Support NeoGAF

[NoMoChokeSJ]

No, it works like this.

Someone "matter of factly" misrepresents the "facts" as you say, while fanboys quote it and +1 it. It's terrible.

I especially love the "Ha, just as I predicted" overtones. As if anyone cares

I don't disagree that the dog piling is happening. But to be fair... You kinda reap what you sew.

 

[Faithless83]

Generally having a smaller pool of any resource is a Good Thing. Whether that is a pre-allocated resource, or a enumeration of things like, connections and so forth. The idea is you use them up as quickly as possible and return them to / reuse the pool. It conserves the underlying resource.

I was thinking back then once the demo was revealed, since the PS5 architecture is so different (and the UE5 demo details proves it) this may drive exclusives on the plataform just due to it not being possible in other hardware.
We may have a PS1 vs N64 FF7 situation again guys...

 

[JeloSWE]

He presents facts from a video by Epic, one that has new information, and you presented none.

DynamiteCop! doesn't seem to understand the meaning of the fact he presents. He ascribes meaning to the 768mb pool. No one outside of Epic or with decent experience with game development will fully comprehend how that potentially relates to the read speed of the SSD.

But lets try. 768mb is what sits in RAM at all times and is what holds the current assets needed to render the Nanite parts in the image.
At 60fps, 5.5GBs means we are reading 92mb per frame. You would need a read speed of 46GB/s to fill the 768mb pool every frame.
The point of a pool is not to fill it every frame but having quick access to needed assets.
But if we really need to because of entering a new area or map it would take 8 frames to do so at 60fts.
The latency/speed of the SSD thus helps in how fast you can bring in new data to the pool. Being able to swap the entirely pool at 8 frames is incredible, this means you could read in object behind you as you rotate the camera around or as objects emerge behind walls or building that were earlier obscured/culled. This greatly increases the potential for more world complexity.

Having a super fast SSD is going to be revolutionary, just not in all the obvious ways we are used to. The faster access you have to any storage the better it is.

Further, UE5, Nanite and Lumen will run beautifully on XSX but may not be able to quite match the amount/variation/details of assets brought in to RAM for rendering.

 

[Entroyp]

UE5 is turning to be mighty impressive. Can’t wait to see games using it.

 

[funking giblet]

But lets try. 768mb is what sits in RAM at all times and is what holds the current assets needed to render the Nanite parts in the image.
At 60fps, 5.5GBs means we are reading 92mb per frame. You would need a read speed of 46GB/s to fill the 768mb pool every frame.
The point of a pool is not to fill it every frame but having quick access to needed assets.
But if we really need to because of entering a new area or map it would take 8 frames to do so at 60fts.
The latency/speed of the SSD thus helps in how fast you can bring in new data to the pool. Being able to swap the entirely pool at 8 frames is incredible, this means you could read in object behind you as you rotate the camera around or as objects emerge behind walls or building that were earlier obscured/culled. This greatly increases the potential for more world complexity.

Having a super fast SSD is going to be revolutionary, just not in all the obvious ways we are used to. The faster access you have to any storage the better it is.

Further, UE5, Nanite and Lumen will run beautifully on XSX but may not be able to quite match the amount/variation/details of assets brought in to RAM for rendering.

Q.E.D

 

[IntentionalPun]

Ehh.. I REALLY don't think you are interpreting that correctly?

I think it's saying that they only need to reserve 768MB of RAM for streaming data in and out of memory... meaning most of the RAM was being used to render the view, and because you can pull data at 5.5GB / second you can fill that cache in a fraction of a second as the player turns.

That's only possible with extreme I/O; with less I/O you'd need a much bigger cache pool to compensate for not being able to pull data as the player moves/turns.

 

[mckmas8808]

It's going to be a shocker for some and for others like me not so much, because I've been saying this for months now. I hate to break it to some of you but that demo's data streaming could be handled by a 5 year old SATA SSD.

768MB is the in view streaming requirement on the hardware to handle that demo, 768 MEGABYTES... COMPRESSED. And what was the cost of this on the rendering end?

Well, this is the result...

This confirms everything I've said, not that these SSD's are useless, because they're 100% not. That data streaming would be impossible with mechanical drives, however, and this is a big however. That amount of visual data and asset streaming is already bottlenecking the renderer, it's bringing that GPU to its knees. There's very little cost to the CPU as you will see below, but as noted about 100 different times on this website and scoffed at constantly by detractors; the GPU will always be the limiting factor..

I've maintained this since square one, Microsoft and Sony both went overkill on their SSD's. That amount of I/O increase is not capable of aligning with the rendering pipeline in terms of the on demand volume of data streaming these SSD allow.

So what's the point here? You've got two systems with SSD's far more capable than their usefulness, but one came at a particularly high cost everywhere else in the system. I'll let you figure out which one that is and where.

Sounds like you know more about creating a console than Mark Cerny and the WHOLE MS hardware engineering team. Congrats

Spoiler: guess what

(on being the most simple-minded person on NeoGAF)

 

[thicc_girls_are_teh_best]

You’re conflating amount of data transferred with speed to transfer that data. It wouldn’t matter if the amount was 50 megabytes. It’s about latency and throughput.

aybe you don’t remember but N64 had a memory subsystem that could transfer 500 MB/second. What was the biggest cartridge that ever released on that system? 64MB? It was about speed and latency not the “amount” of data. This slide doesn’t change that at all.

You do not know more about the UE5 demo than Epic.

You do not know more about computer and system engineering than the PlayStation team.

