Support NeoGAF

[Panajev2001a]

You are correct, you didn't see a baseline. That's because those slides were not a technical specification of the component. Just a description of what it can achieve.
Cerny did not provide a baseline for the clocks rates in the PS5. We know the max speeds but nobody knows how low they can fall. They're "variable" after all. Yet I don't see too many people scream smoke screen maneuver and winge for a baseline, at least not you.
All we have so far is the claims from the platform holders. I will never hold against hem to make claims. Just to make claims that could be disproved. I actually believe that this claim from MS (100GB instantely accessible) is too specific to be made up, as any indie developer with access to a dev kit would be in a position to disprove it the following day and affect the platform's reputation.
Personally I'm done watching the continuous grandstanding in speculation threads for the sake of ego and platform honor. There are more important things happening in the world.

Again, I am not claiming MS’s claims are made up, I do not appreciate my arguments being twisted as such.
I said that the baseline taken as a game already taking advantage of virtual texturing and streaming (with or without SF or PRT) is putting words in MS’s engineers mouth as they did not say that.

Compared to PC’s, the improvements are there and massive for both new consoles... and they should with how much they have invested. MS says instant access to 100 GB: that meaning that programs can access that memory directly (as you do with virtual memory on PC, another claim they made about treating the SSD as virtual memory, now they defined how much they are using as a page file of sorts) and thanks to a fast SSD channel and quite well thought out HW decompression engines as well as a very advanced new API (DirectStorage) they can do very quickly and at a low CPU cost.

XVA is very interesting tech without having to exaggerate things.

 

[Trueblakjedi]

This is the crux of your misunderstanding (unintentional or not). I am taking Cerny and Goossen/Stanard at their word just like you are doing to all three of them.

They are not claiming what you are trying to make them claim, so your appeal to authority does not hold in this case: I am taking them at their word and if they gave more context we could understand what their word means instead of building up fantasies. 2-3x is a relative improvement claim: if you think the baseline includes a game that already implements virtual texturing (purely in software/shaders or using PRT/tiled resources), what is the evidence in what those engineers have been saying?

I am also not calling people fool, so I would like words not being put in my mouth, thanks .

Considering that I am the one on twitter asking James Stanard the questions regarding the multiplier ... i think his explanation and my understanding are in sync.

We are just out of sync with your belief. And that's OK. Thats not on us.

Its fair for you to wait until further information is made available.

 

[Panajev2001a]

Considering that I am the one on twitter asking James Stanard the questions regarding the multiplier ... i think his explanation and my understanding are in sync.

We are just out of sync with your belief. And that's OK. Thats not on us.

Its fair for you to wait until further information is made available.

It could be Pluto asking him the questions and it would make no difference. You know the questions you are asking and he knows what he is answering. It would not be the first time people misinterpret each other.

You are putting words in his mouth and/or in the mouth of the people who wrote the DF piece you quoted. You are claiming a baseline they have not specified (“we have taken XYZ games that already implemented virtual texturing efficiently and SFS delivers a 2-3x I/O bandwidth and memory savings improvement on top of that”), not sure how it is not on you to prove that.

You made a claim, you are not supporting it with evidence or supporting it with evidence that is fair game to discuss on (and that so far cherry picks several different tweets, articles, tech talks, etc... and what we disagree is how they are assembled, not what each of them says), and that is it...

 

[polybius80]

Unless... the memory addresses are independent NVME locations.


Plus... don't forget the "+"

What does the plus refer to?

Its PRT plus something(s) which do not currently exist in PRT.

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

the "+" its because it fixes some problems as explained with the mip maps, if there was not a mip map available then it has to be resolved to continue, with SFS there is at least a lower mip map available that will be used so it can be resolved a couple frames later and be replaced with the correct mip map(higher)

 

[Dodkrake]

When was the first time Sony officially mentioned this? That their SSD will change the way game worlds are made?

First time I'm aware? In Cerny's talk. Why is that relevant?

 

[polybius80]

I think that the initial perspective on how we analyze claims should not be from the perspective of "x claims a boost so apply that same boost to another system."

The boosts stated are only for the system being discussed. The other system its own boosts achieved differently so XSX boost tech doesn't apply to PS5.

Once we stop with that initial frame of reference we can evaluate xsx on its own stated merits.

Stanard says that the boosts are stacked. So you apply the benefits of SFS then apply bcpack and BC7 LZ compression for what your gains to throughput mean.

Simple clear and seemingly obvious.

2-3 gains means either a reduction of almost 75% in texture size or and equivalent throughput rate of 2.5 times the stated rate or 2.4 * 2.5 =6gb/s

Halving that again is the equivalent of 12...

The same exact way the PS5 can mathematically achieve its compressed throughput gains.

Everyone is making this harder than it really is.

the 2-3 gains is not on top of existing PRT techniques, its in the eurogamer article

"We observed that typically, only a small percentage of memory loaded by games was ever accessed," reveals Goossen. "This wastage comes principally from the textures. Textures are universally the biggest consumers of memory for games. However, only a fraction of the memory for each texture is typically accessed by the GPU during the scene. For example, the largest mip of a 4K texture is eight megabytes and often more, but typically only a small portion of that mip is visible in the scene and so only that small portion really needs to be read by the GPU."

