Support NeoGAF

[Mozza]

I gave you guys a very thorough explanation of the role of the SSD. I'm not going to keep repeating myself. Believe what you want to believe.

Just run it by me again.

 

[sircaw]

There is nothing about PS5 SSD helping out graphics I doubt from this tech demo. What I do doubt is him making this outrageous claim that this is simply not possible because SSD's in PC and XsX is just not fast enough. That is dangerous to 'believe' such a thing, because even talented developers have hidden agendas with marketing ploys on gullible and impressionable consumers. The best rebuttal to this tech demo is to DEMONSTRATE it with another unreal 5.0 tech demo from MSFT and other developers on XsX and PC. That will eventually come in time.

If i recall, nx gamer in his latest video stated he is going to release a video on/about ssd's this weekend. Perhaps we both will find something useful in some of his thoughts.

That is my exiting flag of peace, have a good night.

 

[ethomaz]

Oh I see where you are getting confused.
The word software there is referring to the compute shader they wrote-> it a piece of software.
GPUS have dedicated hardware rasterization blocks, but from what I understand passing data through these blocks can happen in parallel with other things.
what he is saying(and I would like someone else to verify) is they don't go through the hardware rasterizer, because for more of their use cases the software compute route is faster and in other cases the hardware rasterizer is.

compute units are on the GPU, so it's the GPU doing the processing just with their code/software.

Oh and guess which has more compute units todo this sort of thing. (just a little twist on the knife there, you like that)

Yes... the vast majority os the triangles are generated by software rasterization (CPU or Async Compute) that is why it can generate billions of triangles without have GPU performance issues.
In specific cases where the hardware rasterization is faster it will use the GPU to render these triangles but they said it is very specific cases.

What defines the performance of Nanite is how faster the data is feed to generate theses triangles.
Faster streaming data = more triangles.

 

[ethomaz]

So what? It's using the GPU to compute math. That resource isn't being done by the SSD dude. So I'm not sure what you are getting at here.

But what defines the performance is how fast you can feed the data to generate these triangles... not the GPU power.
They said their solutions can generate billions of triangles since there is enough data being feed.
This tech is made to free up GPU resources for others tasks.

Faster streaming data = more triangles being generated.

That is what Epic said and explained.

 

[Kazekage1981]

But what defines the performance is how fast you can feed the data to generate these triangles... not the GPU power.
They said their solutions can generate billions of triangles since there is enough data being feed.

That's when the extra 112GB/sec of XsX 10GB of GDDR6 can help right? Assuming that XsX SSD speed is 1/2 of the PS5.

 

[VFXVeteran]

Oh I see where you are getting confused.
The word software there is referring to the compute shader they wrote-> it a piece of software.
GPUS have dedicated hardware rasterization blocks, but from what I understand passing data through these blocks can happen in parallel with other things.
what he is saying(and I would like someone else to verify) is they don't go through the hardware rasterizer, because for more of their use cases the software compute route is faster and in other cases the hardware rasterizer is.

compute units are on the GPU, so it's the GPU doing the processing just with their code/software.

Oh and guess which has more compute units todo this sort of thing. (just a little twist on the knife there, you like that)

Thanks!! You win a prize!

 

[VFXVeteran]

But what defines the performance is how fast you can feed the data to generate these triangles... not the GPU power.
They said their solutions can generate billions of triangles since there is enough data being feed.
This tech is made to free up GPU resources for others tasks.

Faster streaming data = more triangles being generated.

That is just completely untrue. And I'm tired of arguing with you.

 

[ethomaz]

That's when the extra 112GB/sec of XsX 10GB of GDDR6 can help right? Assuming that XsX SSD speed is 1/2 of the PS5.

The data is on storage media not RAM... it has way lower bandwidth.

In a perfect world if you have the all assets on the RAM you don't need to use these tricks to fast use the asset on HDD/SDD to generate additional detail to the render.

That is just completely untrue. And I'm tired of arguing with you.

It is what Tim/Epic is saying.
You have no argument to counter what they said... that is the truth.

I trend to believe more in them than you.
Did you created Nanite?