There is a serious need to try and disprove Epics plainly stated facts about UE5 and PlayStation 5 for some reason. Do you really think they made the statements they did just a few weeks ago knowing, if they were lies, that it would then be discovered when they talked about it a little bit more? These people are not idiots. But they would have to be to be doing what you, the OP, is ascribing to them.

So yeah. Pretty sure everything is in the same place as before this presentation.

Damn, that was a fierce shifting of the goalpost!

Ref? How many points is that?

"It's exactly one point."

WAT?!? ONE MEASLY POINT?!!? Was it even worth it?

"Nope".

 

[mckmas8808]

Damn, that was a fierce shifting of the goalpost!

Ref? How many points is that?

"It's exactly one point."

WAT?!? ONE MEASLY POINT?!!? Was it even worth it?

"Nope".

But you know the OP is wrong right? So why fake the funk?

 

[Faithless83]

If I was the OP, I would have edited the post with:
"Sorry, english is not my first language and I misinterpreted that. Please move on."

 

[thicc_girls_are_teh_best]

But you know the OP is wrong right? So why fake the funk?

Even if the OP might be wrong on some aspects of their interpretation, it doesn't make the people using that to reaffirm their prior biases correct, either. So I'm seeing some people go "Well if the OP was wrong, it means my extreme opinion is still valid!!". When by and large, it was probably never valid, itself.

So I just it all kinda funny. Some people really seem to only want this demo to be capable on only one next-gen system, and maxing out its I/O capabilities at that. A tech demo...before the generation even begins....doing that. Do some people know what they are even asking for? That wouldn't bode particularly well for usage of this in actual games once the generation begins, not at the level seen in the demo, anyway.

Hell we can go back to the T-Rex demo on PS1. Amazing T-Rex for its day, but that clearly maxed out the system's capabilities just for that one thing. Not a single actual game came anywhere close to that visual fidelity in real-time. You'd THINK people would want real games pushing for that level of asset streaming to be a thing next-gen in actual gameplay scenarios, but that isn't necessarily going to happen if the demo is already doing so much it can only work on one system and is stressing out even that system's I/O.

DynamiteCop! doesn't seem to understand the meaning of the fact he presents. He ascribes meaning to the 768mb pool. No one outside of Epic or with decent experience with game development will fully comprehend how that potentially relates to the read speed of the SSD.

But lets try. 768mb is what sits in RAM at all times and is what holds the current assets needed to render the Nanite parts in the image.
At 60fps, 5.5GBs means we are reading 92mb per frame. You would need a read speed of 46GB/s to fill the 768mb pool every frame.
The point of a pool is not to fill it every frame but having quick access to needed assets.
But if we really need to because of entering a new area or map it would take 8 frames to do so at 60fts.
The latency/speed of the SSD thus helps in how fast you can bring in new data to the pool. Being able to swap the entirely pool at 8 frames is incredible, this means you could read in object behind you as you rotate the camera around or as objects emerge behind walls or building that were earlier obscured/culled. This greatly increases the potential for more world complexity.

Having a super fast SSD is going to be revolutionary, just not in all the obvious ways we are used to. The faster access you have to any storage the better it is.

Further, UE5, Nanite and Lumen will run beautifully on XSX but may not be able to quite match the amount/variation/details of assets brought in to RAM for rendering.

So we're assuming it's pulling in 183 MB per frame? Just because?

Yes it's 183 MB because the demo ran at 30 FPS, not 60 FPS. Where are you getting the demo was running at 60 FPS? The 30 was capped.

 

[mckmas8808]

Even if the OP might be wrong on some aspects of their interpretation, it doesn't make the people using that to reaffirm their prior biases correct, either. So I'm seeing some people go "Well if the OP was wrong, it means my extreme opinion is still valid!!". When by and large, it was probably never valid, itself.

So I just it all kinda funny. Some people really seem to only want this demo to be capable on only one next-gen system, and maxing out its I/O capabilities at that. A tech demo...before the generation even begins....doing that. Do some people know what they are even asking for? That wouldn't bode particularly well for usage of this in actual games once the generation begins, not at the level seen in the demo, anyway.

The bolded is extremely weird to me. I don't get it. It's odd.

 

[Elog]

So I just it all kinda funny. Some people really seem to only want this demo to be capable on only one next-gen system, and maxing out its I/O capabilities at that. A tech demo...before the generation even begins....doing that. Do some people know what they are even asking for? That wouldn't bode particularly well for usage of this in actual games once the generation begins, not at the level seen in the demo, anyway.

You know that is not true.

This whole debate has been going through the following loops:

1) Microsoft claims they have the best system for graphics the coming generation as the PS5 specs are released. This is also reinforced by every forum warrior out there. This analysis is more less only focused on Tflops.
2) Cerny's talk starts to disseminate and he claims that I/O will be a more important part of upgrading graphics the coming generation than what people think and spells out how much silicon Sony has dedicated in the PS5 to this.
3) Forum warriors laugh at Cerny as if he does not know what he is talking about since I/O does nothing according to them apart from faster load times.
4) UE5 demo is shown and TS says that I/O is key in making the demo as graphically rich as it is.
5) Slow realization that Cerny might (!) know what he is talking about rolls through the internet's underbelly.
6) MS starts to move their marketing from Tflops to Velocity Architecture etc - they seem to realize this as well.
7) Today the OP erroneously tries to state that the demo is not I/O limited in any way apart from requiring an SSD which both systems have. He reaches this conclusion by adding an apple to an orange and dividing it by pi (basically BS).