As textures have ballooned in size to match 4K displays, efficiency in memory utilisation has got progressively worse - something Microsoft was able to confirm by building in special monitoring hardware into Xbox One X's Scorpio Engine SoC. "From this, we found a game typically accessed at best only one-half to one-third of their allocated pages over long windows of time," says Goossen. "So if a game never had to load pages that are ultimately never actually used, that means a 2-3x multiplier on the effective amount of physical memory, and a 2-3x multiplier on our effective IO performance."

he is basically describing the possible gains from using granularity vs not using, and that depends the game, there is no implication this make gains on top of already granular, unless his intention is to not have most mip maps in RAM and only keep low mip maps and get the ones nedded on deman wich in any case can be implemented on any PRT algorithm and have to be used carefully as you are introducing pop up for every texture and relying in SSD speed to resolve before is too notorious

 

[Three]

No. Everyone including MS says and "effective 2x or 3x (or higher) multiplier on both amount of physical memory and SSD performance."

Effective is key word here.

Now at 4 lanes we know that lane utilization is the key to the speed of the spec. You only need one lane to get to 2.4GB/s... what are they using the other 3 lanes for?

The 2x multiplier is not about lanes. It is specifically mentioned that it is in reference to only loading textures that are needed for the scene. Even confirms it twice with one third of the textures. Why confuse the subject by talking about lanes? It's fine to discuss the merits of a system but as long as we are actually discussing its merits and not looking for secret sauce we claim that we don't actually know about. People shouldn't just be linking everything.

Nothing good can come of that. If we discuss features and even speculate on features and do so separately and logically instead of muddying the water by mixing every sort of unrelated tech under the sun to win an argument (not you specifically) then this thread can stop going in circles. At this point I think maybe this is just to piss some people off and keep bumping this thread.

 

[Panajev2001a]

the 2-3 gains is not on top of existing PRT techniques, its in the eurogamer article


it basically is describing the possible gains from using granularity vs not using, and that depends the game

Indeed, the quote says “For example, the largest mip of a 4K texture is eight megabytes and often more” —> entire texture and not a texture tile/block... not what you would have in memory with tiled resources/PRT.

 

[OptimistPrime]

You are claiming a baseline they have not specified, not sure how it is not on you to prove that.

It‘s pretty obvious that the baseline is “...compared to not using Sampler Feedback Streaming.“

It’s reasonable to assume this includes other common techniques to improve efficiency. I would say that it’s a pretty specific and unproven claim to say that this improvement excludes PRT/tiling. I don’t remember reading from Microsoft that it’s an “effective multiplier of 2X or 3X for games that don’t stream textures.” You’re arguing against a very plain reading.

 

[OptimistPrime]

Indeed, the quote says “For example, the largest mip of a 4K texture is eight megabytes and often more” —> entire texture and not a texture tile/block... not what you would have in memory with tiled resources/PRT.

That quote doesn’t mention texture compression either, so do you want to also add that to your undeniable claim that 2X-3X performance improvement cannot be attributable to texture streaming? That way you can say the 2X-3X improvement they were talking about was compared to uncompressed textures that were not streamed.

You know what, come to think of it, they also didn‘t mention that the games that constituted their baseline were made by professional developers. Probably the real baseline is unoptimized, uncompressed games developed by (unruly) children. Everyone must prove that this is not the baseline.

 

[Trueblakjedi]

the 2-3 gains is not on top of existing PRT techniques, its in the eurogamer article




he is basically describing the possible gains from using granularity vs not using, and that depends the game, there is no implication this make gains on top of already granular, unless his intention is to not have most mip maps in RAM and only keep low mip maps and get the ones nedded on deman wich in any case can be implemented on any PRT algorithm and have to be used carefully as you are introducing pop up for every texture and relying in SSD speed to resolve before is too notorious

What you guys are arguing is nonsensical.

These are absolute claims vs

The size of textures maps that would normally be transported.

The number of textures that could be transported.

In the 2.4/GBs pipe.

Select Only transport half or a third of a texture then compress by 50%.

The end.

Very simple.

 

[Panajev2001a]

It‘s pretty obvious that the baseline is “...compared to not using Sampler Feedback Streaming.“

No, it is not. 200% improvements over PRT / developer implemented efficient virtual texturing is jaw droppingly massive. You are understanding it and confirming MS never claimed, at least we moved one step foeward

It’s reasonable to assume this includes other common techniques to improve efficiency. I would say that it’s a pretty specific and unproven claim to say that this improvement excludes PRT/tiling. I don’t remember reading from Microsoft that it’s an “effective multiplier of 2X or 3X for games that don’t stream textures.” You’re arguing against a very plain reading.

I am taking them at their word knowing that they were painting as a positive and realistic enough picture as they could without disappointing or misleading people (hence why they went with 4.8 GB/s or equivalent compressed data rate instead of using the BCPack decompressor unit’s max throughout of 6 GB/s).

They were very specific about their TFLOPS claims, the CPU and GPU frequency claims, etc... everything that really did well to hype the console’s potential up... yet they were cagey about the detail on this. You are the one claiming a much better than best case scenario here and claiming that the 2-3x improvement is on top of PRT/virtual texturing schemes. The burden of proof for this is on you.

 

[Trueblakjedi]

It could be Pluto asking him the questions and it would make no difference. You know the questions you are asking and he knows what he is answering. It would not be the first time people misinterpret each other.

You are putting words in his mouth and/or in the mouth of the people who wrote the DF piece you quoted. You are claiming a baseline they have not specified (“we have taken XYZ games that already implemented virtual texturing efficiently and SFS delivers a 2-3x I/O bandwidth and memory savings improvement on top of that”), not sure how it is not on you to prove that.

You made a claim, you are not supporting it with evidence or supporting it with evidence that is fair game to discuss on (and that so far cherry picks several different tweets, articles, tech talks, etc... and what we disagree is how they are assembled, not what each of them says), and that is it...

So you are saying that you, "Gafs pleasant genius" has more insight into what I asked and what was answered than either myself or Stanard does.