 

[Thirty7ven]

Either you guys can support with sources how nanite works or you guys are making up theories to fit your favorite narratives. That shit goes for all of you. People are trying to get informed here, I don’t give a fuck if you wanna sweet talk about your pc, your Xbox, your ps, your switch, let’s get facts up in here. A bit tired of know it alls with no sources.

SOURCES

 

[sendit]

The data is on storage media not RAM... it has way lower bandwidth.

In a perfect world if you have the all assets on the RAM you don't need to use these tricks to fast use the asset on HDD/SDD to generate additional detail to the render.


It is what Tim/Epic is saying.
You have no argument to counter what they said... that is the truth.

I trend to believe more in them than you.
Did you created Nanite?

This is true, but not cost effective. There are multiple solutions to a problem, adding more RAM is an incredibly narrow way to look at it.

Just thinking out loud here:

Additionally, RAM is volatile. Are you going to reload assets in to a massive ram pool every time you want to play another games or turn off the device? Think about how long it would take for gigabytes of data to load up all assets in to a pool of RAM. A constant/fast stream is the better option.

 

[Three]

I gave you guys a very thorough explanation of the role of the SSD. I'm not going to keep repeating myself. Believe what you want to believe.

No you didn't you went on about the CPU/RAM and GPU and wrongly downplayed the role of the SSD in loading that data. Affecting the texture quality you can load, player speed, level design. Sure lots and lots of RAM can replace the advantages gained by fast SSD but again who is saying it can't? Nobody.

 

[ethomaz]

This is true, but not cost effective. There are multiple solutions to a problem, adding more RAM is an incredibly narrow way to look at it.

I will say limitations (and that include budget) trend to drive new solutions, tricks, evolution in technology.

 

[JareBear: Remastered]

Splash Damage VP of Tech talks next-gen SSDs and the Megatextures Technique

Next-Gen SSD Tech Could Mean the Resurgence of Megatextures Technique, Says Splash Damage VP of Tech

According to Splash Damage's VP of Technology, the Megatextures technique could make a comeback thanks to the next-gen SSDs in PS5 and XSX.

wccftech.com

There would be a single, incredibly huge (32768×32768 pixel at first, though the size got increased to 128Kx128K with id Tech 5) texture covering the entire polygon map, hence the name. Megatextures technology allowed totally unique textures, without any repeated tiles on the terrain for instance. In short, it promised much more visually varied virtual worlds, particularly for the open world genre which was in its infancy at the time.

According to Splash Damage VP of Technology Marc Fascia, though, Megatextures (or something like it, at least) might have a resurgence soon thanks to the next-gen SSD technology featured in both PlayStation 5 and Xbox Series X, which is set to vastly diminish, if not outright wipe out, those very issues.

Not only are the SSDs in PlayStation 5 and Xbox Series X much faster, both consoles also feature a fully customized Input/Output interface and onboard compression/decompression blocks whose purpose is to virtually eliminate any run-time decompression overhead. The PS5 supports both zlib and the slightly faster Oodle Kraken protocol from RAD Game Tools, while the Xbox Series X supports zlib for general textures and a new, reportedly very fast compression system called BCPack, tailored specifically to handle GPU textures.

 

[sendit]

Splash Damage VP of Tech talks next-gen SSDs and the Megatextures Technique

Next-Gen SSD Tech Could Mean the Resurgence of Megatextures Technique, Says Splash Damage VP of Tech

According to Splash Damage's VP of Technology, the Megatextures technique could make a comeback thanks to the next-gen SSDs in PS5 and XSX.

wccftech.com

Wonder why they couldn't have just have increased the RAM requirement to create the game they want?

/s

"In previous generations, it required a lot of very bespoke optimisation, but things could simplify drastically with faster drives. It’s interesting to see that both hardware vendors such as NVIDIA or AMD as well as Unreal have been doing work in this area lately. "

"How fast the data can be streamed in is vital for such a system to work. "

Just add more RAM bro.

 

[Thirty7ven]

Thanks!! You win a prize!

So the software rasterizer is being run by the CUs, therefore it will scale with number of CUs and clock frequency. But what about the detail? That will be IO dependent considering there’s only so much data you can store in 16 GB of ram (in the case of the PS5, where the demo was running).

In a system with a limited amount of Ram/Vram, the amount of detail you can use in a scene will be dependent on how fast you can move data in and out of ram. Right?