Questions still remain how much impact the I/O differences between the systems will have - both for third party and first party titles (personally I do not believe it will have an impact unless the game is designed to take advantage of it which most likely will be hard outside of first party titles). However, no one argues that UE5 will be able to run on a variety of systems but unless you chose not to believe TS the graphics in the demo will take some sort of hit on the XSX due to lower I/O capabilities.

 

[JeloSWE]

So we're assuming it's pulling in 183 MB per frame? Just because?

Yes it's 183 MB because the demo ran at 30 FPS, not 60 FPS. Where are you getting the demo was running at 60 FPS? The 30 was capped.

I'm not assuming, I gave a math example for UE running at 60fps which Nanite can do. It's only Lumen that were currently limiting the fps.

184 MB or 92 MB doesn't matter too much as it's the same amount of data per second. It might also be much lower in reality as the SSD budget may be spent for other thing to be read.

 

[Allandor]

You know that is not true.

This whole debate has been going through the following loops:

1) Microsoft claims they have the best system for graphics the coming generation as the PS5 specs are released. This is also reinforced by every forum warrior out there. This analysis is more less only focused on Tflops.
2) Cerny's talk starts to disseminate and he claims that I/O will be a more important part of upgrading graphics the coming generation than what people think and spells out how much silicon Sony has dedicated in the PS5 to this.
3) Forum warriors laugh at Cerny as if he does not know what he is talking about since I/O does nothing according to them apart from faster load times.
4) UE5 demo is shown and TS says that I/O is key in making the demo as graphically rich as it is.
5) Slow realization that Cerny might (!) know what he is talking about rolls through the internet's underbelly.
6) MS starts to move their marketing from Tflops to Velocity Architecture etc - they seem to realize this as well.
7) Today the OP erroneously tries to state that the demo is not I/O limited in any way apart from requiring an SSD which both systems have. He reaches this conclusion by adding an apple to an orange and dividing it by pi (basically BS).

Questions still remain how much impact the I/O differences between the systems will have - both for third party and first party titles (personally I do not believe it will have an impact unless the game is designed to take advantage of it which most likely will be hard outside of first party titles). However, no one argues that UE5 will be able to run on a variety of systems but unless you chose not to believe TS the graphics in the demo will take some sort of hit on the XSX due to lower I/O capabilities.

They always mentioned the velocity architecture even before ps5 'specs' were released.
both have more than capable storage solutions. But fast storage (like other systems specs) are only useful up to a specific point where other parts of the system become a bottleneck. MS took that solution and ps5 that. Both are good solutions as far as we can tell.

 

[Psykodad]

Even if the OP might be wrong on some aspects of their interpretation, it doesn't make the people using that to reaffirm their prior biases correct, either. So I'm seeing some people go "Well if the OP was wrong, it means my extreme opinion is still valid!!". When by and large, it was probably never valid, itself.

So I just it all kinda funny. Some people really seem to only want this demo to be capable on only one next-gen system, and maxing out its I/O capabilities at that. A tech demo...before the generation even begins....doing that. Do some people know what they are even asking for? That wouldn't bode particularly well for usage of this in actual games once the generation begins, not at the level seen in the demo, anyway.

Hell we can go back to the T-Rex demo on PS1. Amazing T-Rex for its day, but that clearly maxed out the system's capabilities just for that one thing. Not a single actual game came anywhere close to that visual fidelity in real-time. You'd THINK people would want real games pushing for that level of asset streaming to be a thing next-gen in actual gameplay scenarios, but that isn't necessarily going to happen if the demo is already doing so much it can only work on one system and is stressing out even that system's I/O.



So we're assuming it's pulling in 183 MB per frame? Just because?

Yes it's 183 MB because the demo ran at 30 FPS, not 60 FPS. Where are you getting the demo was running at 60 FPS? The 30 was capped.

I don't think anybody claimed that the tech-demo can't run on XSX or PC or a laptop at all and those that do, completely miss the purpose of the UE5 tech.
It's made to run on any system.

So either you have the odd one out in mind (that did in fact claim the demo is only possible on PS5), or you're misinterpreting whatever's being said.

 

[Tesseract]

 

[FacelessSamurai]

Not only that, they also say that during the scene where she flies through the city, they were streaming in 500K objects and they can easily stream 1 Million.

I don't know too much about tech, but I'm pretty sure, going by this, that they weren't maxing out PS5 at all.
Or that the GPU was a bottleneck, if that's a better way of saying it.

Then why couldn’t the dynamic resolution reach 4K? If it wasn’t an issue the GPU would have pushed 4K, but it couldn’t. I somehow doubt they locked it at a max of 1440p just because.

 

[Deleted member 17706]

Awesome news for everyone. If stuff like that is possible on conventional SSDs, then I'm really excited to see what's possible this gen.

 

[Mister Wolf]

I

You clearly don't understand how Lumen works. It' only uses low resolution Voxels for far away geometry, closer up and medium sized ones uses Signed Distance Fields and the the finer details uses Screen Space GI. Also the GI has infinite bounces by doing a new bounce each frame and constantly accumulating over time. This solution will require little to NO effort on the artists side and will just work. They are aiming for 60fps on next gen and is very likely an achievable target.

No I do get it. The voxels they use are the most demanding/accurate and the closer the objects get they switch to even more simpler representations of the geometry. I simply call it voxel GI because I don't give a fuck about the screen space or signed distance field aspect. It is just a cost cutting poor man's alternative to triangle raytracing global illumination(Metro, Control, Cyberpunk, Dying Light 2) and not as accurate same as the Voxel GI used in CryEngine games like Hunt Showdown.

 

[Psykodad]

Then why couldn’t the dynamic resolution reach 4K? If it wasn’t an issue the GPU would have pushed 4K, but it couldn’t. I somehow doubt they locked it at a max of 1440p just because.