He is always willing to clarify his statements and he did not feel the need to do so.

Now I've heard everything.

Me and several other people asked if the gains were stacked so, SFS+compression techniques to yield greater gains than either technique separately.

He said yes. Asked and answered.

I asked him why the stated compression rate was substantially below the hw decompressison rating.

He said because it was a typical rate but not potentially the maximum that the system was capable of.

He also mentioned under separate cover, that he and his team were making great strides in the development of the Xbox Texture Compression codec.

Asked. Answered. Very clear.

Its only the folks in this thread debating the meaning of "is" and its not fruitful.

 

[Dee_Dee]

Didn’t know we would have to wait until August for MS to go over the architecture...that’s so far from now smh

 

[polybius80]

What you guys are arguing is nonsensical.

These are absolute claims vs

The size of textures maps that would normally be transported.

The number of textures that could be transported.

In the 2.4/GBs pipe.

Select Only transport half or a third of a texture then compress by 50%.

The end.

Very simple.

sorry but claims require a technical explanation to back them up, specially after "the power of the cloud" claims and the results I am not saying it is wrong just for the loss of credibility of past claims but they require an explanation and it happens they explained a bit in DF article and it differs from what you say they claim

in the DF interview yes there are gains about 2-3x but they are not on top of the 2-3x gains already from PRT and similar techniques, also this gains depend from the game they are not absolute on every game and they dont make those claims in DF article they specifically explain the situation that allows to cull textures that is what they claim and they dont claim its over an already partially resident algorithm they mention they counted how many textures where accessed in many game in current gen in typical scenes and there is an oportunity to cull them achieving 2-3x less texture but again they dont claim it on already PRT textures, PRT algorithms already do that by themselves in games this gen and even past generations, and PRT techniques dont apply to every game and not every game may require to cull texture space or administer its textures equally as it can introduce pop-up, blurry textures and even stalls, in fact SFS fixes a problem where there wasnt available a mip map and had to wait to resolve

 

[Trueblakjedi]

sorry but claims require a technical explanation to back them up, specially after "the power of the cloud" claims and the results I am not saying it is wrong just for the loss of credibility of past claims but they require an explanation and it happens they explained a bit in DF article and it differs from what you say they claim

in the DF interview yes there are gains about 2-3x but they are not on top of the 2-3x gains already from PRT and similar techniques, also this gains depend from the game they are not absolute on every game and they dont make those claims in DF article they specifically explain the situation that allows to cull textures that is what they claim and they don't claim its over an already partially resident algorithm they mention they counted how many textures where accessed in many game in current gen in typical scenes and there is an opportunity to cull them achieving 2-3x less texture but again they don't claim it on already PRT textures, PRT algorithms already do that by themselves in games this gen and even past generations, and PRT techniques dont apply to every game and not every game may require to cull texture space or administer its textures equally as it can introduce pop-up, blurry textures and even stalls, in fact SFS fixes a problem where there wasnt available a mip map and had to wait to resolve

Wrong. We are not bringing anything from your perception of Xbox One into Xbox Series X.
Just like we shouldn't bring any perceptions of PS4 Pro into PS5.

These devices and teams stand alone.

Your yearning to douse MS with your negative perception is unnecessary. In the future more will be explained.

We went through this process when Cerny did his marketing talk as well... and everything was accepted as true unique and new.

And while not extraordinarily technical it gave us enough insight to research how it would work. We are trying to do the same here. Except, what you and others call "misunderstandings", "exaggerations" and "marketing" which is simply a rhetorical tool to denying its validity as information, is what has been provided by the people responsible for building the system.

The principal architect of the PS 5 is ostensibly Cerny.
For MS its Goossen. We can talk to Goossen and he does direct interviews as well.
There are questions that Cerny DIDN'T answer when asked by DF directly. You have no problem with that.

Tell me who the architect of compression is for Sony? I dunno but we know James Stanard from MS and he talks to the public to answer questions not still under NDA.

Who designed their Back Compat for Sony? For MS its Bill stillwell and we can talk to him.

So we have 1) posted official material that 2) matches with multiple interviews from different sources, that 3) match with what the men responsible for developing system, say directly to the public when asked.. Yet we are still albatrossed by your displeasure with the cloud claim from 7 years ago.

Don't you think you are being extra?

 

[polybius80]

Wrong. We are not bringing anything from your perception of Xbox One into Xbox Series X.
Just like we shouldn't bring any perceptions of PS4 Pro into PS5.

These devices and teams stand alone.

Your yearning to douse MS with your negative perception is unnecessary. In the future more will be explained.

We went through this process when Cerny did his marketing talk as well... and everything was accepted as true unique and new.

And while not extraordinarily technical it gave us enough insight to research how it would work. We are trying to do the same here. Except, what you and others call "misunderstandings", "exaggerations" and "marketing" which is simply a rhetorical tool to denying its validity as information, is what has been provided by the people responsible for building the system.

The principal architect of the PS 5 is ostensibly Cerny.
For MS its Goossen. We can talk to Goossen and he does direct interviews as well.
There are questions that Cerny DIDN'T answer when asked by DF directly. You have no problem with that.

Tell me who the architect of compression is for Sony? I dunno but we know James Stanard from MS and he talks to the public to answer questions not still under NDA.

Who designed their Back Compat for Sony? For MS its Bill stillwell and we can talk to him.

So we have 1) posted official material that 2) matches with multiple interviews from different sources, that 3) match with what the men responsible for developing system, say directly to the public when asked.. Yet we are still albatrossed by your displeasure with the cloud claim from 7 years ago.

