Support NeoGAF

[Riky]

Keep going in circles and move from point to point, I agree it is fun for you to think about others wasting their time thinking they are engaging in an actual not disingenuous discussion.

I mentioned BC and cross generation games again as you said my remarks on it belittled XVA hoping by trying to be even clearer you would read it and understand it but .
You are belittling XVA and MS approach to BC under appreciating how difficult it is to improve performance meaningfully without compromising your peak potential and without requiring developers to invest a lot of time in changing their code for you… which is ironically delicious.

We agree that MS could have pushed for an even faster SSD solution and decided not to. They had a good enough solutions and wanted to fight other battles / invest their R&D or console BoM (look at it however way you want) in other parts of the design.
The problem is that you still see the XSX as this unlimited budget design and cannot accept any engineering compromise and anything that might not feel like the PS5 trouncing monster you thought you had been promised.
So there comes the stretching of the SFS advantage and the console war features list war… I noticed now your brought RDNA2 advantages (next post maybe you can add some bold “full” to it as people may not get one console has RDNA 1.5 apparently ).

I haven't even mentioned PS5, that's all in your console warrior mindset, I was talking about how MS have approached the problem of moving data and saying that we haven't seen the full potential until we see games using all the features with hardware support including SFS which is a cornerstone of XVA. You start talking about something else entirely for some reason.

 

[Allandor]

Oodles Textures and Kraken are compression formats. And in the PS5s case it can dramatically decrease the install size of a game. And also the patch sizes as well.

Not really. As compression improves, higher res textures more than equalize that. The only thing that reduces game sizes so far is the packaging process. To make it simple: Instead of big archives that contain many files (and different archives contain the same files) the SSD allows to use small packets (or even files directly). This dramatically decreases game sizes and also patch sizes, as patches no longer need to deliver the big archives. Instead individual files (or small archives) can be patched.

The other thing that seem to reduce game sizes on PS5 is, that the "Other" part on your SSD increases with different games. My guess so far, there are the versioned files from the SDK needed to run the game. So not every game needs to deliver its own SDK-file package. Therefore the "Other"-data on your PS5s SSD contains shared SDK data for the installed games. At least that is the only reason that comes to my mind why the size is very variable and changes with the combination of the installed games.

 

[FranXico]

Block diagrams.

To show how things work under the hood in modern games.

Jesus.

Timing diagrams, state flow diagrams, class diagrams, etc.

"Block diagrams" real developers use in their work.

 

[MasterCornholio]

They indeed are fantastic systems and I feel more unique compared to each other than Xbox One and PS4 actually weirdly enough.

Neither system is superior the other in every single way. It's why it's so fascinating to discuss the strengths and weaknesses of each one.

Still trying to nab myself a PS5 though.

 

[MasterCornholio]

Not really. As compression improves, higher res textures more than equalize that. The only thing that reduces game sizes so far is the packaging process. To make it simple: Instead of big archives that contain many files (and different archives contain the same files) the SSD allows to use small packets (or even files directly). This dramatically decreases game sizes and also patch sizes, as patches no longer need to deliver the big archives. Instead individual files (or small archives) can be patched.

The other thing that seem to reduce game sizes on PS5 is, that the "Other" part on your SSD increases with different games. My guess so far, there are the versioned files from the SDK needed to run the game. So not every game needs to deliver its own SDK-file package. Therefore the "Other"-data on your PS5s SSD contains shared SDK data for the installed games. At least that is the only reason that comes to my mind why the size is very variable and changes with the combination of the installed games.

But not all compression techniques are the same. Some do can compress more than others which can help explain the reduction in file size. Might not be the only thing but it certainly is a factor.

 

[Panajev2001a]

I haven't even mentioned PS5, that's all in your console warrior mindset, I was talking about how MS have approached the problem of moving data and saying that we haven't seen the full potential until we see games using all the features with hardware support including SFS which is a cornerstone of XVA. You start talking about something else entirely for some reason.

I like the “technically… but wink wink nudge nudge” approach here… sure, it is the others with a console warring mindset, just gotta accuse them of it first and make it stick.

All of this because your “3x the efficiency vs anything that does not have SFS (wink wink… let’s forget about PRT/Tiled Resources based streaming)” was challenged. Not a new argument just cyclical regurgitation of it.

 

[Lethal01]

There won't be, Microsoft just attacked the same issue in a different way with hardware support for Sampler Feedback Streaming, Intel just did a presentation on it and it gives up to three times the performance.
It's going to be a game changer for Xbox consoles.

Three times compared to what? I remember seeing them talk about it being three times better than a system that's not using partially residential textures but the didn't specify how got it is versus one that does use that.

Can't really say that SFS is a big advantage over PS5 with PRT for now.

Could you link me the talk? maybe there is a more recent one I missed.

 

[Allandor]

But not all compression techniques are the same. Some do can compress more than others which can help explain the reduction in file size. Might not be the only thing but it certainly is a factor.

As I already wrote, the new texture sizes will more than compensate the better compression
Textures and PS4 were also compressed with data- and texture-compression. There is not so much to gain here as you will also increase texture resolution or do you want PS4 games just running on PS5 with the same assets?.
Another thing that might also reduce game-size is reduction of lower-res assets. As you no longer need some low-quality textures for the new system, those can also be left (e.g. you might not need every mip-level of a texture etc).