Nobody ever said there isn't a ceiling for the GPU.

The discussion right now is to what extent PS5 benefits from it's SSD.
From what I understand, it still seems that PS5 will likely be able to stream/load more assets/details/whatever, but at a lower res/performance.
Some people explained some stuff that also seems to support this and that is probably what Epic was talking about when they say it's only possible on PS5.
I.e. the tech-demo can run on other platforms, but the loading/streaming (or whatever it's called) will be better on PS5 and thus will only be reached there.

In any case, I personally don't care much for the demo being 1440p, because the visuals are so ridiculously good imo, that it easily blows away any current-gen native 4K game.
Resolution isn't everything.

 

[thicc_girls_are_teh_best]

The bolded is extremely weird to me. I don't get it. It's odd.

Don't get it as in it's weird to present the idea, or don't get it as in it's weird there are people who actually seem to discuss the tech along these terms?

You know that is not true.

This whole debate has been going through the following loops:

1) Microsoft claims they have the best system for graphics the coming generation as the PS5 specs are released. This is also reinforced by every forum warrior out there. This analysis is more less only focused on Tflops.
2) Cerny's talk starts to disseminate and he claims that I/O will be a more important part of upgrading graphics the coming generation than what people think and spells out how much silicon Sony has dedicated in the PS5 to this.
3) Forum warriors laugh at Cerny as if he does not know what he is talking about since I/O does nothing according to them apart from faster load times.
4) UE5 demo is shown and TS says that I/O is key in making the demo as graphically rich as it is.
5) Slow realization that Cerny might (!) know what he is talking about rolls through the internet's underbelly.
6) MS starts to move their marketing from Tflops to Velocity Architecture etc - they seem to realize this as well.
7) Today the OP erroneously tries to state that the demo is not I/O limited in any way apart from requiring an SSD which both systems have. He reaches this conclusion by adding an apple to an orange and dividing it by pi (basically BS).

Questions still remain how much impact the I/O differences between the systems will have - both for third party and first party titles (personally I do not believe it will have an impact unless the game is designed to take advantage of it which most likely will be hard outside of first party titles). However, no one argues that UE5 will be able to run on a variety of systems but unless you chose not to believe TS the graphics in the demo will take some sort of hit on the XSX due to lower I/O capabilities.

No, I've observed as such. Some people may not necessarily outright state it that clearly, of course. But there's a tone you can pick up in a person's use of language, what type of points they stick to talking to, when they do so and how, and it can start to paint a picture. It's just something people do a lot, granted that happens on both sides of the aisle too.

1) Yeah that happened, but that was the fault of forum warriors only focused on TFs, and that was their choice. I was there in those threads. I kept trying to say "TFs don't mean everything", because both sides were having an e-dick competition for biggest TF count. It only got worst when the Github leak happened and the testing data kept coming forward. There are people to this day who pretend Ariel and Oberon are not PS5 chips simply because final TF and clock speeds are different. I'm looking forward to the phantom PS5 chip to finally make its appearance if that's the case.

2) Yes he did this, and it was a great thing to see. It was what was needed at the time, for sure. Granted, he also did try downplaying TFs in the process and imply that a smaller, narrower GPU is better at faster clock than a wider GPU at slightly lower clock, despite evidence of almost all GPU benchmarks between cards of the same architecture showing the opposite in almost every category aside from pixel fillrate and cache speeds.

So he basically did a downplay by giving a half-truth, on a particular architecture feature they didn't have an advantage in regards pure numbers. That is something you do as PR control no matter how technical the rest of your dissertation is.

3) It was wrong of people to dismiss the advantages of SSDs based simply on PC benchmarks, that's true. Was clear even before Road to PS5 that SSDs would bring a lot of I/O advantages going into next-gen, but that presentation opened the floodgates on encouraging that discussion in earnest.

TBF yes, some of that kickback was from PC and Xbox zealots who probably felt a bit cut down after seeing Sony's SSD I/O presentation. But from what I observed, they only got a bit particular about it when pro-Sony fans tried spinning the SSD I/O as a means of closing the graphics gap, and in many cases exceeding graphics on Xbox and PC platforms, without understanding the actual role of the SSD in a console design hierarchy. It just all kept feeding into a vicious feedback loop.

4) Yeah, TS said this...although he did a LOT of PR fluff into his statements as well. That went conveniently overlooked by people who used his statements to solidify their perception on the next-gen SSD I/Os even going as far to say that non-Sony platforms could indeed never run a demo like that (which I mean if you want to argue semantics yeah, because the demo has no binary compiler to run on a system other than PS5 xD).

5) Here's the thing; people (especially certain people) take Cerny's view on resolving the I/O issues as being the ONLY solution. That is where all of those people are wrong. Throughout the history of technology there have always been MULTIPLE solutions to the same problem, some just being more successful than others and not always because they were the objectively better solution, at that. x86's success, in fact, can be argued as being down more to ubiquity and mass market saturation (thanks to PCs) than being just an objectively superior solution, since there were always other architectures that did certain things much better for specific tasks that required them (M68K for example having 32-bit registers at a time when x86 processors did not).

The same thing applies with the SSD I/O solutions. Neither Sony's nor Microsoft's approaches are the ONLY approaches valid in being taken. They can each work and offer comparable results based on their design principals. I feel that some people don't understand the nuances in this which is why they keep clinging to seeing these as apples-to-apples solutions and clinging to paper specs so much, when in fact the paths the approaches take are pretty divergent and have different guiding philosophies to them.