Don't you think you are being extra?

that is a lot of text just to say you dont know, I dont care if Ms employees are very active in tweeter or the name of who is involved with back compat in sony, so far the DF interview dont say what you claim


James Stanard has not made such claim in his twitter(I have just checked every one from dec 2019 to this day, I dont use twitter in my PC so have to use someones else) he actually mention that he cannot share lot of things so is not as open to questions as you say, in fact he says sampler feedback is part of DX12 and reffer to it as similar or equal to SFS so he says "windows games can take advantage too" and I already provided the DX12 description for sampler feedback and how it works and what it do compared to PRT, he actually mentions the filtering that allows to fallback to the resident mip map(low mip map) in XBSX I assume is the only advantage over SF in DX12, but that does not provide such advantage and there is no claim of 2-3x on top of already PRT or SF

Sampler Feedback

Engineering specs for DirectX features.

microsoft.github.io

apparently I touched a nerve with my comment of the cloud, I apologize for that

 

[Panajev2001a]

So you are saying that you, [...] has more insight into what I asked and what was answered than either myself or Stanard does.

I do not claim insights on what he thinks or what you meant to ask. I am just reading what has been written like anyone else. You are reading into things... you are adding meaning and connecting data to support your foregone conclusion. IMHO, at the moment at least, this is more a case of confirmation bias than anything else.

Me and several other people asked if the gains were stacked so, SFS+compression techniques to yield greater gains than either technique separately.

He said yes. Asked and answered.

As I said in my previous posts, I agree. This was and is not being disputed: SFS and compressione benefit do stack up.
Not sure what is this supposed to prove though... having those two factor stacking up does not change the question of what the baseline is and you getting more aggressive does not change it.

I asked him why the stated compression rate was substantially below the hw decompressison rating.

He said because it was a typical rate but not potentially the maximum that the system was capable of.

Agreed, you got him to state that the BCPack decompressor has a higher than 4.8 GB/s theoretical max throughput, which people already knew from the DF article, and that when given the equivalent compression rate speed they went for a realistic average estimate instead of the theoretical max of the unit (same thing that Sony did when gong with 8-9 GB/s instead of 22 GB/s or the theoretical throughput of the Kraken HE decompressor).

He also mentioned under separate cover, that he and his team were making great strides in the development of the Xbox Texture Compression codec.

That is very good and exciting to see more data about it in August .

Again, the baseline for the 2-3x SSD I/O bandwidth and memory savings improvements was not addressed in any of the comments you just made. We are back to it being your own interpretation of what has been said, so your claim to prove not his. What he did claim checks out.

 

[ToadMan]

We went through this process when Cerny did his marketing talk as well... and everything was accepted as true unique and new.

That’s because Cerny was careful in his tech talk to specify what his comparison baselines were - usually PS4.

The numbers in use here 2x 3x are relative to? .... hmmm not clear.

Ok nevermind, 100gb “instant access” - what does instant mean? Ok I know what it means - but some others think that this is a physics defying special memory solution for which there is 0 latency.

The problem is MS has introduced marketing into its “technical” descriptions - XVA is a marketing term itself - so the question is how should anyone understand what is technically possible/achievable if MS are playing fast and loose with the terms they’re using.

 

[Deleted member 775630]

XVA is a marketing term itself

XVA is as much as a marketing term as SmartShift which has been used constantly by Cerny. These things just need to have names so people can clearly say what they are talking about.

 

[Panajev2001a]

XVA is as much as a marketing term as SmartShift which has been used constantly by Cerny. These things just need to have names so people can clearly say what they are talking about.

Smartshift is an AMD patented and trademarked technology so he just really referenced a third part term after describing how it works... used it constantly? Really read it once in the article and heard it like once in the video: taking about how it works and being careful about what he said and the baselines of his comparisons may cause some fans to be upset at not having more details, but he does not strike me as the Jobsian’s RDF type either.

 

[Deleted member 775630]

Smartshift is an AMD patented and trademarked technology so he just really referenced a third part term after describing how it works... used it constantly? Really read it once in the article and heard it like once in the video: taking about how it works and being careful about what he said and the baselines of his comparisons may cause some fans to be upset at not having more details, but he does not strike me as the Jobsian’s RDF type either.

It's a marketing term, but that doesn't mean it doesn't have any value. There's a reason why we give these things proper names, because otherwise you might not properly understand what someones talking about beause it's too technical. SmartShift and XVA are both marketing terms, but there's nothing wrong with that. That's my point.

 

[Panajev2001a]

It's a marketing term, but that doesn't mean it doesn't have any value. There's a reason why we give these things proper names, because otherwise you might not properly understand what someones talking about beause it's too technical. SmartShift and XVA are both marketing terms, but there's nothing wrong with that. That's my point.

It has tons of value marketing wise, but it can help muddle the waters too: as someone pointed out it was clever but a bit sneaky to do such a thing too (which is what marketing is about, from MS or Sony).

XVA was a great way to convey that it was cool tech and to generate hype and I am sure some people think it means additional positive things, such as it all being a mastermind idea by MS Sony must have copied at the last minute or something, statements which MS did make nor endorses but I bets its marketing department loves for people to run wild with them.

Still, the point stands. On the SSD I/O they went a bit more marketing than data driven/fully open as they were with other details of their machine and it muddled the waters as made the baseline of their comparisons shifty and under defined and it shows.

 

[longdi]

Anyway we seen snippets of fast next gen i/o, spiderman demo, xbox games switching and UE5.

I guess they are nice and all, but loading is still loading. I mean UE5 demo can be done visually with lower i/o speeds according to Epic China.

 

[Panajev2001a]

Anyway we seen snippets of fast next gen i/o, spiderman demo, xbox games switching and UE5.

