Next-Gen PS5 & XSX |OT| Console tEch threaD

Status
Not open for further replies.
Wow. The compressed youtube video really doesn't do the UE5 demo justice. Those stills are just unbelievable! It's incredible to see how much impact a super fast SSD can have on the level of detail!

Put in mind that's 1440p upscaled, that's why some grain helps masking that and it's very, very slightly blurry compared to native 4K. Give me 1440p everyday in that case.
 
Last edited:
Quick question.

What do you mean by arriving late?

Your displaying a higher res textuire over a lower one and blending it in so you dont get that POP.

The higher res texture has arrived late, or you would not be blending it over the poorer texture - there would be no need lol.

SFS is a blended pop in for data misses....But MS have PR and dressed it up to a multiplier - its sort of is, in that you can have a smaller streaming pool and you have a band aid so it does not look as bad when streaming fucks up.....

What do you think it is ? Must be magic :messenger_beaming:
 
Last edited:
Your displaying a higher res textuire over a lower one and belding it in so you dont get that POP.

The higher res texture has arrived late, or you would not be blending it over the poorer texture - there would be no need lol.

SFS is a blended pop in for data misses....But MS have PR and dressed it up to a multiplier - its sort of is, in that you can have a smaller streaming pool and you have a band aid so it does not look as bad when streaming fucks up.....

What do you think it is ? Must be magic :messenger_beaming:

I see so it's really a way to stop pop in from happening. I'm pretty sure if your I/O is fast enough Pop-in shouldn't happen in the first place.
 
Your displaying a higher res textuire over a lower one and belding it in so you dont get that POP.

The higher res texture has arrived late, or you would not be blending it over the poorer texture - there would be no need lol.

SFS is a blended pop in for data misses....But MS have PR and dressed it up to a multiplier - its sort of is, in that you can have a smaller streaming pool and you have a band aid so it does not look as bad when streaming fucks up.....

What do you think it is ? Must be magic :messenger_beaming:

That sounds like only good for old LOD's method. Most PS5 games should skip LOD's going forward.
 
Saw this image floating around, XVA better than PS5's I/O complex confirmed. Misinformation is around the corner. People just outright refuse to accept PS5 has an advantage over XSX. Funny enough, this is the same people that gets mad at others who can't accept the GPU advantage on XSX.
RCkP4lD.jpg

Holy shit! I'm impressed by Jason Ronald ( Director of Program Management at Xbox ) and his buzzwords, instead of Andrew Goossen ( Xbox System Architect ) who said that XSX SSD is 6 GB/s theoretical max. for decompression method. /s
 
Last edited:
There's no need to takie into account slower memory pool in real game scenario, because no developer will use more than 10 GB just for textures with only 13.5 GB available. So dont worry, XSX GPU will be well feed with tasty 560 GB/s BW :).
Yes XSX bus is capable of 560GB/s, but while the GPU is getting fed at that speed can you explain how the CPU and rest of the system gets access to the memory?

(Apologies I know this is from 30 pages back, I got shit loads of catching up to do!)
 
Last edited:
Theoretical (peak) is 22GB/s, 5.5GB/s is raw sustained, compressed without Oodle Texture is 8-9GB/s which is also sustained on average.

We don't know what throughput is with Oodle Texture. But im sure someone can extrapolate, without going wild.

To me it's pretty clear who focused on I/O more, evident by fixed function hardware, choice of SSD and Kraken license (you dont buy a license for every dev, for shits and giggles).

Sony wanted to close the gap between real performance and theoretical max. That's a hard sell and they most likely have a reason for this. (Horizon and Demon Souls, look to have pretty insane geometry).

Let me help you in some way :




I think it can.

Like twit says, it was used without Oodle

"The Sony test sets I know of (that were used to eval the Kraken HW during development) were all assets from (now several year old) PS4 games. So that figure is probably on non-Oodle Texture data."

127 MB block compressed GPU textures, mix of BC1-7

78 MB with zip/zlib/deflate

70 MB with Oodle Kraken

40 MB with Oodle Texture + Kraken

And this :




So, this means compression ratio can be more than 3 to 1. That also means compression can be around 17 GB/s
 
That's BC game not optimized to use VA features (different compression metchod, therefore XSX HW decompression chip is not working here). State of decay 2 loads equally as "fast" on PC with slow 500MB/s SSD and it's proof this game is not using VA features. Game loads faster compared to Xbox One X with standard HDD, but that's it.
So Microsoft did not even bother to optomize their demo to show off their velocity architecture , sfs and their ssd speed multiplier.
 
Your displaying a higher res textuire over a lower one and blending it in so you dont get that POP.

The higher res texture has arrived late, or you would not be blending it over the poorer texture - there would be no need lol.