So in theory on XSX because it has around 18% more compute power you could be running it at higher resolution, but considering it has the same amount of Ram/Vram as the PS5, it would display less detail. Makes sense?

On a PC there wouldn’t be this limitation as long as you had the right config, as you can feed it a lot of ram and Vram on top of throwing a lot of compute at it. On PC it becomes about economics and not how performant you can get.

 

[sendit]

Please don't shoot here asking genuine question

Will that not just ware the drive out continuesly decompressing etc

Decompression is handled by the CPU. which is a read request (a read request, has little impact to the wear of the device) from the SSD. In the PS5s case (I beleive the XSX has this as well), decompression cycles are offloaded from the CPU to a hardware de-compressor (effectively reducing CPU load equivalent to the use of 9 Zen 2 cores):

""By the way, in terms of performance, that custom decompressor equates to nine of our Zen 2 cores, that's what it would take to decompress the Kraken stream with a conventional CPU," Cerny reveals. "

Reference:

Inside PlayStation 5: the specs and the tech that deliver Sony's next-gen vision

www.eurogamer.net

 

[pyrocro]

Yes... the vast majority os the triangles are generated by software rasterization (CPU or Async Compute) that is why it can generate billions of triangles without have GPU performance issues.
In specific cases where the hardware rasterization is faster it will use the GPU to render these triangles but they said it is very specific cases.

What defines the performance of Nanite is how faster the data is feed to generate theses triangles.
Faster streaming data = more triangles.

you say yes then you go on to say the exact opposite. what a silly way to argue a point.

THERE IS NO MENTION OF CPU IN THAT SOURCE YOU PROVIDED. YOU PROVIDED the source and you didn't understand what was there

you are getting it wrong, read it again software there does not mean CPU.

don't be so thicc in the head.


Again let's see a SOURCE for as you are saying "what defines the performance of Nanite".

 

[ethomaz]

More to come.



They confirmed to run the two software rasterizers via Async Compute:

 

[pyrocro]

More to come.



They confirmed to run the two software rasterizers via Async Compute:

Yes VIA Async Compute, We all know this. Where is your CPU claim? were you wrong? did you fully understand what you read?

remember compute units are on the GPU.

Again let's see a source for that other thing you're saying "what defines the performance of Nanite".

 

[ethomaz]

Yes VIA Async Compute, We all know this. Where is your CPU claim? were you wrong? did you fully understand what you read?

remember compute units are on the GPU.

Again let's see a source for that other thing you're saying "what defines the performance of Nanite".

Claim? It was an OR.

“CPU or Async Compute”

The only way to do software rasterization is via CPU or Async Compute... Epic confirmed they are using Async Compute something not shared before.

SSD speeds is what defines how much triangles/details Nanite can generate.

 

[pyrocro]

Claim? It was an OR.

“CPU or Async Compute”

The only way to do software rasterization is via CPU or Async Compute... Epic confirmed they are using Async Compute something not shared before.

SSD speeds is what defines how much triangles/details Nanite can generate.

What silly FUDDY DUDDY thing you are saying here.

THERE is NO OR the last real attempt to do real graphics on the CPU was the CELL processor, and we all know how that went.
There is a reason why CPU software renders are used for research purposes now. it is SLOW.
There is a reason why SIMD GPU exist-> they are 100s to 1000s of times faster.

THERE IS NO MENTION Of CPU IN ANY Of YOUR SOURCES. DROP it you were wrong.


Again let's see a source for that other thing you're saying "what defines the performance of Nanite".

 

[Rentahamster]

Linus weighs in on the SSD conversation.

 

[Ascend]

Wonder why they couldn't have just have increased the RAM requirement to create the game they want?

/s

"In previous generations, it required a lot of very bespoke optimisation, but things could simplify drastically with faster drives. It’s interesting to see that both hardware vendors such as NVIDIA or AMD as well as Unreal have been doing work in this area lately. "

"How fast the data can be streamed in is vital for such a system to work. "

Just add more RAM bro.