Sony's approach maximizes a focus on bandwidth, Microsoft's maximizes a focus on latency. This doesn't mean they are necessarily lacking in the other area, just that their main priority is in one of them. They are both very valid approaches but a large contingent of people view Cerny's approach as the only possible solution. The truth is that he is not the first person to notice "Hm, there's some bottlenecks here. Let's try fixing them!", not even by a long shot. And he is not the only one who has found solutions to this problem, either.

Yes, certain solutions may excel in specific areas and use-cases over others; the fact remains there are always multiple valid solutions to addressing similar problems and it has been this way in technology since dang near its inception. It will remain that way into the future as well, minus some market capital tomfoolery or anything driven by money pushing people to a standardized solution.

6) MS was actually talking about XvA at least from the time Sony talked about their SSD I/O. In fact in that same blog post the day of Sony's Road to PS5, they mention XvA and other things like DirectML right there. But guess what? It's everyone else who were obsessed over TFs still, that skipped past that stuff and only focused on the TF count.

So Road to PS5 gets underway and people are still clinging to TFs. Then Sony announces theirs and (keep in mind I was in the Youtube stream when this happened) there's just a flood of "WTF!", "Lol", and other type of comments. So that shows you where these people's headspace was at, even up to that point. However, once Sony starts delving deep into their SSD I/O, you get some people who might've actually remember MS's stated performance numbers, kept up with some of the other rumors etc...they see Sony's stuff here and now that presents a new avenue for them to focus their discussions on (some doing so as more console warrior BS, obviously).

It's just very ironic that it took Road to PS5 for people at large to finally focus on something aside from TF, so the reason MS played into TFs early on is actually because the majority of gamers themselves were only paying attention to TFs! They were just giving people the talking points they wanted, but they did find ways to drop in mentions of things aside from TFs well before Road to PS5, including XvA, Project Acoustics, DLI etc. You probably just didn't pay attention to that since you might've been obsessed over GPU TFs. It's essentially tunnel vision, and a lot of people had it (and still do, they've just shifted that tunnel vision to SSD I/O instead).

7) I'll admit OP looked at the 768 MB figure incorrectly, but what do we see people doing now? Doubling down again on "you can only do this on PS5 after all", or some variation of that. It's like there is zero middle ground in this discussion anymore. Some people implying you'd need "magnitudes larger RAM reserve" for "other systems" to pull off the same thing...sneaky doublespeak like that it's been creeping back up once again.

Now, I still assert that there's nothing in this demo that cannot be done on at least the Series systems; they have pretty much all of the same focuses on smart bandwidth, very low latency (perhaps moreso), decompression etc. as PS5, just with lower overall maximum bandwidths. PC could be a case of where it may not be possible due to I/O and file system limitations, at least temporarily, but those will likely be resolved in the near future.

The thing is it would serve the community best if indeed this type of demo can run on other platforms because if this is the peak of asset streaming, and we're getting it in a demo before the generation even starts, that's kind of blowing the load too early. I'd like to think the generation, in terms of fast asset streaming capabilities, will have more to offer in real gameplay than what was seen in the UE5 demo.

I'm not assuming, I gave a math example for UE running at 60fps which Nanite can do. It's only Lumen that were currently limiting the fps.

184 MB or 92 MB doesn't matter too much as it's the same amount of data per second. It might also be much lower in reality as the SSD budget may be spent for other thing to be read.

You're still applying your math as an assumption, the assumption being they are constantly streaming in new data the whole time. If that's indeed what they're doing, it's ultimately a waste of resources WRT to an actual gameplay scenario.

As a controlled demo with limited physics and no advanced AI, physics, world logic etc. systems running in tandem (nor NPCs), on the merits of being technologically impressive it certainly fits that bill.

 

[Handy Fake]

I still don't think folk are understanding the advantages and implications of the PS5 IO and SSD.

 

[Esppiral]

But, but but I've been told that the SSad is magic and it can make render Pixar like graphics no matter the GpU I've been fooled?

 

[Jigsaah]

It's going to be a shocker for some and for others like me not so much, because I've been saying this for months now. I hate to break it to some of you but that demo's data streaming could be handled by a 5 year old SATA SSD.

768MB is the in view streaming requirement on the hardware to handle that demo, 768 MEGABYTES... COMPRESSED. And what was the cost of this on the rendering end?

Well, this is the result...

This confirms everything I've said, not that these SSD's are useless, because they're 100% not. That data streaming would be impossible with mechanical drives, however, and this is a big however. That amount of visual data and asset streaming is already bottlenecking the renderer, it's bringing that GPU to its knees. There's very little cost to the CPU as you will see below, but as noted about 100 different times on this website and scoffed at constantly by detractors; the GPU will always be the limiting factor..

I've maintained this since square one, Microsoft and Sony both went overkill on their SSD's. That amount of I/O increase is not capable of aligning with the rendering pipeline in terms of the on demand volume of data streaming these SSD allow.

So what's the point here? You've got two systems with SSD's far more capable than their usefulness, but one came at a particularly high cost everywhere else in the system. I'll let you figure out which one that is and where.

I mean the SSD's are far more capable than their usefulness in a demo. What about a completed game at 60 fps 4k. Also we don't expect the UE5 demo to be the pinnacle of performance next gen right? I mean I would think things would get better than that eventually, maybe requiring more SSD data streaming, right?

 

[sendit]

Good morning, this thread aged well.

 

[Hendrick's]

I still don't think folk are understanding the advantages and implications of the PS5 IO and SSD.

That's because they are all theoretical. Until we see this supposed advantage in action, it's just marketing.