SFS is a blended pop in for data misses....But MS have PR and dressed it up to a multiplier - its sort of is, in that you can have a smaller streaming pool and you have a band aid so it does not look as bad when streaming fucks up.....

What do you think it is ? Must be magic :messenger_beaming:
This isn't completely true. You still have the efficiency that you only load a sub portion of the mip. This still results in x2.5 memory efficiency that can be used for other stuff.
 
New shots of the UE 5 demo assets, It's still the best showcase of next gen possibilities by a wide margin.


bastian-hoppe-artstation-bh-07.jpg

bastian-hoppe-artstation-bh-08.jpg

bastian-hoppe-artstation-bh-09.jpg

bastian-hoppe-artstation-bh-10.jpg

bastian-hoppe-artstation-bh-04.jpg


jerome-platteaux-cave-screenshot-0025.jpg


Love those lips

I want to know yet, how megascans will be used in games.

Sony has their Films Division and now PlayStation can use those assets to input in a game. This also implies to their special effects division, and PlayStation can handle VFX in real time too.

But for now the Nanite belongs only to Epic and we don't have much information about Decima and other Sony Studios Engines about these technologies.
We know that Horizon FW is not using these cinematic scans, but we Know SpiderMan Miles Morales is using 4D scans (For the characters if I'm not wrong).

If we can input Sony Films Division with games, I think we will reach "movie quality" games and it will be amazing. I hope they show something like that to our next games in the next event (imagine a GOW and a ND like a movie it would be incredible).
 
That's BC game not optimized to use VA features (different compression metchod, therefore XSX HW decompression chip is not working here). State of decay 2 loads equally as "fast" on PC with slow 500MB/s SSD and it's proof this game is not using VA features. Game loads faster compared to Xbox One X with standard HDD, but that's it.




o5yvwbv.png

MS need to start showing, they blabbed about XVA on the demo.....
 
Yeah but its good XSX has it so third parties can push streaming and smaller streaming pool of data - the 1 second Cerny talks about....Ps5 does not need a back up as its IO is crazy.

From the sounds of it I'm guessing the PS5s I/O is easier to use since developers don't have to do anything unusual to take advantage of it.
 
This isn't completely true. You still have the efficiency that you only load a sub portion of the mip. This still results in x2.5 memory efficiency that can be used for other stuff.

Yes it is true, SFS is a hardware blended shader. Thats what it is to help when stuff arives a few frames late (Goosen).

All the benefits of 2.5 x is simply an estimate of a smaller streaming pool and people just dont get it.

Goosen quote :

A technique called Sampler Feedback Streaming - SFS - was built to more closely marry the memory demands of the GPU, intelligently loading in the texture mip data that's actually required with the guarantee of a lower quality mip available if the higher quality version isn't readily available, stopping GPU stalls and frame-time spikes. Bespoke hardware within the GPU is available to smooth the transition between mips, on the off-chance that the higher quality texture arrives a frame or two later.

Note the 768 mb streaming pool for UE5 demo that was billions of polygons.....But that is more complex than mips, memory saving = how little streaming pool you can play with. XSX has a back up plan, Ps5 is so fast it does not need one.

At least we can have 3rd party games that push streaming next gen...as long as they dont run on ps4 / xb1...

Anyway, off to bed byes, good night chaps.
 
Last edited:
I want to know yet, how megascans will be used in games.

Sony has their Films Division and now PlayStation can use those assets to input in a game. This also implies to their special effects division, and PlayStation can handle VFX in real time too.

But for now the Nanite belongs only to Epic and we don't have much information about Decima and other Sony Studios Engines about these technologies.
We know that Horizon FW is not using these cinematic scans, but we Know SpiderMan Miles Morales is using 4D scans (For the characters if I'm not wrong).

If we can input Sony Films Division with games, I think we will reach "movie quality" games and it will be amazing. I hope they show something like that to our next games in the next event (imagine a GOW and a ND like a movie it would be incredible).
Sony could use Atom View instead.

 
Sony could use Atom View instead.



You know whats funny. There is Atom view UE4 plugin. And then all of sudden we get a UE5 demo doing exact features that Atom view plugin did & more. All coated as collaboration with SONY for "years". The whole thing stinks.Its just all 1 big Sony production.

That acquisition happened was few years ago behind the scenes. Would be interesting to see how much more advance Atom View got since 2017. Its obviously going to be in the PS5/VR2. But boy PS Studios are going to pull off spectacular things. We are In for a treat with this tech.
 
Last edited:
Yes it is true, SFS is a hardware blended shader. Thats what it is to help when stuff arives a few frames late (Goosen).

Al the benefits of 2.5 x is simply an estimate of a smaller streaming pool and people just dont get it.