Additional RAM can compensate for a slower SSD speed. But you can't use RAM to compensate for the slow loading of the HDD. Just as a reference...
It was stated in a Eurogamer article that one statue in the UE5 demo uses 24 8K textures. That would amount to 24 x 7680 x 4320 x 32 / (8,000,000,000) = 3.2GB of assets for one statue. If all statues are the same, no duplicate data is necessary. In any case, the full uncompressed statue can be fully streamed in less than a second (0.6 seconds) by the PS5, while the XSX would take 1.3 seconds. An HDD at 50 MB/s would take 64 seconds. That means that your initial loading on the SSDs will be fast, while the HDD will take a full minute to do it. No amount of RAM will fix this.

However... Assume you have to render four different statues that happen to be the exact same size, in one second. The GPU would need to receive 4 x 3.2 GB = 12.8 GB in that second. Assume for a second also that the GPU can output the statues instantly after all the data is received. No compression is used.
In order to achieve that on the PS5, you would need to have 12.8 - 5.5 = 7.3GB of data already in RAM, to reach the target of 2 statues in one second by streaming the remaining data.
For the XSX, the same would be 12.8 - 2.4 = 10.4 GB of RAM required.
For a system with an HDD, it would be 12.8 - 0.05 = 12.75 GB of RAM required, or basically, everything.

Ok. So now, if you want to swap those statues for another 4 completely different statues...;
The PS5 would take 12.8 / 5.5 = 2.3 seconds
The XSX would take 12.8 / 2.4 = 5.3 seconds
The system with an HDD would take 12.8 / 0.05 = 256 seconds.

So in this case, the RAM of the PS5 needs to hold data for the upcoming 2.3 seconds of gameplay, the XSX for the upcoming 5.3 seconds of gameplay, and the HDD system for up to 256 seconds of gameplay. I think we can understand why the additional RAM argument is sound for slower and faster SSDs, but not for HDDs.

The bottom line is that the PS5 needs less RAM than the XSX. If a game is optimized for the PS5 to use both its SSD speed, and the max amount of RAM, the XSX will not be able to cope with it. A PC with 32GB of RAM for example, will have zero issues compensating for the slower SSD though with its RAM.

 

[MasterCornholio]

The judge. How did this even get to my courtroom.

He's probably just enjoying it.

 

[Exodia]

When it takes 30 seconds to load a game even on fast SDD it's clearly a sign of bottleneck, because with lets say 3.5 GB/s SDD you should be able to fill system ram in just a few seconds.

Tim Sweeney from Epic has more knowledge than everyone here, and even he wrote the reason why PC SDDs are slower is because of bottlenecks. They only reason why you think PC SDDs have no bottlenecks is because you dont want to accept the truth and you are willing willing to lie to yourelf, just to not admit PC technology can be sometimes worse compared to cheap consoles. Other PC "fans" are doing the same thing, so you can hear them saying the dumbest things possible like there's no 16x MSAA on consoles (the problem is MSAA is no longer supported in the majority of games even on PC).

No the bottle neck is easily resolved by Direct Storage API which releases on PC later this year.

Microsoft’s Goossen told Digitial Foundry that doing decompression on the 4K textures to match the speed of the SSD rate would have consumed three Zen 2 CPU cores, plus an additional two more just for the I/O overhead. With DirectStorage, Microsoft reduced that down to just a tenth of one core. All that CPU power can now be repurposed for other things.

DirectStorage works in conjunction with the Sampler Feedback Streaming (SFS) technology also built into the Velocity Engine, reducing the number of textures that actually need to be loaded.

This increased efficiency translates into two to three times improvement on the effective amount of physical memory, and two to three times more I/O bandwidth, Goossen said.

You still need hardware in order to remove these bottlenecks he is talking about.

Using software you can make things as close to a wash.

A storage api such as the upcoming DirectStorage can only mitigate the bottleneck to a certain degree, you need custom hardware blocks to get rid of the overhead in a way closer to the PS5 architecture. This might be available on PC next year, but it'll cost money. So as of right now, there is no equivalent of what PS5 has on the PC side.

How is it only mitigating the bottleneck when there's literally no bottleneck? Going from need 5 zen 2 cores to using only 10% of 1 zen 2 cores seems like more than mitigating to me. More like eliminating. Now for PC, while they don't have HW decompressor, Direct Storage still gives them a huge boost because it allows developers to access the low levels of the nvme controller. It eliminates mostly the issues that Tim mentioned.