I guess they are nice and all, but loading is still loading. I mean UE5 demo can be done visually with lower i/o speeds according to Epic China.

... and the cycle starts anew: pendulum swings from the debunked Chinese video evidence meant to downplay SSD speed important and in a day or less to a claim that actually speed matters and a secret feature is tripling or quadrupling it.

It is starting to look like a predictable script, but decently coordinated perhaps...

 

[Redlight]

... and the cycle starts anew: pendulum swings from the debunked Chinese video evidence meant to downplay SSD speed important and in a day or less to a claim that actually speed matters and a secret feature is tripling or quadrupling it.

It is starting to look like a predictable script, but decently coordinated perhaps...

This is a thread about a Series X feature where a number of interested people are discussing the possibilities as they know them.

Strangely this Xbox thread is haunted by a small number of hardcore Sony cheerleaders who relentlessly argue, largely in bad faith, that anything seen as a potential Xbox positive is really nothing at all. When called on it some, like yourself, play the victim card as if any closing of the I/O gap would rain all over your Sony shaped parade and must not be allowed.

Ad infinitum.

You know what? I guess that... "It is starting to look like a predictable script, but decently coordinated perhaps..."

 

[jimbojim]

Strangely this Xbox thread is haunted by a small number of hardcore Sony cheerleaders who relentlessly argue, largely in bad faith, that anything seen as a potential Xbox positive is really nothing at all. When called on it some, like yourself, play the victim card as if any closing of the I/O gap would rain all over your

Yet the OP is also crowded with comparison against PS5 and Kraken compression method. /s

 

[ZywyPL]

It is starting to look like a predictable script, but decently coordinated perhaps...

Oh the irony...

 

[Panajev2001a]

This is a thread about a Series X feature where a number of interested people are discussing the possibilities as they know them.

Strangely this Xbox thread is haunted by a small number of hardcore Sony cheerleaders who relentlessly argue, largely in bad faith, that anything seen as a potential Xbox positive is really nothing at all. When called on it some, like yourself, play the victim card as if any closing of the I/O gap would rain all over your Sony shaped parade and must not be allowed.

Ad infinitum.

You know what? I guess that... "It is starting to look like a predictable script, but decently coordinated perhaps..."

You keep accusing me of a perceived slight against your platform of choice, throwing big words like “bad faith”, etc... and defending the choice to make claims up, not providing supporting evidence and/or being willing discuss it but essentially demand people accept it just because... reasons... am I the one playing the victim here? How exactly?

I am discussing the possibilities just as you are, Looking at the tech, willing to dive deeply into news reported. If you want this thread as a safe space where it is great to just post marketing data with your own spin to it, but are willing to take the onus of proof, which is a overemphasised way to say being willing to discuss it... fine too .

 

[oldergamer]

People here need to stop arguing and holding up in their corners.

1. Yes this is not a entirely new implementation of Megatextures/PRT/SFS by microsoft. It is however solving some of the problems with the feature that have existed since 2011
2. Yes all of the specific changes made by MS to now be in hardware, could be implemented in software in some manner. After all, what graphical feature on these video cards couldn't be? However, do this in software is at the sacrifice of performance, quality, memory, bandwidth, etc...
3. PRT in hardware has existed for nearly a decade, but in most cases it has rarely been used. Why don't all games use it? Well that's because it can be a challenge to implement if you didn't set yourself up correctly from the beginning, and it still comes with its own set of visual challenges that needs to be resolved.
4. This isn't a solution in search of a problem. The issues with handling mip-maps existed since PRT hardware was added to video cards. Bottom line is this is a hardware feature that makes it easier for developers to make full use of PRT, without as many drawbacks.
5. Games that already stream textures can make use of the new hardware features, and still get a benefit of improved performance and less visual artifacts. Games that don't support PRT, can now add it and get a benefit of improved performance, and pushing more textures into their games (improving visuals)and lowering bandwidth usage

Can we stop arguing about this now? The new hardware supported feature will still benefit any developer that wants to use it.

Id recently moved away from megatextures/PRT. They said that it made their textures blurry and the decision to move away from it allowed them to compete with other games using higher res textures. Texture pop-in is one of the worst visual artifacts we see on a regular basis today. We've basically seen this on multiple engines when streaming textures. So anyone saying, "oh it just fixes pop-in" is trying to minimize the impact. I for one will be happy if i don't see this in 1080p or 4K. IMo this is as bad as screen tearing, and happens often in games that stream. Note i've seen worst cases then this, but couldn't find a good gif.



So, a few pretty thought out posts at B3D from Ronaldo. I'm going to quote them below as they seem well thought out.:

An interesting thought experiment resulting from sampler feedback streaming is to try and estimate the bandwidth requirementfor a typical case of use.
80 MB is a fair estimate for a 4K framebuffer.
The automatic fallback to a lower MIP and seamless transition to the required one provides us with n frametimes to stream the required tiles from the SSD. (note that this is due to the fact that the bandwidth of RAM is orders of magnitude higher than that of an SSD and plagues all implementations of virtual texture streaming . Having a "superfast SSD" (as per MS and Sony PR) won't eliminate it.)
Let n = 2 and the frame be 1 frame/s.
Scenario 1: 100% of the textures are not immediatley available in memory and have to be streamed

Estimated throughput requirement = (Framebuffer/(frametime x 2) = (80/(2/60)) = 2.4 GB/s

However, this scenario represents an overly aggresive form of texture sampling and most of the data required will already be found in memory (that's why a 99% culling is recommended in the github whitepaper for sampler feedback.No free lunch.)
Let's say 50% of the framebuffer is actually required (extremely generous). Then,

Estimated throughput requirement = (40/(2/60)) = 1.2 GB/s.