Goosen quote :



Note the 768 mb streaming pool for UE5 demo that was billions of polygons.....But that is more complex than mips, memory saving = how little streaming pool you can play with. XSX has a back up plan, Ps5 is so fast it does not need one.

At least we can have 3rd party games that push streaming next gen...as long as tghey dont run on ps4 / xb1...
You're not properly understanding what he is saying. The most important part is "intelligently loading in the texture mip data that's actually required". Also before that quote he also says "We found a game typically accessed at best only one-half to one-third of their allocated pages over long windows of time," says Goossen. "So if a game never had to load pages that are ultimately never actually used, that means a 2-3x multiplier on the effective amount of physical memory, and a 2-3x multiplier on our effective IO performance."

So sure the PS5 has faster speeds but it still loads the whole mip instead of only what is needed. So the PS5 is losing 60% of their bandwidth to stuff that is not needed. That's a lot...
 
So sure the PS5 has faster speeds but it still loads the whole mip instead of only what is needed.

You saying tiled resources,. partially resident textures and all that stuff is now proprietary to XSX.:messenger_winking:


Go read about megatextures, this is old shit. We had this all last gen.

Sounds like we are rereading the same stuff from 8 years ago..
 
Last edited:
I want to know yet, how megascans will be used in games.

Sony has their Films Division and now PlayStation can use those assets to input in a game. This also implies to their special effects division, and PlayStation can handle VFX in real time too.

But for now the Nanite belongs only to Epic and we don't have much information about Decima and other Sony Studios Engines about these technologies.
We know that Horizon FW is not using these cinematic scans, but we Know SpiderMan Miles Morales is using 4D scans (For the characters if I'm not wrong).

If we can input Sony Films Division with games, I think we will reach "movie quality" games and it will be amazing. I hope they show something like that to our next games in the next event (imagine a GOW and a ND like a movie it would be incredible).

I'm just hype for Final Fantasy 7:RE

Here is a snippet of how the assets are made useable in the demo

Timestamped

vKkHCF.gif
 
Last edited:
You saying tiled resources,. partially resident textures and all that stuff is now proprietary to XSX.:messenger_winking:


Go read about megatextures, this is old shit. We had this all last gen.

Sounds like we are rereading the same stuff from 8 years ago..
They put dedicated hardware in the Xbox one X to track this. So now Goossens is lying and this doesn't free up 60% of the memory bandwidth? They created something that was completely unnecessary and already existed? And on top of that they chose to rebrand it and market it? Yeah Ok... If you want to go all nutty fanboy go ahead, but I'll take the word of a lead Xbox engineer over you.
 
They put dedicated hardware in the Xbox one X to track this. So now Goossens is lying and this doesn't free up 60% of the memory bandwidth? They created something that was completely unnecessary and already existed? And on top of that they chose to rebrand it and market it? Yeah Ok... If you want to go all nutty fanboy go ahead, but I'll take the word of a lead Xbox engineer over you.

Goosen said the SFS hardware is a blended shader for mipmap, thats what it is.

Go read below. Partial loading of mipmaps was tiled resources / PRT, its already done in doom 4 etc and has been around since ps4 / xb1 .


From watchdogs

When using virtual texturing, you will only load the texture tiles into memory that are actually viewed by the virtual camera. The main benefit is that you can save on the amount of video memory that you need for you texture data, but there are a bunch of side benefits. Loading times are reduced and disk access is more constant and predictable. The bottom-line is that you can really increase the graphical fidelity of games, while staying within the limited of the current hardware.

If you can find anything in a link where MS have developed new hardware on silicon to do the well proven virtual textures / PRT (it has lots of names as been around a long time) then we can talk more.

Its mostly a sales pitch from MS and they chuck everything into the pot, it takes time to correctly pick out what MS actually did out of the sales pitch and it does confuse.
 
Last edited:
You are talking about 48 CUs GPU but with much slower clock, while I was thinking about 48 CUs with the same 2200 MHz clock speed

48 CUs 2200 MHz = 13.5 TF
48 CUs 2100 MHz = 12.9 TF
48 CUs 2000 MHz = 12.2 TF
48 CUs 1900 MHz = 11.6 TF
48 CUs 1800 MHz = 11 TF
48 CUs 1700 MHz = 10.4 TF

You want to tell me Sony would go even below 1700 MHz on 48 CUs GPU?
Honestly I'm not picking on you:messenger_peace:
Just wanna say 48 CU'S @ 2.2GHZ is not likely possible at all! They'll have to design for the theoretical maximum power it can draw and heat it will generate if all CU's get fully utilised (no matter how unlikely that is). The power supply will be bigger and cooling solution will likely be expensive and louder (if cool-able at all?), and can forget about Cerny's 'thermal equilibrium' optimisation

But after all that, yea its a good exercise to consider the what-if's :unsure:

Edit: I forgot to mention, even with Sony's variable frequency tech (I believe) it still won't be possible to push that many CU's at that frequency
 
Last edited:
This gif only shows how unoptimized the demo is. You should always delete what can't be seen, this only shows that if optimized it can easily run at native 4K instead, maybe 60fps as well. What a waste of resources, more like a stress test.