 

[Kazekage1981]

I know this is not the case, and no one implying it is, it just makes it seem from all these developers especially from EPIC and Sony themselves, that the SSD is some sort of extension of the GPU when its not. Its basically being used as additional RAM with a lot lower bandwidth then the main RAM to free up the main RAM for more pertinent and prioritized graphical and game assets (and that's a good thing).

I am still kind of perplexed at this new established 'bottleneck':

"[The level of detail such as textures, light, appearance of shapes, colors, patterns + the generation of billions of triangles = assets] is limited by stream of data of those assets coming from where it is stored (SSD)"

thicc_girls_are_teh_best can you please chip in? You are good at explaining things

 

[thicc_girls_are_teh_best]

I know this is not the case, and no one implying it is, it just makes it seem from all these developers especially from EPIC and Sony themselves, that the SSD is some sort of extension of the GPU when its not. Its basically being used as additional RAM with a lot lower bandwidth then the main RAM to free up the main RAM for more pertinent and prioritized graphical and game assets (and that's a good thing).

I am still kind of perplexed at this new established 'bottleneck':

"[The level of detail such as textures, light, appearance of shapes, colors, patterns + the generation of billions of triangles = assets] is limited by stream of data of those assets coming from where it is stored (SSD)"

thicc_girls_are_teh_best can you please chip in? You are good at explaining things

The truth is it may be a mixture of both. Yes, the SSDs are being used to stream data into system memory faster, but I also think the systems are using some form of AMD's SSG tech, which does provide a bypass of on-board NAND storage directly to the GPU. How specifically that works I don't know, I have to look into it a bit more (PC World has a pretty good article about it).

The SSD is basically fast enough with raw data sequential speeds that it can replace contents of RAM very quickly, but I think the part people are getting hung up on is simply looking at the paper specs of PS5's SSD and XSX's SSD and making conclusions from that. At least IMO, MS's taking a more software-focused approach to resolving certain data pipeline bottleneck issues whereas Sony's taken a more hardware-focused approach. MS's will probably need more targeted optimization from devs to utilize at high levels but if the software API and toolset side are sufficient enough the learning curve should be lowered.

Comparatively, Sony's approach will be the easier for devs to utilize out of the gate since the hardware is basically automating a lot of the process for them and having that done by hardware essentially provides at least some degree of "free" performance. But that also means devs don't need to think as much about optimizations. Time will tell if that ends up encouraging bad/lazy coding practices, but it does mean they will have to be a bit smarter with XSX to achieve similar asset streaming results.

Thing is, XSX has the tools to do that. It has the APIs to leverage them, too.

Additional RAM can compensate for a slower SSD speed. But you can't use RAM to compensate for the slow loading of the HDD. Just as a reference...
It was stated in a Eurogamer article that one statue in the UE5 demo uses 24 8K textures. That would amount to 24 x 7680 x 4320 x 32 / (8,000,000,000) = 3.2GB of assets for one statue. If all statues are the same, no duplicate data is necessary. In any case, the full uncompressed statue can be fully streamed in less than a second (0.6 seconds) by the PS5, while the XSX would take 1.3 seconds. An HDD at 50 MB/s would take 64 seconds. That means that your initial loading on the SSDs will be fast, while the HDD will take a full minute to do it. No amount of RAM will fix this.

However... Assume you have to render four different statues that happen to be the exact same size, in one second. The GPU would need to receive 4 x 3.2 GB = 12.8 GB in that second. Assume for a second also that the GPU can output the statues instantly after all the data is received. No compression is used.
In order to achieve that on the PS5, you would need to have 12.8 - 5.5 = 7.3GB of data already in RAM, to reach the target of 2 statues in one second by streaming the remaining data.
For the XSX, the same would be 12.8 - 2.4 = 10.4 GB of RAM required.
For a system with an HDD, it would be 12.8 - 0.05 = 12.75 GB of RAM required, or basically, everything.

Ok. So now, if you want to swap those statues for another 4 completely different statues...;
The PS5 would take 12.8 / 5.5 = 2.3 seconds
The XSX would take 12.8 / 2.4 = 5.3 seconds
The system with an HDD would take 12.8 / 0.05 = 256 seconds.