If n is increased to 3,

Estimated throughput requirement = (40/(3/60)) = 0.8 GB/s.

So this is an interesting thought experiment. Especially when you look at what Epic did with unreal 5. They are rendering what is visible, but at the same time saying, this is made possible with the PS5 SSD. In all those threads where sony fanys were saying "this is an advantage over xbox", i said a few times, we don't know how much bandwidth Epic is using. So realistically, is sony's 5.5GB/s giving them an advantage over 2.4 GB/s? Especially in the case where you are rendering only what you see? I think the answer to that is no. Not unless the system with 2.4GB/s SSD is bandwidth starved. Epic was rendering only what was visible, so there is no chance the rest of the bandwidth was used up by, audio or geometry. Brings me to this Ronaldo quote:

Maybe its a failing on my side...but what I was in fact addressing is the adequacy of the XSX's SSD for implementig a top-end texture streaming solution. Both a competent VT solution and SFS will decrease BW requirements and that's the point (but SFS will be the more efficient and effective solution because of a much lower computational overhead due to being a hardware implementation.) Boasting about having an X Gb/s capable SSD is meaningless.

Right, this is an important point and goes to what i was thinking right after i had time to digest what i saw from the epic demo. They are doing all they can to save bandwidth, this kinda goes to the speed of the SSD being meaningless. Fair point FAF & Panjev? Also ties into what epic said afterwards about this running great on other SSD's or NVME drives on PC or Xbox. Can we all agree to stop the dick waving now? Can we all stop reposting the same damn xbox articles/links and agree that that the epic demo likely is NOT using all the SSD bandwidth available? Which makes the most sense since its a multiplatform engine.

Also regarding latency Ronaldo recently wrote:

The discussion about cache scrubbers for the PS5 has got me thinking about the importance of latency for shader performance. Obviously, this was a prticular area of focus for Cerny and SIE R&D. It turns out that MS was also fretting about latency and cache misses but in a fashion typical of engineering, went about providing a solution in a completely different way. Andrew Goosen (that guy again), Ivan Nevraev and others came up with a novel method to improve cache prefetching in a patent titled : "Prefetching for a Graphics Shader" (US10346943B2).
They claim to be able to be able to "greatly reduce or eliminate the latency issues by ensuring that the subsequent cache lines of the shader are available in a cache before the shader executes those subsequent cache lines without pauses/delays in the execution of the shader for any further requests of cache lines" by using a "purposedly configured GPU" with a prefetcher block that can execute contemporaneously with the shader.

Can we discuss the above point, I'd like to hear more thoughts on latency.

 

[longdi]

Im kind confused, i thought next gen i/o can stream polygons or reyes. You know like UE5 said, no need for fake texture maps, most of the time?

 

[longdi]

... and the cycle starts anew: pendulum swings from the debunked Chinese video evidence meant to downplay SSD speed important and in a day or less to a claim that actually speed matters and a secret feature is tripling or quadrupling it.

It is starting to look like a predictable script, but decently coordinated perhaps...

imo, i think xcloud stuff is going to be more important to give gamers a better quality of life.
many enterprises i worked with, have move to the cloud.
I havent really spent time thinking of the ways, but the possibilities are out there to be exploited

 

[oldergamer]

Im kind confused, i thought next gen i/o can stream polygons or reyes. You know like UE5 said, no need for fake texture maps, most of the time?

Well when you have solved the bigger bandwidth issue with the biggest hog being textures (by rendering only what you see) , you still have lots of bandwidth to handle other items. I don't see why ultra high res geometry would be a problem

 

[Thirty7ven]

Can we all agree to stop the dick waving now? Can we all stop reposting the same damn xbox articles/links and agree that that the epic demo likely is NOT using all the SSD bandwidth available? Which makes the most sense since its a multiplatform engine.

You want to talk about Xbox I/O solution, but once again everything presented is in some way how it diminishes gap to PS5. There's a multipurpose thread that would better fit these comparisons.

I've been following up on the B3D thread, and your focus on what Ronaldo says is also starting to look a bit obvious. He's challenged in that thread but you don't bring those up. I mean so far everything Ronaldo posts is about propping up MS's solution while lessening the solution of the other guy. It's starting to get a bit obvious what his agenda was when he signed to B3D in May 18th.

 

[oldergamer]

You want to talk about Xbox I/O solution, but once again everything presented is in some way how it diminishes gap to PS5. There's a multipurpose thread that would better fit these comparisons.

We're talking about xbox hardware features that affect I/O bandwidth, and you take this as diminishing PS5? Feel free to discuss something else for all i care.

I've been following up on the B3D thread, and your focus on what Ronaldo says is also starting to look a bit obvious. He's challenged in that thread but you don't bring those up. I mean so far everything Ronaldo posts is about propping up MS's solution while lessening the solution of the other guy. It's starting to get a bit obvious what his agenda was when he signed to B3D in May 18th.

First problem is you are looking at that in a way that points to having an agenda yourself. Instead of seeing his points or trying to argue on the merits of discussion, you are trying to point out that HE has an agenda based on when he signed up? Gimme a break. Try reading what he wrote there, and looking at his argument. Otherwise, don't comment at me with that agenda based nonsense.

 

[martino]

imo, i think xcloud stuff is going to be more important to give gamers a better quality of life.
many enterprises i worked with, have move to the cloud.
I havent really spent time thinking of the ways, but the possibilities are out there to be exploited

atm this is even more nebulous than an I/O advantage

 

[oldergamer]

atm this is even more nebulous than an I/O advantage

Yeah lets leave the Xcloud stuff for a different thread. Very different set of issues.