Unless it's doing what it's selling itself on.. Only rendering what is seen.
Ofcourse I agree this demo is a stress test and the tech is unoptimizaed but if they are actually demonstrating that building this way has negligible cost then that's a great thing. it shows how much faster people will be able to work.

it's all wait and see. can't wait to play around with it myself.
 
Last edited:
When compressing data, you choose a block size and potentially a number of other blocks which contain relevant require data.

think key frames in video compression...

yes you will have to have decompressed the key frame and likely one whole block.. but that might be very small depending on how the codec is configured. There is nothing stopping you knowing which blocks need to be decoded and only choose those ...

every codec has the same balancing issues... including of course ps5 ...
Well if that is the solution they are going for re: random access in zlib-esq Bcpack files - 64KB block decompressed per texel sample in a worst case - then the gains in texture residency and reduced compute shading of SFS aren't the multiplier benefit being stated.
The additional workload to enable random access will add latency in the computational cost of decompressing at such a coarse level for such fine granular data. Accessing each mipmap will also be equally costly and the effective gain in I/O from stacking variable and fix rate compression with virtualized access starts to look dubious IMHO.

The whole point of variable rate compressors used for gaming is to get excellent compression ratios (which RDO is focused on) and fast/lightweight processing decompression which the zlib derivatives like kraken, oodle, etc are doing.

In reality I don't see a typical usage for stacking SFS with random access streaming BCpack files from disk on XsX. Instead I see SFS used effectively with decompressed BCpack files resident in memory, or SFS used with random access streaming BCn files from disk. Which in both of those scenarios the multiplier is an either or not both stacking.
 
Unless it's doing what it's selling itself on.. Only rendering what is seen.
Ofcourse I agree this demo is a stress test and the tech is unoptimizaed but if they are actually demonstrating that building this way has no cost then that's a great thing. it shows how much faster people will be able to work.

it's all wait and see. can't wait to play around with it myself.

You are right. 20 million polygons is more than good enough for 6K natively, they could push that down to 8.5 million for less GPU taxing native 4K. I think that what Horizon FW is doing. Zoom in very close and you will see aliasing that is not visible at native resolution:

49996575967_22d44a7637_o.png


But you zoom in here you can't see that because it has overwhelming polygons that gives it a more film-like, realistic look.

1920x-1
 
They put dedicated hardware in the Xbox one X to track this. So now Goossens is lying and this doesn't free up 60% of the memory bandwidth? They created something that was completely unnecessary and already existed? And on top of that they chose to rebrand it and market it? Yeah Ok... If you want to go all nutty fanboy go ahead, but I'll take the word of a lead Xbox engineer over you.
What the Lead XBOX engineer leaves out is what he is using for comparison; is it a system not using SFS? Not using PRT/Tiled Resources? You are making assumptions on what he is comparing it against
 
This gif only shows how unoptimized the demo is. You should always delete what can't be seen, this only shows that if optimized it can easily run at native 4K instead, maybe 60fps as well. What a waste of resources, more like a stress test.
That's not how it works. First off, if the mesh isn't visible because it's being obscured it's not taking up any rendering performance, secondly, all the hidden meshes you saw are still being used elsewhere in the cave so they must remain in the RAM pool. Thirdly, the most likely reason it ran at 1440p is because Lumen isn't yet fully optimized for 60fps or 4K. Lastly, the game being rendered at 1440p is being up sampled to 4K through temporal accumulation. If the camera is still the image would resolve to true 4K over a couple of frames. This is a very smart and highly cost effective approach which I think many games will continue to utilize instead of wasting render power on brute forcing true 4K in the next gen.
 
I just sent £399 cash, fs man, that was my ps5 money.

And with that night night all, keep up the good fight.
Remember no shit posting.......with out me present. hugs.
Oh wow! I'm so jealous of you! I missed my window! You're gonna get SO MUCH MONEY BACK!
 
I don't think he's this dumb, I him and the other FUD suspects are doing it in purpose. If he's that dumb and the others, then I really feel sorry for them, still doubt it.

Just more 'fake news' to bring in hits? Yes, I suppose that's very likely. More and more seem to be doing the same these days. And I guess, if it's something you can always hide behind 'i made a mistake' or 'my sources were wrong/I misunderstood my sources' and it makes you money meanwhile, maybe it's worth it to them?
 
Status
Not open for further replies.
Top Bottom