So in this case, the RAM of the PS5 needs to hold data for the upcoming 2.3 seconds of gameplay, the XSX for the upcoming 5.3 seconds of gameplay, and the HDD system for up to 256 seconds of gameplay. I think we can understand why the additional RAM argument is sound for slower and faster SSDs, but not for HDDs.

The bottom line is that the PS5 needs less RAM than the XSX. If a game is optimized for the PS5 to use both its SSD speed, and the max amount of RAM, the XSX will not be able to cope with it. A PC with 32GB of RAM for example, will have zero issues compensating for the slower SSD though with its RAM.

Does any of this factor into account what role (if any) the wider data bus and faster memory bandwidth on XSX plays in terms of how the data in RAM is actually being accessed by the GPU? Also is there any scenario where utilizing lower-quality textures to stream into memory and/or to the GPU and then upscaled in real-time assist XSX in the asset streaming department?

 

[pawel86ck]

No the bottle neck is easily resolved by Direct Storage API which releases on PC later this year.

Microsoft’s Goossen told Digitial Foundry that doing decompression on the 4K textures to match the speed of the SSD rate would have consumed three Zen 2 CPU cores, plus an additional two more just for the I/O overhead. With DirectStorage, Microsoft reduced that down to just a tenth of one core. All that CPU power can now be repurposed for other things.

DirectStorage works in conjunction with the Sampler Feedback Streaming (SFS) technology also built into the Velocity Engine, reducing the number of textures that actually need to be loaded.

This increased efficiency translates into two to three times improvement on the effective amount of physical memory, and two to three times more I/O bandwidth, Goossen said.



Using software you can make things as close to a wash.



How is it only mitigating the bottleneck when there's literally no bottleneck? Going from need 5 zen 2 cores to using only 10% of 1 zen 2 cores seems like more than mitigating to me. More like eliminating. Now for PC, while they don't have HW decompressor, Direct Storage still gives them a huge boost because it allows developers to access the low levels of the nvme controller. It eliminates mostly the issues that Tim mentioned.

Interesting, but let me ask you one question, if MS can remove these bottlenecks just with software optimizations, then why sony build hardware for that? Cerny dont know what he is doing?

 

[Persian buttercup]

I love that the presence of this thread has annoyed a lot of people.

 

[phil_t98]

There is nothing about PS5 SSD helping out graphics I doubt from this tech demo. What I do doubt is him making this outrageous claim that this is simply not possible because SSD's in PC and XsX is just not fast enough. That is dangerous to 'believe' such a thing, because even talented developers have hidden agendas with marketing ploys on gullible and impressionable consumers. The best rebuttal to this tech demo is to DEMONSTRATE it with another unreal 5.0 tech demo from MSFT and other developers on XsX and PC. That will eventually come in time.

He never claims Xbox sx or pc isn’t fast enough all he says is pc with HDD isn’t fast enough. That is all he said on that

 

[Omega Supreme Holopsicon]

He never claims Xbox sx or pc isn’t fast enough all he says is pc with HDD isn’t fast enough. That is all he said on that

Some are claiming even sata ssd is insufficient, you need at least nvme ssd.

But the team definitely underlines that this demo cannot be played using a SATA SSD. -play4

Epic also appreciated the benefits of the PS5's SSD, noting that such graphics cannot be achieved using a hard drive or even a SATA SSD as they do not have enough speed to load the high resolution textures as they are needed.If we want to use those textures on PC, we will need if or if a high speed NVME SSD. -guru3d

 

[phil_t98]

Some are claiming even sata ssd is insufficient, you need at least nvme ssd.

where is that from? can you link please

 

[Omega Supreme Holopsicon]

where is that from? can you link please

"Epic also appreciated the benefits of the PS5's SSD, noting that such graphics cannot be achieved using a hard drive or even a SATA SSD as they do not have enough speed to load the high resolution textures as they are needed.If we want to use those textures on PC, we will need if or if a high speed NVME SSD. "

For PC at least a GeForce RTX 2070 Super to run the Unreal Engine 5 demo smoothly

And that's not bad at all when you think about it. Epic discussed the requirements to run this demo properly on PC, revealing that at least one RTX 2070 Super is required....

www.guru3d.com

RTX 2070 or SATA SSD which graphics card does the Unreal Engine 5 demo require?

