Support NeoGAF

[iHaunter]

Microsoft will try as hard as they can to downplay the PS5 SSD!! The difference is much bigger than the TFlops difference.

Correct, the PS5 SSD is still far superior. Click-bait title again? Either way, I'm curious how much devs take adv on both consoles.

 

[SenjutsuSage]

Bogus? LOL

Where do you think the narrative about 10 million copies sold from a 100 million install base originated from?

Before the Xbox One X came out, people were saying people cared about the best version of multiplats and exclusives were only for the hardcore games. That's the reason why you see Xbox posters on here repeat the same stuff because they got their terrible logic from them. lol

I know you like the, but they are a very toxic community.

Incorrect. They've criticized xbox for years for lack of good new first parties on the level of what sony put out. You aren't paying attention. God of War and Spider-Man are among their most heavily praised games period. Nobody claims exclusives aren't important. They do not do it, but they do also admit that multi-platforms are what most people play and buy, which is true.

 

[Vasto]

Series x is just on another level compared to PS5

 

[DForce]

Incorrect. They've criticized xbox for years for lack of good new first parties on the level of what sony put out. You aren't paying attention. God of War and Spider-Man are among their most heavily praised games period. Nobody claims exclusives aren't important. They do not do it, but they do also admit that multi-platforms are what most people play and buy, which is true.

It's right there in the video. I know you like them, but don't be in straight denial. They used bad metrics to show their viewers that no one was buying exclusives.

Yes, they like exclusives, but they still said it didn't factor in when it comes to selling consoles.

 

[dxdt]

Regardless of whether the PS5 is weaker or not (most likely weaker), TF as a general measurement is extremely wanting, and he’s not wrong.

Take this Digital Foundry piece, wherein they make the case for a 4TF Xbox Series S, and repeatedly get lower TF hardware to turn out better results than higher TF builds:



Likewise, I thought Cerny made a good case for his CU argument as well.

Teraflops are a good general measurement, but we’ve come to rely on them to an obsessive degree. A number has been simplified down to the number for ease-of-use among console warriors, when the reality is quite a bit more complex.

The video is comparing 4 TF RDNA vs. 6 TF GCN. He's saying that MS having a 4 TF RDNA2 cheaper box vs 12 TF RDNA2 might be valid if the resolution is scaled to 1080p instead of 4K.

 

[martino]

The video is comparing 4 TF RDNA vs. 6 TF GCN. He's saying that MS having a 4 TF RDNA2 cheaper box vs 12 TF RDNA2 might be valid if the resolution is scaled to 1080p instead of 4K.

but what will happen when the resolution is not 4k on XsX ? because i think it will happen.

 

[StreetsofBeige]

Bogus? LOL

Where do you think the narrative about 10 million copies sold from a 100 million install base originated from?

Before the Xbox One X came out, people were saying people cared about the best version of multiplats and exclusives were only for the hardcore games. That's the reason why you see Xbox posters on here repeat the same stuff because they got their terrible logic from them. lol

I know you like the, but they are a very toxic community.

And how do sales wars relate to the thread?

 

[DForce]

And how do sales wars relate to the thread?

Someone posted a video from Colteastwood. I'm pointing out how often he tires to spread misinformation.

 

[rntongo]

But I thought PS5 was more balanced according to a thread I saw.

Curious why PS5's slightly-lower tflops is an "extreme" but XsX SSD performing at half the speed is "balanced".

Because at the end of the day you cannot deliver next gen graphical fidelity with an SSD

 

[mckmas8808]

The video is comparing 4 TF RDNA vs. 6 TF GCN. He's saying that MS having a 4 TF RDNA2 cheaper box vs 12 TF RDNA2 might be valid if the resolution is scaled to 1080p instead of 4K.

I thought you guys said less CUs would hurt Ray Tracing and other compute stuff.

 

[ZywyPL]

I thought you guys said less CUs would hurt Ray Tracing and other compute stuff.

RT performance scales with resolution (rays per pixels), so going from 4K all the way down to 1080p should require require just a quarter of RT capabilities of XBX.

 

[mckmas8808]

RT performance scales with resolution (rays per pixels), so going from 4K all the way down to 1080p should require require just a quarter of RT capabilities of XBX.

Hmmmm....so I'd assume this same logic applies if say the GPU needed to scale the resolution down say 18% then right? Let's say if a machine wanted to display the same RT capabilities as the XSX, but was 18% weaker.....it could run the game at a resolution around 3600 x 2000 (instead of 3840 x 2160) and get the same RT performance (rays per pixel).