 

[Handy Fake]

That's because they are all theoretical. Until we see this supposed advantage in action, it's just marketing.

Well, you'll get no argument from me there chief.

 

[JeloSWE]

You're still applying your math as an assumption, the assumption being they are constantly streaming in new data the whole time. If that's indeed what they're doing, it's ultimately a waste of resources WRT to an actual gameplay scenario.

As a controlled demo with limited physics and no advanced AI, physics, world logic etc. systems running in tandem (nor NPCs), on the merits of being technologically impressive it certainly fits that bill.

I'm not sure we are at odds here, I simply gave an explanation of how I think the 768mb pool is being utilized and how quickly it can be filled with uncompressed SSD speed of 5.5GB/s. By no means do I think it will use the full bandwidth exclusively as there are more things needed to be read than simply the Nanite geometry data. And it being able to output at 4.5ms to GBuffer is really great, it's well withing the frame time of 16.7ms for 60fps and even 8.3ms for 120fps.

Those performance number leaves us with a lot of time to dedicate to the other tasks you mentioned.

I should probably state that I'm not for or against PS5, I also don't expect Nanite to require a PS5 SSD to do impressive stuff. A faster or slower SSD will only dictate what detail level you can bring in new geometry, on the fly, for rendering, that's all.

 

[DynamiteCop!]

I mean the SSD's are far more capable than their usefulness in a demo. What about a completed game at 60 fps 4k. Also we don't expect the UE5 demo to be the pinnacle of performance next gen right? I mean I would think things would get better than that eventually, maybe requiring more SSD data streaming, right?

The issue isn't necessarily the amount of data although there is still a rendering hit, it's the knock on effect and hit to the rendering pipeline which is the concern.

All of this requires more accurate and a larger volume of GI, shadows, AF, AA, and case dependent RT. It's crippling to a GPU.

 

[jaysius]

I'm guessing it's all just future proofing then, and justfying the price tags, they didn't want to up the GPU/CPU to a reasonable level this gen so they both came up with the SSD lie. It's the worst marketing yet.

 

[thebigmanjosh]

Man this got out of hand quickly.

The 768MB is referring to the geometry budget in frame, and the SSD speeds will only affect how quickly that data can be replaced.

Its impossible to make accurate direct comparisons without knowing more, but here is the breakdown on zone budgets (they had up to 5 preloaded) in Spider-Man for reference:

Seeks are gone, data duplication is gone, RAM bandwidth has increased by 2X+, I/O throughout has increased by 50X-100X, render budgets could be increasing by 40X, and some of you are still focused on marketing and buzzwords.

 

[itsjustJEFF]

The pool is in RAM, it's per updated view from disk with just whats needed, nothing about "per second". It's data that's in view, and potentially per frame updates to this pool.
The latency of the SSD helps more than the pure throughput here, but you didn't get those numbers so...
Also 4.5ms is well within frametime (so not a bottleneck), I think Lumen is the bottleneck here, as it pushed it down to 30fps.

Why did you make this thread?

No one in here will reply to you.

Pure ignorance.

The 768MB working set of RAM dedicated to nanite streaming is so small, precisely because the I/O bandwidth is so high + they could leverage dedicated hardware compression.

What this means is you have a shit ton of headroom for everything else:

- Lumen acceleration structures, shaders, materials, BDRFs
- Non-nanite assets (mirror-reflective or translucent assets, skinned assets etc)
- Animation data, collision data, auxiliary physics data
- Audio data, gameplay scripts
- Niagra data, scripts, materials, anim data
- Everything else a modern game may need

Without the kind of I/O bandwidth next gen consoles provide that 768MB would likely be something like ~12-18GB to make up for the extreme latency, in-order to ensure you can stream the data to make available for rendering.

So, what is the correlation between 768MB of that data in RAM and the SSD speed? I'm not a tech wizard or anything, not like 80% of the people in this thread. The main point of this thread is that the 768MB in RAM is proof that the SSD are over engineered and the console makes are idiots. In the video the guy specifically says "streaming from RAM" do I just not understand how that relates to SSD and I/O?

So far, I don't think anyone has explained how 768MB of data in RAM is connected to SSD and I/O speeds.

 

[JeloSWE]

No I do get it. The voxels they use are the most demanding/accurate and the closer the objects get they switch to even more simpler representations of the geometry. I simply call it voxel GI because I don't give a fuck about the screen space or signed distance field aspect. It is just a cost cutting poor man's alternative to triangle raytracing global illumination(Metro, Control, Cyberpunk, Dying Light 2) and not as accurate same as the Voxel GI used in CryEngine games like Hunt Showdown.

But that is the impressive part. It will greatly increase visual fidelity for a lower rendering cost than brute forced triangel raytracing. This tech will probably scale well across hardware as well. The visual quality of the demos GI is incredible looking with all the onscreen detail brought on by Nanite. You are having very smal surface detail casting shadows and receiving indirect lighting. This can be done in a similar fashion with RTX and path tracing but those examples uses far less complex scene geometries such as Quake2 and Minecraft. The furute will certainly do perfect raytracing but this approach is very smart and yields great results.

 

[hyperbertha]

If 768 mb is per frame, 768x30 is about 22 gb/s which lines up perfectly with the PS5's IO 22 Gb/s throughput. If that means the IO is maxed out and fully utilized, it can only mean the Xbox can't run that demo without downgrades, provided this is the right interpretation.

 

[Kerlurk]

Gaming Discussion

Video game news, industry analysis, sales figures, deals, impressions, reviews, and discussions of everything in the medium, covering all platforms, genres, and territories.

www.neogaf.com

 

[funking giblet]

No one in here will reply to you.