 

[oldergamer]

Sorry if I'm cranky guys, i've become grumpy older gamer again

 

[semicool]

Saw this from Ronaldo8 on B3D, thought it was pertinent and interesting about MS approach to the need of cache misses and latency:

Start:

The discussion about cache scrubbers for the PS5 has got me thinking about the importance of latency for shader performance. Obviously, this was a prticular area of focus for Cerny and SIE R&D. It turns out that MS was also fretting about latency and cache misses but in a fashion typical of engineering, went about providing a solution in a completely different way. Andrew Goosen (that guy again), Ivan Nevraev and others came up with a novel method to improve cache prefetching in a patent titled : "Prefetching for a Graphics Shader" (US10346943B2).
They claim to be able to be able to "greatly reduce or eliminate the latency issues by ensuring that the subsequent cache lines of the shader are available in a cache before the shader executes those subsequent cache lines without pauses/delays in the execution of the shader for any further requests of cache lines" by using a "purposedly configured GPU" with a prefetcher block that can execute contemporaneously with the shader.

End

 

[ToadMan]

They are rendering what is visible, but at the same time saying, this is made possible with the PS5 SSD.

The problem of I/O latency isnt, and hasn't been since the 2d era, what is visible - the issue is what is not visible yet, where the data for that is stored and how it gets from there to the rasterizer. That is where data throughput becomes critical.

Your thought experiment only has meaning if your goal is displaying a static image - the UE5 demo was not static it was dynamic.

As soon as something is animated in 3D space dynamically, the movement of data becomes critical to fidelity. A lot of current gen processor cycles are spent giving the illusion of detail because loading more detail from the HDD will take too long - especially when one has to hold about 30secs worth of gameplay data in (limted) RAM.

With xsex SSD latency, it'll be putting the next 3 or 4 seconds of gameplay data in RAM (specifically in a subset of RAM to stay within the low latency RAM space) with a consequent boost in fidelity comapred to curent gen.

The XVA bits will assist - to what extent we don't know because there's insufficient data about its performance and what numbers MS have suggested are open to interpretation.

I imagine the reason Epic went with a PS5 is because XVA isn't ready for prime time and they wanted a solution baked into hardware to avoid any embarrasing moments during the demo.

 

[oldergamer]

The problem of I/O latency isnt, and hasn't been since the 2d era, what is visible - the issue is what is not visible yet, where the data for that is stored and how it gets from there to the rasterizer. That is where data throughput becomes critical.

Your thought experiment only has meaning if your goal is displaying a static image - the UE5 demo was not static it was dynamic.

I'm not sure i follow your logic on that. The thought experiment accounts for 60fps. I don't see how he was referring to a static scene.

 

[semicool]

The problem of I/O latency isnt, and hasn't been since the 2d era, what is visible - the issue is what is not visible yet, where the data for that is stored and how it gets from there to the rasterizer. That is where data throughput becomes critical.

Your thought experiment only has meaning if your goal is displaying a static image - the UE5 demo was not static it was dynamic.

As soon as something is animated in 3D space dynamically, the movement of data becomes critical to fidelity. A lot of current gen processor cycles are spent giving the illusion of detail because loading more detail from the HDD will take too long - especially when one has to hold about 30secs worth of gameplay data in (limted) RAM.

With xsex SSD latency, it'll be putting the next 3 or 4 seconds of gameplay data in RAM (specifically in a subset of RAM to stay within the low latency RAM space) with a consequent boost in fidelity comapred to curent gen.

The XVA bits will assist - to what extent we don't know because there's insufficient data about its performance and what numbers MS have suggested are open to interpretation.

I imagine the reason Epic went with a PS5 is because XVA isn't ready for prime time and they wanted a solution baked into hardware to avoid any embarrasing moments during the demo.

Wat? Hyperbolic, wild, ungrounded, unevidenced conjecture much? Well you did say you were using your imagination when you said "I imagine...".

 

[ToadMan]

XVA is as much as a marketing term as SmartShift which has been used constantly by Cerny. These things just need to have names so people can clearly say what they are talking about.

Except we know what SmartShift is, what it does and it's a third party term.

XVA is a name given to an aspiration. Same as "instant" access to 100gb.

They don't mean anything without details except the spin people choose to attach to the words.

 

[ToadMan]

I'm not sure i follow your logic on that. The thought experiment accounts for 60fps. I don't see how he was referring to a static scene.

I was talking specifically about the sentence you wrote - I didn't pay attention to the quote from other sources.

The contents of the framebuffer (that which is visible) is a solution to a problem - the problem is how to get a dynamic 3D logical model translated to 2D for display. It's also a solution which has to be updated at least 30 times per second in the face of changing parameters (the parts which are not currently visible but may be in the next frames).

Measuring the size of the solution - that would be the framebuffer - is not a valid way of determining the process or data manipulation that arrived at that solution. We don't know how the UE5 demo was running or the details of the paradigm they're persuing with UE5 - we just have some comments from Tim Sweeney.

For all we know, Epic abused memory and dumped texture caches here and there only to have to reload them and that's why needed a PS5 right now - their engine is a year away from release and isn't optimised. Or maybe they want their engine to run dynamically - to load meshes and textures on the fly from storage more often than not and will scale down for lower throughput machines.

In fact Sweeney talked specifically about SSD and hardware decompression being important - perhaps MS's software just isn't getting it done for Epic yet and that's why they chose PS5 to show this now.

Looking at what's "visible", estimating it's memory footprint and then leaping from there to a pronouncement about how it's rendered, is a collosal oversimplification on one hand and a massive assumption about Epic's tech on the other.