That's good to know. Thanks!

 

[rntongo]

But I thought PS5 was more balanced according to a thread I saw.

Curious why PS5's slightly-lower tflops is an "extreme" but XsX SSD performing at half the speed is "balanced".

SSDs mainly improve load times and can enable an increase in the amount of RAM used to render current scenes. Throughput above 2GB/s would mean data can instantly be loaded into RAM from secondary memory. At such speeds the user will only notice minor differences in load times but will not see any differences when playing a game.

 

[rntongo]

So probably 100 gb for os 100+ gb for this virtual ram, and games 100+ Gb, I think the 1 tb drives are going to make it difficult to have more than 4 or 5 large games installed at any time, anyone see any mention of the disk drive speeds as it seems like managing space this gen will be way worse than last

The OS won't take up 100GB. The virtual RAM is the installed game itself. Data from the game is instantly loaded into RAM, so the game installs on the SSD are a virtual RAM.

 

[Panajev2001a]

SSDs mainly improve load times and can enable an increase in the amount of RAM used to render current scenes. Throughput above 2GB/s would mean data can instantly be loaded into RAM from secondary memory. At such speeds the user will only notice minor differences in load times but will not see any differences when playing a game.

Not sure about the bolded there. It implies 2 GB/s is the maximum usable speed to produce perceivable differences (more interested in data you can transfer in 0.25s or even how
much data you can get transferred in RAM in 33-50ms so data you could use in the next frame, 30 FPS games, for per frame streaming). Not convinced it is the case.

 

[Trueblakjedi]

So "any chosen 5GB of 100GB is accessible in about a second" right?
That's cool, now, why only 100GB? I thought SSD in XSeX will be 10 times bigger.

The way I read it, some assholes tried to mislead the public by making nonsensical statements.

The amount could technically be as high as 2TB. The Radeon Pro SSG was built in by AMD in 2016 with a refresh in 2017 with what appears to be the exact same technology. However devs will tell you that access to 50-60Gb of compressed data "instantly" is more than enough.. you can read about the implementation here and here. Its pretty impressive if this is what the architecture can truly implement.

 

[Ashoca]

Not sure about the bolded there. It implies 2 GB/s is the maximum usable speed to produce perceivable differences (more interested in data you can transfer in 0.25s or even how
much data you can get transferred in RAM in 33-50ms so data you could use in the next frame, 30 FPS games, for per frame streaming). Not convinced it is the case.

why would a multiplatform game developer optimize their games for a single SSD? There are a lot of players on PC that don't even use SSD. Also, Sony didn't even show us anything besides less loading times. No video, no demo, nothing.

So, sony should show us something first maybe?!

It's safe to assume that for multiplatform games the only difference will be reduced loading times, that's it.

 

[rntongo]

Not sure about the bolded there. It implies 2 GB/s is the maximum usable speed to produce perceivable differences (more interested in data you can transfer in 0.25s or even how
much data you can get transferred in RAM in 33-50ms so data you could use in the next frame, 30 FPS games, for per frame streaming). Not convinced it is the case.

It takes half a second for a player to turn in a video game. Within half a second, the PS5 can load 2GB of data into RAM with a throughput of 5.5GB/s. The series X can load close to 1GB in the same time frame if we take a throughput of 2.4 GB/s. We're talking about 50-100 times faster speeds here so it changes the way RAM is used. instead of loading into RAM what will be needed, only load what is going to be used within that second. So you're feeding 1 second of gameplay with 8GB RAM on PS5 and 4GB of RAM on Series X. Thats crazy!!

 

[Trueblakjedi]

It takes half a second for a player to turn in a video game. Within half a second, the PS5 can load 2GB of data into RAM with a throughput of 5.5GB/s. The series X can load close to 1GB in the same time frame if we take a throughput of 2.4 GB/s. We're talking about 50-100 times faster speeds here so it changes the way RAM is used. instead of loading into RAM what will be needed, only load what is going to be used within that second. So you're feeding 1 second of gameplay with 8GB RAM on PS5 and 4GB of RAM on Series X. Thats crazy!!

It would be more than 4GB of Ram because you would only be loading whats visible on screen... instead of entire mipmap texture you would only load the element to the player... so you could really load a lot of other data rather then the entire texture map. I dont know if the PS5 has this feature but its a key feature of the XSX... IN others words what the PS5 would have to use 8 Gb to move through the PCIE bus you would only need 3 or so Gb on the XSX if it wasnt already found in the SSD Vram.