It is not possible to play Unreal Engine 5 demo with SATA SSD. The demo can be played on the PC with RTX 2070 Super graphics card.

play4.uk

 

[synce]

You probably can't achieve these graphics with standard nvme now... just extremely similar ones once you tone down a couple settings that 99% of people wouldn't notice anyway. I also found it funny that the tech demo had a classic shimmy animation to disguise load times, despite all the hype about PS5's SSD eliminating them. If devs have to make a tradeoff between no loading or stupidly hi-res textures, I'll bet they'll always choose the latter.

 

[Omega Supreme Holopsicon]

Linus weighs in on the SSD conversation.

Poor linus. First a lot of this demo is geometry which I've heard is ridiculously compressible, which if true means even the 22GB/s ps5 ssd figure that trounces the raid ssds he speaks of would be a reality. Second those are raid ssds, not single ssds, if I'm not mistaken. Third the i/o complex of the ps5 means it is essentially faster also do to no overheads that those pc ssds will have.

You probably can't achieve these graphics with standard nvme now... just extremely similar ones once you tone down a couple settings that 99% of people wouldn't notice anyway. I also found it funny that the tech demo had a classic shimmy animation to disguise load times, despite all the hype about PS5's SSD eliminating them. If devs have to make a tradeoff between no loading or stupidly hi-res textures, I'll bet they'll always choose the latter.

They clarified that was not to hide loading but to show close detail an artistic and unecessary choice.

 

[MMaRsu]

Yeah not like Epic would lie or anything

 

[Rentahamster]

Poor linus. First a lot of this demo is geometry which I've heard is ridiculously compressible, which if true means even the 22GB/s ps5 ssd figure that trounces the raid ssds he speaks of would be a reality. Second those are raid ssds, not single ssds, if I'm not mistaken. Third the i/o complex of the ps5 means it is essentially faster also do to no overheads that those pc ssds will have.

They clarified that was not to hide loading but to show close detail an artistic and unecessary choice.

Poor Linus?

 

[Omega Supreme Holopsicon]

Poor Linus?

He's trying hard to downplay Tim's comment, and ps5's ssd superiority. But even with his uber expensive, raid ssds if I'm not mistaken. He still seems to fail to show absolute superiority of the pc, once you take the full specs of the ps5 into account.

The I/O complex and the 22GB/s compressed figure for some content, still surpasses even these raid ssds he tried to bring to the table.

 

[Rentahamster]

The I/O complex and the 22GB/s compressed figure for some content, still surpasses even these raid ssds he tried to bring to the table.

You're comparing the marketing speak of an untested and unreleased platform vs a tangible, tested, piece of hardware in his hands.

 

[Omega Supreme Holopsicon]

You're comparing the marketing speak of an untested and unreleased platform vs a tangible, tested, piece of hardware in his hands.

You can claim it is marketing speak, but it was claimed to be achievable to developers in a developer conference not a marketing promotion. It would be senseless to tell developers this is what can be achieved if it is false.

edit:
As for those tested platforms, I've heard, but not confirmed, that raid ssds fall short of their theoretical maximum multiple speed ups.

 

[phil_t98]

Some are claiming even sata ssd is insufficient, you need at least nvme ssd.

I totally get that it won’t run on Sata SSD but there are no quotes I have seen on that, the only thing that’s printed as a quote is Tim saying it would be downgraded on HDD. He did say it would run awesome on pc to.

 

[Omega Supreme Holopsicon]

I totally get that it won’t run on Sata SSD but there are no quotes I have seen on that, the only thing that’s printed as a quote is Tim saying it would be downgraded on HDD. He did say it would run awesome on pc to.

we would have to ask the authors of the quote, they claim epic said sata ssd is also unviable

And they are not the only article claiming that

Finally, Epic noted that these graphics were made even more possible by the use of the PS5’s SSD, even going as far to say that a normal hard drive or even a SATA SSD wouldn’t be fast enough to load the high res textures used in the demo, and an NVME SSD would be needed in order to load the textures at the required speed on a PC. -game-debate

