiked really hard. 448GB/s has to handle a GPU more powerful than the 5700XT (which has 448GB/s all to itself), a powerful CPU, 8-9 GB/s of streaming SSD data, an up to 20GB/s audio chip, RT which is a HUGE bandwidth hog and denoising which loves bandwidth just as much. That sounds like a potential bottleneck.

But again, each and every PS5 bottleneck or shortcoming can be canceled by lowering the resolution.

Regarding cache, L0 is inside the CU and XSX has 44% more CUs, hance more L0 cache. In RDNA the L2 is tied to the memory controller so each 32-bit PHY has 512KB. PS5 has 8 PHYs which means 4MB of L2, XSX has 10 PHYs which means 5MB. One of them might have made some custom changes, but that's what we know right now.


Both had 64 ROPs in Github and it is unlikely to change so late in the process.


Kill the bandwidth? Replacing

Cerny focused on bottlenecks in his presentation because he wanted to emphasize the SSD, but just because he said that bottlenecks are a priority doesn't make the PS5 bottlenecks-less console. According to every leak we've had it seems as if Sony had experimented with higher than 500GB/s memory bandwidth for months if not years and back then the PS5 GPU profile was just 9.2TF, now it's over 2200Mhz and they only have 448GB/s, probably because GDDR6 prices spiked really hard. 448GB/s has to handle a GPU more powerful than the 5700XT (which has 448GB/s all to itself), a powerful CPU, 8-9 GB/s of streaming SSD data, an up to 20GB/s audio chip, RT which is a HUGE bandwidth hog and denoising which loves bandwidth just as much. That sounds like a potential bottleneck.

But again, each and every PS5 bottleneck or shortcoming can be canceled by lowering the resolution.

Regarding cache, L0 is inside the CU and XSX has 44% more CUs, hance more L0 cache. In RDNA the L2 is tied to the memory controller so each 32-bit PHY has 512KB. PS5 has 8 PHYs which means 4MB of L2, XSX has 10 PHYs which means 5MB. One of them might have made some custom changes, but that's what we know right now.


Both had 64 ROPs in Github and it is unlikely to change so late in the process.


Kill the bandwidth? Replacing the 1GB chips with 2GB chips will improve their bandwidth, not reduce it. It will make the whole 20GB run at 560GB/s, no split speeds or weird memory addressing, just a unified 20GB running at 560GB/s. So yes, it's probably MS's only solution to combat Sony's SSD if they want to combat it just like clock speed was Sony's only tool vs MS's CU count.

Regarding the SSD, both getting a huge IO leap in the shape of the SSD is the big win for next-gen. But you can't ignore Sony's ~2x advantage in that area. Will it matter? Time will tell, we haven't had games designed around something like this since the N64.


When I say node based GI I mean methods used today in games like Control. It's a rasterized effect using prebaked nodes in order to do GI. I know of two ways of enhancing node based GI. The first is using RT in order to remedy (pun intended!) node based GI artifacts which Control uses in the PC RTX version. The other is using RT in order to keep updating the nodes so the nodes are not baked but updated every few frames. Both methods still use rasterization while utilizing RT in order to make the effect better. It's not RT GI, but it is something worlds better than the current generation and it can look very impressive. But it's not RT GI which is currently the heaviest RT effect I know, much more demending than RT reflections or RT shadows.

With regards to ROPs, the reason I asked is because I've seen comments made by some, that the XSX has 80 + ROPs, and I'm curious as to why that would be.

All I can think is they are going off Big Navi, which I assume has a higher ROP count, and they think the XSX has a lighter version of Big Navi.

 

[Metroiddarks]

It can, depending on what MS put in the controller. If they do it software based, it's going to be a lot slower. It would be the wisest thing. I don't think doing it any other way will be efficient or easy for developers.

If this was true Microsoft should have said they change the bases about how the memory works, they mentioned every new technology that they implemented,
so this sounds more like a theory based on a wish than a real basis.

Normally, L2 is indeed larger than L1. But that's for speed and cost reasons. Normally, L1 is the fastest and the smallest. L2 is slower than L1 but larger etc. In this case, the L2 will be smaller,
but equal in speed to L1. It's not literally cache, but, it will likely work as one.

But also is because you don't want your 'L1' is starving while is waiting the 'L2' filled it again with data, in the end I don't see this practical for third parties and even for first parties to deal
with that headache, because the third want also to have their game compatible for other system and also the same Xbox want to translate that game to PC.

So I think the way the dev will work about this is just know before to get some info from SSD if that will a kind of Tier1 data which should be loaded in the faster part of the memory since
the beginning, which is almost exactly what the crytek engineer said.