 

[rntongo]

It would be more than 4GB of Ram because you would only be loading whats visible on screen... instead of entire mipmap texture you would only load the element to the player... so you could really load a lot of other data rather then the entire texture map. I dont know if the PS5 has this feature but its a key feature of the XSX... IN others words what the PS5 would have to use 8 Gb to move through the PCIE bus you would only need 3 or so Gb on the XSX if it wasnt already found in the SSD Vram.

True. I was just pointing out how much data would be possible to load in a second. I think you're talking about sampler feedback streaming. And yes that would make it much more efficient by only loading data that is being used in that instance.

 

[Panajev2001a]

It takes half a second for a player to turn in a video game. Within half a second, the PS5 can load 2GB of data into RAM with a throughput of 5.5GB/s. The series X can load close to 1GB in the same time frame if we take a throughput of 2.4 GB/s. We're talking about 50-100 times faster speeds here so it changes the way RAM is used. instead of loading into RAM what will be needed, only load what is going to be used within that second. So you're feeding 1 second of gameplay with 8GB RAM on PS5 and 4GB of RAM on Series X. Thats crazy!!

Yes, it is... crazy good .

 

[Bernkastel]

 

[Goliathy]

maybe that's why they had to partner with seagate for the SSD? Can't get this sweet stuff with the off-the-shelf ssds

 

[Deto]

This is because SX will run Azure applications while not running xcloud games.

maybe that's why they had to partner with seagate for the SSD? Can't get this sweet stuff with the off-the-shelf ssds

Like the xbox 360 HD?
They did it because it is more profitable to sell proprietary memory

 

[Bernkastel]

maybe that's why they had to partner with seagate for the SSD? Can't get this sweet stuff with the off-the-shelf ssds

Seagate only made the enclosure and is the exclusive launch partner. There will be more later.

 

[Goliathy]

Like the xbox 360 HD?
They did it because it is more profitable to sell proprietary memory

what? no. the xbox 360 hdd wasn't anything special.

 

[D.Final]

what? no. the xbox 360 hdd wasn't anything special.

Welp

 

[Bernkastel]

 

[The Hamburglar]

So backpak is their secret, secret sauce?

 

[Jigsaah]

I'm confused as hell. Is this something to counter the SSD in the PS5? Too many numbers, somebody break it down for me.

 

[Jigsaah]

absolutely great explanation.. are you a dev from sony?

what happens when 399 people come to you station and you fast car has only 200 seats?
i think... big problems happen... .

I'd rather take the bus...That's too many damn people in a car.

 

[rntongo]

I'm confused as hell. Is this something to counter the SSD in the PS5? Too many numbers, somebody break it down for me.

Depends on how good the texture streaming on the PS5 is. But basically it means with SFS, at the least it will close the gap between the SSDs because the XSX will be able to cut down the amount of textures being used in a scene by up to 75% for example. It does this by only sending certain pieces of a texture into RAM.

 

[Panajev2001a]

Depends on how good the texture streaming on the PS5 is. But basically it means with SFS, at the least it will close the gap between the SSDs because the XSX will be able to cut down the amount of textures being used in a scene by up to 75% for example. It does this by only sending certain pieces of a texture into RAM.

It does not close the gap, it would only close the gap if PS5 did not have barely decent texture streaming which SFS makes easier / more efficient but is, as you yourself posted, a way to give feedback to the same mechanism powering PRT/existing virtual texturing implementations. Also you are quoting best case scenario vs an implementation that likely is doing little to no streaming (and use of PRT). At this point I may as well take Kraken’s 22 GB/s theoretical peak and claim that your 75% memory savings or whatever multiplier you throw at it is still not enough .

 

[Jigsaah]

Depends on how good the texture streaming on the PS5 is. But basically it means with SFS, at the least it will close the gap between the SSDs because the XSX will be able to cut down the amount of textures being used in a scene by up to 75% for example. It does this by only sending certain pieces of a texture into RAM.

And so...this wouldn't like...lessen the quality of the textures?

SFS?

EDIT:
Ok Sampler Feedback Streaming. Had to look that up. Ok so basically it's lightening the load and making the system more efficient...therefore not needing a faster SSD?

 

[Panajev2001a]

I'm confused as hell. Is this something to counter the SSD in the PS5? Too many numbers, somebody break it down for me.

Marketing trying to spin an improvement in efficiency as a revolution that would close the gap with PS5’s SSD betting on the fact that people forgot Partially Resident Textures were a thing (and give you this memory and bandwidth multiplier in that fashion).