So, what is the correlation between 768MB of that data in RAM and the SSD speed? I'm not a tech wizard or anything, not like 80% of the people in this thread. The main point of this thread is that the 768MB in RAM is proof that the SSD are over engineered and the console makes are idiots. In the video the guy specifically says "streaming from RAM" do I just not understand how that relates to SSD and I/O?

So far, I don't think anyone has explained how 768MB of data in RAM is connected to SSD and I/O speeds.

As you do not need to store data for a very long time, just for the frame you are drawing, you can store less of it as you can just swap (some of) it out when required. If you had a slow disk, this wouldn't be possible as you risk not loading the data in time, so you would store more of it instead. You don't need a lot of data to draw a frame if you are smart around the assets that are loaded, (partial textures / meshes etc), but you need to be able to swap it out quickly to react to changes between frames.

 

[funking giblet]

If 768 mb is per frame, 768x30 is about 22 gb/s which lines up perfectly with the PS5's IO 22 Gb/s throughput. If that means the IO is maxed out and fully utilized, it can only mean the Xbox can't run that demo without downgrades, provided this is the right interpretation.

It's the pool of data they store data in, they may not replace all of it per frame, but certainly there can be times where they replace a lot of it. A lot of data would also be preserved between frames too.

 

[Neo_game]

IMHO the unreal demo is blow way out of proportion by both parties. It is just a tech demo. That games is not going to release. So I do not think too much importance should be given. What I like was that it was 1440P which IMO is what the next gen console should target. LOD was very impressive and lighting as well there was no need for RT. Hopefully devs will look into this.

 

[ThatOneGrunt]

Good morning, this thread aged well.

I was reading the first few pages of comments and thought, I love this post. I'm so happy. That maybe it will stop all the meandering discussions of the PS5's ssd. And then I jumped to the last page of comments and felt sad again.

 

[JeloSWE]

No one in here will reply to you.

So, what is the correlation between 768MB of that data in RAM and the SSD speed? I'm not a tech wizard or anything, not like 80% of the people in this thread.

It's the other way round, 80% here aren't tech-wizards.

The main point of this thread is that the 768MB in RAM is proof that the SSD are over engineered and the console makes are idiots. In the video the guy specifically says "streaming from RAM" do I just not understand how that relates to SSD and I/O?
So far, I don't think anyone has explained how 768MB of data in RAM is connected to SSD and I/O speeds.

You are right, because no one truly knows besides Epic how they are related in the specifics, but a couple of people and I have tried to give plausible explanations how SSD speed, 768mb RAM pool could relate.

The way it works is: SSD -> RAM, CPU then prepares data for the GPU to render -> send data to GPU and it renders it.
There are two bandwidth speeds there, one is SSD to RAM 5.5GB/s and the other one is CPU and GPU access to RAM at 448GB/s.
What is being discussed is how quickly the 768mb RAM pool can be refreshed, which is misleading as it won't be completely refilled every frame. But a high SSD speed means you can add large amount of new data to the pool compared to slower SSDs. Whether the PS5 SSD speeds are the bare minimum for the UE5 demo isn't entierly know yet. All we are doing is speculating but some make worse attempts than others.

 

[thicc_girls_are_teh_best]

If 768 mb is per frame, 768x30 is about 22 gb/s which lines up perfectly with the PS5's IO 22 Gb/s throughput. If that means the IO is maxed out and fully utilized, it can only mean the Xbox can't run that demo without downgrades, provided this is the right interpretation.

It's not though, and it's a somewhat flawed interpretation for a couple of reasons.

1)There's no guarantee they are moving in new assets every single frame. That goes for not only textures but poly models too.

2)PS5's 22 GB/s is a theoretical peak based on "particularly good compression", but essentially lossy compression. Data that can be handled well at very high compression rates, like video files, can hit this range. Most types of texture asset files will do extremely poorly in this range because of lossiness that comes with compression beyond 2:1 (PS5's 8-9 GB is more around 1.75:1 compression range but still lossless), plus the fact the ratio is just very high so data and quality are lost.

By no means do I think it will use the full bandwidth exclusively as there are more things needed to be read than simply the Nanite geometry data.

I would hope it didn't though if it did, at least that would be justified by it being a demo.

In terms of how systems with lower or different I/O systems can handle it, that really just depends. We know MS have talked a lot about texture upscaling. Neither system has Tensor cores for DLSS operations, but I'd picture MS were aware enough of that feature and the needs it could fulfill in data markets (such as servers, which is where they also want to leverage Series X) that they'd have designed some assisting silicon in addition to the various INT support they've done with AMD on the CUs to accommodate for texture upscaling at as small a performance penalty as possible.

So even with lower bandwidth, systems employing that type of hardware could essentially reproduce that level of quality by upscaling low-resolution textures. I've also been thinking this might be something being leveraged by the Series' GPUs in SFS hardware implementation of mip blending transitions from lower-quality mipmaps to the higher-quality ones. But I guess we shall shee.

 

[mckmas8808]

Don't get it as in it's weird to present the idea, or don't get it as in it's weird there are people who actually seem to discuss the tech along these terms?

It's weird that people discuss the tech along those terms. Like both consoles are SUPER close to each other. Neither is going to really be able to do things the other can't. It's mainly going to come down to who has the better in-house devs and which console has the better exclusives and cool extras stuff not related to console power (i.e. Gamepass, PSVR, etc).

 

[Allandor]

You know that is not true.