RTX 2070 Super will be required to at least run Unreal Engine 5 demo on PC, says Epic

Epic has stated that an RTX 2070 Super will be needed to at least run the Unreal Engine 5 demo on PC.

www.game-debate.com

 

[phil_t98]

we would have to ask the authors of the quote, they claim epic said sata ssd is also unviable

And they are not the only article claiming that

RTX 2070 Super will be required to at least run Unreal Engine 5 demo on PC, says Epic

Epic has stated that an RTX 2070 Super will be needed to at least run the Unreal Engine 5 demo on PC.

www.game-debate.com

I know they ain’t but as I said the only direct quote is the one Tim said reguarding HDD. I agree Sata SSD probibly won’t run it as well, but around here people were saying the other day it would be impossible to run the demo on pc at the same level as PS5.

now if you use what Tim is quoted in reguards toPChe said that PC top end at the moment can run it “awesome” which people have read to mean not as good as PS5 , now Tim has been quoted about the tech in unreal engine 5 saying both the PS5 and XBSX would both run it awesome as well. I take from That awesome is the same

 

[Omega Supreme Holopsicon]

I know they ain’t but as I said the only direct quote is the one Tim said reguarding HDD. I agree Sata SSD probibly won’t run it as well, but around here people were saying the other day it would be impossible to run the demo on pc at the same level as PS5.

now if you use what Tim is quoted in reguards toPChe said that PC top end at the moment can run it “awesome” which people have read to mean not as good as PS5 , now Tim has been quoted about the tech in unreal engine 5 saying both the PS5 and XBSX would both run it awesome as well. I take from That awesome is the same

probably we don't know for sure. I doubt they'll give figures that would clarify the issue.

But thankfully we have sony first party titles whose developers will likely give figures. That will tell us how much of the ssd speed is actually needed in games.

 

[Rentahamster]

You can claim it is marketing speak

Nearly everything that isn't test numbers from a neutral third party is marketing speak.

but it was claimed to be achievable

I've heard

These types of words do not inspire certainty or confidence.

It would be senseless to tell developers this is what can be achieved if it is false.

Microsoft and Sony are marketing to developers just as much as they're marketing to consumers. Marketing talking points aren't necessarily false, either. They're trying to portray their product in the best light possible, and you need to be able to distinguish between verified and hyperbole.

 

[phil_t98]

probably we don't know for sure. I doubt they'll give figures that would clarify the issue.

But thankfully we have sony first party titles whose developers will likely give figures. That will tell us how much of the ssd speed is actually needed in games.

Would u trust a first party developer on numbers though?
I mean first party developers will always give numbers in favour of the system they are tied to and that works both sides to.

at the moment all we have is Tim saying the breakthrough with PS5 has made this possible.

doesn’t mean as many have thought it isn’t possible on other hardware at exactly the same levels. The breakthrough he “could” be referring to could be between the level of hardware from PS4 to PS5.

it’s just interesting how it can be read so many different ways don’t you think?

 

[FranXico]

Claim? It was an OR.

“CPU or Async Compute”

The only way to do software rasterization is via CPU or Async Compute... Epic confirmed they are using Async Compute something not shared before.

SSD speeds is what defines how much triangles/details Nanite can generate.

You do know that code can be offloaded to the CUs, right?

 

[Omega Supreme Holopsicon]

Would u trust a first party developer on numbers though?
I mean first party developers will always give numbers in favour of the system they are tied to and that works both sides to.

I sincerely doubt they'd make up figures, as this can be tested by other developers with dev kits, and they'd easily be found out and look foolish if they tried to pull anything like that off.

 

[phil_t98]

I sincerely doubt they'd make up figures, as this can be tested by other developers with dev kits, and they'd easily be found out and look foolish if they tried to pull anything like that off.

There will be quotes such as this couldn’t be made without the power of the SSD running at this speed or this couldn’t be made with out 12tf.

it’s easy to quote something like that.

we have already seen what Tim has said been pushed in a million different directions from only a couple of lines he said.

first party Devs will always quote in favour of the hardware they are tied to

 

[Self_Destructive]

PS5 is undoubtedly more efficient in its design for gaming than any PC...

However being more efficient, doesn't necessarily mean more performant.