From the horse’s mouth:

That being said, games have been streaming virtual memory pages for a while. It's called Partially Resident Textures. Using Sampler Feedback to trigger page reads is sort of the "new hotness".

Samplers Feedback, aka PRT+, defined in the specs:https://microsoft.github.io/DirectX-Specs/d3d/SamplerFeedback.html

 

[Bernkastel]

I'm confused as hell. Is this something to counter the SSD in the PS5? Too many numbers, somebody break it down for me.

These things are created after years of R&D. Some of the stuffs in XSX SSD and the velocity architecture are based on Azure data center storage designs. Even the closer look and glossary blog post, where they talked about their SSD and velocity architecture was released coinciding the Digital Foundry and Austin Evans unveil, a day before the "Road to PS5" stream.

 

[rntongo]

It does not close the gap, it would only close the gap if PS5 did not have barely decent texture streaming which SFS makes easier / more efficient but is, as you yourself posted, a way to give feedback to the same mechanism powering PRT/existing virtual texturing implementations. Also you are quoting best case scenario vs an implementation that likely is doing little to no streaming (and use of PRT). At this point I may as well take Kraken’s 22 GB/s theoretical peak and claim that your 75% memory savings or whatever multiplier you throw at it is still not enough .

PRTs don’t hold a candle to SF hardware in SFS. The PS5 is most likely going to use sampler feedback but will still push textures into RAM that may never be used. SFS is supposed to eliminate this and that’s why it has a huge multiplier of 2-3x. Even if you use an absurd 22GB/s figure For the PS5 it 100% doesn’t apply to texture data which is what we’re talking about here. Sony didn’t spend years building a system like SFS as far as we know and they didn’t have to

 

[Panajev2001a]

These things are created after years of R&D. Some of the stuffs in XSX SSD and the velocity architecture are based on Azure data center storage designs. Even the closer look and glossary blog post, where they talked about their SSD and velocity architecture was released coinciding the Digital Foundry and Austin Evans unveil, a day before the "Road to PS5" stream.

Sure, nobody is stupid enough to thinkDirect Storage, the BCPack HW decompressor, and other optimisations to data streaming were copied from PS5 and hastily done. Any console engineer trying to make use of decently fast SSD’s in a console scenario and without the backwards compatibility concerns of Desktop Windows would try to use all the available SSD bandwidth and see that it would saturate your CPU cores just to move data around the system.

You should be proud of what they have achieved, but IMHO you oversell it as if you really were selling it.

 

[rntongo]

And so...this wouldn't like...lessen the quality of the textures?

SFS?

EDIT:
Ok Sampler Feedback Streaming. Had to look that up. Ok so basically it's lightening the load and making the system more efficient...therefore not needing a faster SSD?

SFS is a suite of hardware features built custom for the XSX GPU that include Sampler Feedback. Sampler Feedback is an improvement over PRTs but still even on the XSX, it will have custom texture filters not available in the PC hardware. Basically MSFT took years of PRT data in the Xbox One and found there were large percentages of texture data in ram that were actually never used. SFS eliminates this waste by streaming only pieces of textures as they are needed. This is a huge advantage over all other next gen systems.

 

[Panajev2001a]

PRTs don’t hold a candle to SF hardware in SFS. The PS5 is most likely going to use sampler feedback but will still push textures into RAM that may never be used. SFS is supposed to eliminate this and that’s why it has a huge multiplier of 2-3x. Even if you use an absurd 22GB/s figure For the PS5 it 100% doesn’t apply to texture data which is what we’re talking about here. Sony didn’t spend years building a system like SFS as far as we know and they didn’t have to

No, it does not have a 2-3x multiplier over PRT and developer driven streaming. Re-read what the MS engineer I quoted posted and re read their darn specs: SF is a hardware assisted hint to prefetch textures you are likely to need next frame (aka triggering page reads).

Virtual Texturing, PRT, is what delivers the RAM and bandwidth savings you are talking about and that in itself is GPU support for a technique that, at extra shader cost and game complexity and varying results, developers were implementing on their own.
You are overselling it.

 

[Jigsaah]

Marketing trying to spin an improvement in efficiency as a revolution that would close the gap with PS5’s SSD betting on the fact that people forgot Partially Resident Textures were a thing (and give you this memory and bandwidth multiplier in that fashion).