This whole debate has been going through the following loops:

1) Microsoft claims they have the best system for graphics the coming generation as the PS5 specs are released. This is also reinforced by every forum warrior out there. This analysis is more less only focused on Tflops.
2) Cerny's talk starts to disseminate and he claims that I/O will be a more important part of upgrading graphics the coming generation than what people think and spells out how much silicon Sony has dedicated in the PS5 to this.
3) Forum warriors laugh at Cerny as if he does not know what he is talking about since I/O does nothing according to them apart from faster load times.
4) UE5 demo is shown and TS says that I/O is key in making the demo as graphically rich as it is.
5) Slow realization that Cerny might (!) know what he is talking about rolls through the internet's underbelly.
6) MS starts to move their marketing from Tflops to Velocity Architecture etc - they seem to realize this as well.
7) Today the OP erroneously tries to state that the demo is not I/O limited in any way apart from requiring an SSD which both systems have. He reaches this conclusion by adding an apple to an orange and dividing it by pi (basically BS).

Questions still remain how much impact the I/O differences between the systems will have - both for third party and first party titles (personally I do not believe it will have an impact unless the game is designed to take advantage of it which most likely will be hard outside of first party titles). However, no one argues that UE5 will be able to run on a variety of systems but unless you chose not to believe TS the graphics in the demo will take some sort of hit on the XSX due to lower I/O capabilities.

They always mentioned the velocity architecture even before ps5 'specs' were released.
both have more than capable storage solutions. But fast storage (like other systems specs) are only useful up to a specific point where other parts of the system become a bottleneck. MS took that solution and ps5 that. Both are good solutions as far as we can tell.

 

[Jigsaah]

The issue isn't necessarily the amount of data although there is still a rendering hit, it's the knock on effect and hit to the rendering pipeline which is the concern.

All of this requires more accurate and a larger volume of GI, shadows, AF, AA, and case dependent RT. It's crippling to a GPU.

Even a 12 Teraflop GPU?

 

[Amiga]

It's weird that people discuss the tech along those terms. Like both consoles are SUPER close to each other. Neither is going to really be able to do things the other can't. It's mainly going to come down to who has the better in-house devs and which console has the better exclusives and cool extras stuff not related to console power (i.e. Gamepass, PSVR, etc).

on the contrary the PS5 will be more capable in a technical level. their will be game designs that won't run on XS. but the games that stick to traditional designs will most lily look better on the XS.

PS5 could also have better frame-rates because of the high speed of everything SSD/IO/GPU clock. the lower CU count actually pushes developers towards this.

 

[Mister Wolf]

But that is the impressive part. It will greatly increase visual fidelity for a lower rendering cost than brute forced triangel raytracing. This tech will probably scale well across hardware as well. The visual quality of the demos GI is incredible looking with all the onscreen detail brought on by Nanite. You are having very smal surface detail casting shadows and receiving indirect lighting. This can be done in a similar fashion with RTX and path tracing but those examples uses far less complex scene geometries such as Quake2 and Minecraft. The furute will certainly do perfect raytracing but this approach is very smart and yields great results.

All Lumen is doing is cost cutting. Its no more complex and certainly not better than CryEngine's Voxel GI which uses voxels whether the objects are near or far. Lumen using screen space inherits all the flaws of screen space representation of geometry that was the whole point of realtime triangle raytracing removing on top of being more accurate. Same for Sign Distance Fields which are inferior to using Voxels as well.

 

[thicc_girls_are_teh_best]

It's weird that people discuss the tech along those terms. Like both consoles are SUPER close to each other. Neither is going to really be able to do things the other can't. It's mainly going to come down to who has the better in-house devs and which console has the better exclusives and cool extras stuff not related to console power (i.e. Gamepass, PSVR, etc).

Ah okay, that makes sense then. I think there are enough smaller elements in the systems that differ that does present some interesting advantages one can have over the other in specific use-cases, but overall they're going to be very close in overall system performance. As you say the real pudding will be with the 1st party studios leveraging all of the respective system capabilities in perfect harmony.

I see the systems being more like Super NES and MegaDrive in terms of some of their differences and yet overall having close performance among the top games with devs who can smartly utilize their strengths and work around their weaknesses. You would see bigger differences in the lower-tier of some 3rd-party titles that didn't get as creative with the system architectures, but the top stuff? Yoshi's Island, FF6, Phantasy Star 4, Sonic 3, Alien Soldier, DKC, Vectorman etc? All extremely close to one another in objective and subjective visual terms, because they played to their system strengths and worked around their weaknesses.

PS5 and Series X don't have as much difference between them as those Nintendo and SEGA consoles of course, but it's neat to see there's some of that when you look at the philosophies driving their resolve of things like I/O bottleneck issues, CPU/GPU throughput utilization, etc.

 

[DynamiteCop!]

Even a 12 Teraflop GPU?

You have to think, you're increasing the volume of the scene. This will increase the amount of applied light and shadows on surfaces, the amount of post applied to these areas not to mention the increase draw without culled LoD's. It's a heavy rendering demand. While the amount of what you're streaming may not be a ton, all of those graphical effects have to scale up with it as well.

 

[Jigsaah]

You have to think, you're increasing the volume of the scene. This will increase the amount of applied light and shadows on surfaces, the amount of post applied to these areas not to mention the increase draw without culled LoD's. It's a heavy rendering demand. While the amount of what you're streaming may not be a ton, all of those graphical effects have to scale up with it as well.

So then, you think the systems need more GPU and not to focus so much on SSD? Even when JayzTwoCents is posting video saying the Series X is somewhere between a 2080 super and a Ti?

Somethin don't add up here.