 

[oldergamer]

I was talking specifically about the sentence you wrote - I didn't pay attention to the quote from other sources.

The contents of the framebuffer (that which is visible) is a solution to a problem - the problem is how to get a dynamic 3D logical model translated to 2D for display. It's also a solution which has to be updated at least 30 times per second in the face of changing parameters (the parts which are not currently visible but may be in the next frames).

Measuring the size of the solution - that would be the framebuffer - is not a valid way of determining the process or data manipulation that arrived at that solution. We don't know how the UE5 demo was running or the details of the paradigm they're persuing with UE5 - we just have some comments from Tim Sweeney.

For all we know, Epic abused memory and dumped texture caches here and there only to have to reload them and that's why needed a PS5 right now - their engine is a year away from release and isn't optimised. Or maybe they want their engine to run dynamically - to load meshes and textures on the fly from storage more often than not and will scale down for lower throughput machines.

In fact Sweeney talked specifically about SSD and hardware decompression being important - perhaps MS's software just isn't getting it done for Epic yet and that's why they chose PS5 to show this now.

Looking at what's "visible", estimating it's memory footprint and then leaping from there to a pronouncement about how it's rendered, is a collosal oversimplification on one hand and a massive assumption about Epic's tech on the other.

Again not sure what you are going on about. Which sentence that i wrote are you replying to?

Also btw i spent 20 plus years in the game industry so u don't need to go into a diatribe about what constitutes " real - time " but thanks for that?

Anyway, Sure decompression is important when its needed. It was an example to make a point, of course its an over simplification. Thats all people do around here is make over simplifications, and you dont have a problem with statements like

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

Very clear oversimplification dont you agree?

If you think epic had these massive meshes that takes up so much memory that they had to be compressed and decompressed because what was rendered couldn't be done elsewhere, well be my guest. Im not falling for that.

If you dont like what i said or agree with it, go right ahead. Id be willing to make a ban bet with you that epic wasn't maxing out the bandwidth of the SSD in that demo.

 

[Redlight]

You keep accusing me of a perceived slight against your platform of choice, throwing big words like “bad faith”, etc... and defending the choice to make claims up, not providing supporting evidence and/or being willing discuss it but essentially demand people accept it just because... reasons... am I the one playing the victim here? How exactly?

I am discussing the possibilities just as you are, Looking at the tech, willing to dive deeply into news reported. If you want this thread as a safe space where it is great to just post marketing data with your own spin to it, but are willing to take the onus of proof, which is a overemphasised way to say being willing to discuss it... fine too .

You are perfectly entitled to hose down, concern troll and goal keep, though it would be best to do that in the open rather than pretending that you're just 'interested in discussing the tech'. Wouldn't you say that you're a pretty rusted-on Sony fan? That's certainly the way it reads.

You play the victim when expressing outrage that people would dare to keep discussing this. It's almost as if they're not listening to you! How dare they! Perhaps it's some kind of coordinated conspiracy by evil Xbots...

... and the cycle starts anew: pendulum swings from the debunked Chinese video evidence meant to downplay SSD speed important and in a day or less to a claim that actually speed matters and a secret feature is tripling or quadrupling it.

It is starting to look like a predictable script, but decently coordinated perhaps...

Btw, 'bad faith' is really two rather small words.

 

[Panajev2001a]

[...]

 

[Panajev2001a]

You play the victim when expressing outrage that people would dare to keep discussing this

This is not playing the victim mate... I also disagree that it was a play or even my point, but I am digressing. I am also not concern trolling nor spreading FUD... I am not sure if a fan makes up a claim that MS has not made based on what they think it can be implied from it and that seems to be baseless it is not concern trolling or FUD sparring, not sure how you could stretch it to be as such.

though it would be best to do that in the open rather than pretending that you're just 'interested in discussing the tech'.

I am interested in discussing the tech and do not really like that kind of astroturfing / bad faith trolling or people making stuff up and getting aggressive when their walls of text and stitched up quotes are not taken as proof and their truth accepted as gospel. If. Sony “flat earther” came about and I read it, discussed, and engaged something similar would ensue. Contrary to some’s belief I do not scour the forum to pick shit up...

 

[FranXico]

This always happens. There's a distinct double-standard applied to any claims by Sony and MS when it comes to the next-gen consoles. All claims from Sony (specifically Cerny) are to be assumed true until proven otherwise. All claims from Xbox engineers are to be assumed false until proven otherwise.

I also see double standards, but in the direct opposite direction. Even recently, a thread was made claiming that the PS5 actually still has a RDNA1 GPU. The "but it's not really 10TF, it's 9!" meme lives on. And let's not forget how every day new posts show up trying to prove that the UE5 demo can run on slow laptops with the same settings and at a higher framerate than a PS5.
Cerny and anyone who works with Sony is branded a liar, while every marketing acronym from Microsoft is repeated ad nauseum without any level of critical thinking or understanding.

 

[semicool]

I also see double standards, but in the direct opposite direction. Even recently, a thread was made claiming that the PS5 actually still has a RDNA1 GPU. The "but it's not really 10TF, it's 9!" meme lives on. And let's not forget how every day new posts show up trying to prove that the UE5 demo can run on slow laptops with the same settings and at a higher framerate than a PS5.
Cerny and anyone who works with Sony is branded a liar, while every marketing acronym from Microsoft is repeated ad nauseum without any level of critical thinking or understanding.

This is an Xbox Velocity Architecture thread, please don't derail it with an unrelated Sony TFs discussion. The Sony SSD could be a counterpart discussion but certainly not the TFs of the GPU in the unrelated sense you portrayed it.