From the horse’s mouth:



Samplers Feedback, aka PRT+, defined in the specs:https://microsoft.github.io/DirectX-Specs/d3d/SamplerFeedback.html

So then are you saying because it's been around that this updated version of it could not possibly match what the PS5 is doing with the SSD? Or is it left to be seen maybe?

I guess here's what I'm considering. Xbox has the raw power advantage, but how is this leveraged? Does Raw power = Speed? higher fidelity? Can it do both without compromises?

For the PS5, we know its fast...so no guesses about that, but with 10 TF (variable, though only slightly I suppose) does the SSD make up for the lack of Raw Power.

I'm not choosing a side here, but if anybody has ever watched Dragonball Z when Super Saiyan 2 Trunks go against Super Saiyan Vegeta...well Vegeta won because Trunks was too bulky to move quickly. So he could hit hard but couldn't hit his opponent because he was too slow.

Too bad Microsoft and Sony aren't punching each other. It's more like they're punching the same target, namely multiplatform games or the latest prized example, the UE5 demo.

One can hit really hard but slower, the other can hit not as hard, but faster. So to end this terrible analogy, the question ultimately becomes, what game are the systems taking on. I get the feeling that in some cases Xbox Series X could perform better and in others PS5 might have the advantage.

 

[Panajev2001a]

SFS is a suite of hardware features built custom for the XSX GPU that include Sampler Feedback. Sampler Feedback is an improvement over PRTs but still even on the XSX, it will have custom texture filters not available in the PC hardware. Basically MSFT took years of PRT data in the Xbox One and found there were large percentages of texture data in ram that were actually never used. SFS eliminates this waste by streaming only pieces of textures as they are needed. This is a huge advantage over all other next gen systems.

At this point you are just regurgitating their PR straight.

 

[Bernkastel]

And so...this wouldn't like...lessen the quality of the textures?

SFS?

3:28 onwards

 

[ToadMan]

Because the XSX SSD is fast enough, 100GB is instantly accessible that's MORE than enough. The GPU and CPU are beasts that easily can process all that information. RAM is just normal, a bit better than PS5, but nothing major.

Why I'm saying that PS5 is unbalanced is because the SSD can even get more data, but their CPU/GPU isn't powerful enough to process more. You've heard what Cerny said. The CPU and GPU can't both be at full power, so one is always capped lower. That's why I said extremes. Extremely fast SSD, lesser CPU/GPU, unbalanced.

That isn’t what Corny said.

Cerny said the max clocks are capped and the power usage is capped. The variable then is compute load.

Both the gpu and cpu can run at 100% clock all day. The developer’s job is to design their code to work within the cumulative power cap Sony have targeted.

 

[Panajev2001a]

So then are you saying because it's been around that this updated version of it could not possibly match what the PS5 is doing with the SSD? Or is it left to be seen maybe?

I guess here's what I'm considering. Xbox has the raw power advantage, but how is this leveraged? Does Raw power = Speed? higher fidelity? Can it do both without compromises?

For the PS5, we know its fast...so no guesses about that, but with 10 TF (variable, though only slightly I suppose) does the SSD make up for the lack of Raw Power.

I'm not choosing a side here, but if anybody has ever watched Dragonball Z when Super Saiyan 2 Trunks go against Super Saiyan Vegeta...well Vegeta won because Trunks was too bulky to move quickly. So he could hit hard but couldn't hit his opponent because he was too slow.

Too bad Microsoft and Sony aren't punching each other. It's more like they're punching the same target, namely multiplatform games or the latest prized example, the UE5 demo.

One can hit really hard but slower, the other can hit not as hard, but faster. So to end this terrible analogy, the question ultimately becomes, what game are the systems taking on. I get the feeling that in some cases Xbox Series X could perform better and in others PS5 might have the advantage.

I am saying that SF, without knowing what feedback mechanism PS5 provides, makes it easier for more developers to achieve an efficient self improving texture streaming mechanism.

Fully Software driven virtual texturing (think Mega Textures as one of the many implementations) -> PRT + software driven virtual memory pages management -> SFS or PRT+ or PRT + HW driven pre-fetching (helps you from having to manage virtual texture memory yourself and triggers page reads for data it thinks you may want to read next also avoiding a page fault and extra latency).

So, no, SFS does not close the gap IMHO.

 

[Panajev2001a]

3:28 onwards

Thanks for sharing this link again, she gave a really good presentation indeed.