Depends if it's a hardware or a software solution. Technically, data in the 336GB/s portion will have a longer latency, since the CPU needs to work through the whole 10GB portion first. But I doubt it will be a software solution. They have hardware accelerators for everything, basically. But we'll have to wait and see.

I am sure that will be not eliminate just like that, the data need needs time to be moved from one side to another and you controler need same to identified what part of the memory will
move and which part of the L1 will replace, if the latency exist in real cache for CPU will exists here.

So in the end you will gain some GB/s but also incrementing the latency which could be worst.

That's actually how it is. The OS will use 2.5GB of the 360GB/s portion. So that means the developer gets 3.5GB of the 'slow' L2 RAM. And that is again yet another deep dive we can go into... Because they can split the OS allocation however they want to. They can divide the 2.5GB over all the six 2GB modules, reserving around 417MB per RAM chip. Or they can use two 2GB chips, or anything in between... Most likely they'll split them over the six modules to maximize the free bandwidth of the L2 portion.

Yes sure if all its game is just around 10 GB then yes can put all in the faster part but even then OS of the console needs to be read/write in some moment so not even in the best scenario the
perfect 560 will be reach it, also because as you know cannot read to 560GB/s and then 336GB/s in the same time.

We all agree the memory of XSX is faster but how the memory was composed the teoric limit is far to be real one but even with that should be faster than PS5, the
difference is the Devs in PS5 doesn't need to program all in so low level in order to not be penalized so hard for an much slower memory to be close to its theoretical performance.

Your theory doesn't has bases is closer to be conspiracy plot than a technical reason of why XSX has the perfect memory, we don't heard from anyone to say XSX change the way
how the memory works.

Basically you want Xbox redefine how the memory works while never say nothing about this when they love to show off , also the devs needs to be worried about how to move the
data inside the fast part of the memory always when is necessary, also the same code needs to work in other platforms because Xbox will never has exclusive game of XSX or
XSS, their approach needs to remove the latency is some strange/magic way.

And you get defensive when someone tells you it doesn't make sense because memory doesn't work in that way Sorry but I don't see any sense in that theory is bases many in 'you
cannot prove they won't do that' (fallacy ad ignorantiam)is like say the PS5 increment the IPC for Its GPU in Y% and this should be true because you cannot prove is not true and even
that example is more plausible.

But don't worry about this the Xbox one has a garbage memory compare with PS4 and is not like its games looks awful.

 

[geordiemp]

1) Yes? So?
2) Thank you for not understanding anything I wrote on the last page
3) Which means the 12TF XSX GPU will not be bottlenecked and thus will easily be able to show its 2TF advantage over the PS5 GPU.
4) Uh... I showed in which direction data goes, not what has access to what. But thank you for not understanding.

Read below . Someone on beyond3d suggested there were timing issues trying to do more simultaneuous things between the 6 and 10...so maybe MS did try. There are 4 operations per clock and its complex.

You would think for the premium they would just go with all 2GB, especially with Lockart option.

The signals are complex and timing on memory access using te common bus as well, thats why its better having it all the same apparanetly. Its not so straight forward. See what the signal actually looks like below per memory clock

 

[JerryinSoCal]

Completely disagree, but theres no reason why they couldnt include a camera choice. If Bethesda's developers can figure it out, Im sure CDPR can.

They show the player character standing in front of the mirror in that video, but whats the point of dressing them up if you cant see them? Its a bit like a game that lets you create your face, but then you spend the entire game with a helmet on for max armor.

I've never felt a 1st person view was more immersive, I actually find them more limiting and annoying if I'm being honest. Third person gives you a better sense of where you are in the world and let's you see your "avatar" while first person limits your view and makes it unrealistically narrow and you still see your avatars hands. I don't get why they made the change it wasn't needed, nobody was complaining about the witcher being 3rd person.

 

[TearInto5th]

The main issue with this tired and invalid argument, is that devs build from the top down. Especially in the case of Xbox, where everything will release on PC which will already be far superior by the time these systems launch. Devs won't even be starting at XSX, they will shoot past that. XSX and Lockhart (if such a thing exists) would get settings toned down to whatever level they need. The overall game design won't be an issue since there will be some level of parity between the systems in regards to the CPU and storage. Wide gaps in GPU performance can be bridged with changes in resolution and LOD settings.

How can there be Parity when the PC doesn't use an SSD as standard storage?

They might have all the GPU and CPU grunt in the world, but if the SSD is still 50 times slower than the console due to the number of bottlenecks, the PC market will hold the consoles back. It isn't just about graphical power anymore. The PC won't have the advantage anymore.

Shit we see this generation won't be able to be replicated, whether its on XSX or PS5 due to the lowest common denominator. The HDD.