To go back to to the topic at hand though, when talking about the texture streaming scenario with SF she said:

Consistent with:

 

[rntongo]

No it does not have a 2-3x multiplier over PRT and developer driven streaming. Re-read what the MS engineer I quoted posted and re read their darn specs: SF is a hardware assisted hint to prefetch textures you are likely to need next frame (aka triggering page reads).

Virtual Texturing, PRT, is what delivers the RAM and bandwidth savings you are talking about and that in itself is GPU support for a technique that, at extra shader cost and game complexity and varying results, developers were implementing on their own.
You are overselling it.

Why are you nitpicking things. Let's quote the same engineer since you want to misunderstand things. SF is only one part of SFS and will have a 2-3x multiplier whether you want to believe it or not.

SF is only one part of SFS and even it will have custom texture filters. SFS is not a part of DX12U:


Custom texture filters for XSX offer an advantage over SF in PC builds and users may experience pop in:




Example of cutting a texture by up to 75%, This would reduce the RAM space needed by 4x. That's why the numbers of 2-3x I/O multiplier are not hypotheticals like you suggest.

 

[ToadMan]

Sounds like 2013 repeats itself with the difference of who's in the lead.

Actually that’s an interesting comparison.

Back at PS4/x1 release, Sony had nearly 40% tflops advantage and much faster memory to boot. The PS4 has 18 CUs running at 800Mhz vs xb1 with 12 CUs at 850Mhz. The PS4 had 50% more CUs running about 7% slower.

Comparing multiplats, that got the PS4 one step more resolution and the same FPS as x1.

So a 1080p/30fps game on PS4 was 900p/30fps on x1. Nothing like the PS3/360 differences.

Here we are with a tflop difference of around 18% (PS5 36 CUs at 2.2GHz vs 52CUs at 1.8GHz in Xsex) with that advantage to x1 spread across more CUs individually running slower than the PS5 CUs.

I suspect that overall the power difference won’t be enough for one step of resolution difference between PS5 and Xsex. Maybe there’ll be a small FPS loss but that could go to x1 or PS5 - depends on whether more slower CUs is easier to get performance from than less but faster CUs in practice.

Overall I think straight power is a wash between the 2. Maybe the SSD tech gets Sony some advantages in throwing data around but for multiplats that won’t matter.

 

[rntongo]

At this point you are just regurgitating their PR straight.

You're cherry picking an engineer's comments why don't you include these from him?? You're using a false equivalency of SFS and SF yet SF is only one aspect of SFS in the XSX and will have custom hw not available in other SF implementations!! On top of having extra hardware for more efficient texture streaming.

 

[Panajev2001a]

Why are you nitpicking things. Let's quote the same engineer since you want to misunderstand things. SF is only one part of SFS and will have a 2-3x multiplier whether you want to believe it or not.

SF is only one part of SFS and even it will have custom texture filters. SFS is not a part of DX12U:


Custom texture filters for XSX offer an advantage over SF in PC builds and users may experience pop in:




Example of cutting a texture by up to 75%, This would reduce the RAM space needed by 4x. That's why the numbers of 2-3x I/O multiplier are not hypotheticals like you suggest.

You are cherry picking and mixing things out of context trying to imply the same marketing PR spiel about SF bringing 2-3x or more memory and bandwidth improvements over PRT in say RDNA1.

I am not downplaying SFS nor the extra HW filter they added to reduce the visible impact of LOD/texture switching as higher quality data is streamed in.
You are acting as if something you cannot pinpoint in SFS that is not DX12U SF (which itself is not a super duper revolution over PRT) is the reason for a 3x bandwidth and memory storage multiplier.
It is as if the XSX extensions for SFS not part of DX12U already were Earth shattering and in the same tweets you quote for example you can see this:

I only said Sampler Feedback was part of DX12 Ultimate. You can use it to figure out what texture pages to stream, but without our custom texture filters, you might notice "pop in" at page boundaries.

You are patching together PR statements and cherry picked data left and right to support a 2-3x or more memory storage and bandwidth multiplier over PRT which has no basis in either of those tweets or the technical presentation that was linked beyond well worded marketing PR intended to hype the product up (as it should) and choosing a baseline which is not a state of the art virtual texturing solution in a modern GPU.
Virtual Texturing, to avoid uploading the full quality texture (or the full mio map chain for example) or unused textures, is something developers have been doing since the PS2 days or well before actually (with different levels of quality and efficiency/rendering cost).

More recently an implementation that was trying to unify it under a single banner and keep draw cal cost to a bare minimum was id’s MegaTexture tech... all data points you are conveniently ignoring in your efforts to market XSX around.