 

[kensama]

Please tell me about these awesome games Cerny has made.

Go to wikipedia
I see what you try to do downplay Cerny game with Knack (a good game in my opinion) but if you see his historical game CV you will see that he work on Sony's major IP and Sega's major IP. He also contributed to create ICE team inside Naughty dog studio.

But go to wikipedia!

 

[BluRayHiDef]

Does Sony have enough time to add more VRAM to the PS5's design before manufacturing begins?

 

[B_Boss]

Please tell me about these awesome games Cerny has made.

Objection, subjective lol? That is irrelevant and “awesome” differs from one person to the next.

The major point is that he is a game developer industry veteran who would absolutely understand and know very well the obstacles develops encounter when creating video games. There is a reason the man is in the Academy of Interactive Arts & Sciences Hall of Fame. It lends weight to his credibility and understanding of his profession.

 

[Lunatic_Gamer]

 

[Fdkenzo]

I am glad to hear that. But i must Say, this game does not look like a NexT gen game for me.

 

[Bonsaibäumchen]

That MMO Spy Game sounds like shit.
Have fun trying to be sneaky as fuck with a full fledged plan in your head only to get busted because some 12 year old kid or other simplistic fool has the same mission and fucks everything up.
Also expect lots of trolls and so on.

 

[Bonsaibäumchen]

I wondering, if Sony actually managed to mass manufacture these pcbs with through hole cooling, will we see this kind of cooling solution more often in the future? Also stacked apus in a cooling sandwich?
This might actually be a "proof of concept" for future "console" generations with even more dedicated hardware solutions stacked together.
Now I'm curious, if they can effectivly pull this off, what kind of beast a PS5 Pro might turn out to be.

 

[psorcerer]

If all rumors are true about XSS/XSX, then devs will be building games for XSX and then simply reduce the target resolution for XSS. The visuals, effects and features will be identical.

That's nonsense. If you can easily "downport" it already means that the game was made with the weaker hw in mind.
Nobody ever "downports" anything without huge sacrifices. It's not a matter of architecture it's about engineering "budget": memory, rendering, textures, effects, etc. All of that is rationed. If you can ration it properly for a smaller hw it means you target that hardware.

 

[Shmunter]

How can there be Parity when the PC doesn't use an SSD as standard storage?

They might have all the GPU and CPU grunt in the world, but if the SSD is still 50 times slower than the console due to the number of bottlenecks, the PC market will hold the consoles back. It isn't just about graphical power anymore. The PC won't have the advantage anymore.

Shit we see this generation won't be able to be replicated, whether its on XSX or PS5 due to the lowest common denominator. The HDD.

PC is problematic indeed. You can cut last gen console support with a rubber stamp. But those pesky pc’s will be kicking around with their slow assed hdd’s and ssd’d for the foreseeable future. Sure you can dramatically revise minimum requirements on Pc, but that’s like hitting the reset button on a billion users - not sure how feasible without a prolonged uptake period.

But what’s interesting is the console platform holders mandating their games to execute from the SSD, cannot do so from USB3. This a strong testament to push the envelope forward. Indeed it is counter to MS supporting all gamers on all platforms. They’re going to have to make their minds up.

Next gen, with everything it brings, playing on pc surely must be losing its lustre. Cheaters, no game suspend/resume, slow loading, patches, compatibility. Where’s the attraction anymore?

Edit : I can see game mods being a sizeable strength of pc.

 

[pawel86ck]

Regarding the SSD, both getting a huge IO leap in the shape of the SSD is the big win for next-gen. But you can't ignore Sony's ~2x advantage in that area. Will it matter? Time will tell.

I think 2x difference in SDD speed will help with load times and that's not a big issue for me, because it's not like XSX load times should be long either. For streaming purposes however 9 GB/s should yield diminishing returns with only 13 GB available to developers. Streaming pool will be much below 9 GB for sure and probably even below 5 GB (I believe thats why MS build slower SDD).

 

[psorcerer]

If it works that way, it's actually a really smart design...

It's a good design for bandwidth. But a really bad design for latency.
CPU will access ram. It will access it much less frequently than GPU. Will be placed always in second tier and will always incur 2x latency penalty.
Caches work because their latency is much smaller than each "higher" tier.

 

[Shmunter]

I think 2x difference in SDD speed will help with load times and that's not a big issue for me, because it's not like XSX load times should be long either. For streaming purposes however 9 GB/s should yield diminishing returns with only 13GB available to developers. Streaming pool waill be much below 9 GB for sure and probably even below 5 GB (I believe thats why MS build slower SDD).

? Your streaming reasoning is completely reversed. Lol