 

[Panajev2001a]

As I already wrote, the new texture sizes will more than compensate the better compression
Textures and PS4 were also compressed with data- and texture-compression. There is not so much to gain here as you will also increase texture resolution or do you want PS4 games just running on PS5 with the same assets?.
Another thing that might also reduce game-size is reduction of lower-res assets. As you no longer need some low-quality textures for the new system, those can also be left (e.g. you might not need every mip-level of a texture etc).

True, but I think that adding Oodle Textures + Kraken (which is already an improvement over zlib) may even in some cases get you smaller on disk and yet higher quality assets (see FF VII integrade thread and the comments by S-E).

As well as data redundancy to reduce latency when dealing with mechanics drives, older consoles had also less CPU and GPU grunt to dedicate to additional levels of texture and asset compression (even on PS5 you can apply further optimisations to textures that you need to decode on the CPU or GPU with async compute shaders, BC7PREP). Still, devs will find ways to stuff even more data and we will still get large games .

 

[MasterCornholio]

As I already wrote, the new texture sizes will more than compensate the better compression
Textures and PS4 were also compressed with data- and texture-compression. There is not so much to gain here as you will also increase texture resolution or do you want PS4 games just running on PS5 with the same assets?.
Another thing that might also reduce game-size is reduction of lower-res assets. As you no longer need some low-quality textures for the new system, those can also be left (e.g. you might not need every mip-level of a texture etc).

Well that's assuming that both systems are not using the same textures. When they are the platform with better compression techniques will have a smaller file size.

 

[Riky]

Three times compared to what? I remember seeing them talk about it being three times better than a system that's not using partially residential textures but the didn't specify how got it is versus one that does use that.

Can't really say that SFS is a big advantage over PS5 with PRT for now.

Could you link me the talk? maybe there is a more recent one I missed.

Compared to the software emulated version if you don't have hardware support. Intel GDC.

 

[Lethal01]

Compared to the software emulated version if you don't have hardware support. Intel GDC.

Thanks a lot this is really great and Solid info.

So worst-case scenario is that PS5 can get all the benefits when it comes to streaming but with SFS taking 3x longer to compute. Now if we just knew how many ms it takes we'd be getting a full picture of the advantages of the dedicated hardware.

 

[Panajev2001a]

Compared to the software emulated version if you don't have hardware support. Intel GDC.

Ah, interesting so you are taking a benchmark measuring sampler feedback performance and comparing the speed of the HW implementing it vs a fully software based solution (letting slide the overall contribution to the render time in this synthetic benchmark was 23%) and using it to imply a 3x efficiency (I/O throughput) improvement with Sampler Feedback Streaming (SFS)… as you were using this in relation to the SSD speed differential.

Ok, so you were comparing apples to oranges.

Video in full:

Includes the benchmark commissioned to analyse the impact of SF in their texture space shading demo.

 

[Panajev2001a]

Thanks a lot this is really great and Solid info.

So worst-case scenario is that PS5 can get all the benefits when it comes to streaming but with SFS taking 3x longer to compute. Now if we just knew how many ms it takes we'd be getting a full picture of the advantages of the dedicated hardware.

Partially, this is result on Intel’s own HW in a presentation, we know what they see overall… 23% improvement to rendering time in this benchmark. It is by no means small.

This has little to do with SSD I/O throughput (although this can offset some of the efficiency gains a large SRAM cache in PS5’s I/O complex and the DRAM cache for the SSD controller bring in terms of latency reduction… on the I/O path we are talking about the efficiency of the SF feature vs a lot of dedicated HW/CPU’s in the I/O complex … we do not have the data on PS5 CPU tax for I/O but it is not unreasonable to think it is below the percentage MS quoted for one of the Zen 2 cores used to process I/O transfers) overall but the fact that SFS is an improvement over PRT was not the bit questioned.

The video in full (well it seems cut for brevitiy in some parts or just badly edited):
Includes the benchmark commissioned to analyse the impact of SF in their texture space shading demo.

The other interesting part is that we are in a PS5 SSD thread talking now about a DX12 API/feature in the context of rendering performance (as if PS5 implemented DX12 and we compared feature for feature across different API’s too).

 

[Lethal01]

Partially, this is result on Intel’s own HW in a presentation, we know what they see overall… 23% improvement to rendering time in this benchmark. It is by no means small.

This has little to do with SSD I/O throughput (although this can offset some of the efficiency gains a large SRAM cache in PS5’s I/O complex and the DRAM cache for the SSD controller bring in terms of latency reduction… on the I/O path we are talking about the efficiency of the SF feature vs a lot of dedicated HW/CPU’s in the I/O complex … we do not have the data on PS5 CPU tax for I/O but it is not unreasonable to think it is below the percentage MS quoted for one of the Zen 2 cores used to process I/O transfers) overall but the fact that SFS is an improvement over PRT was not the bit questioned.

The video in full (well it seems cut for brevitiy in some parts or just badly edited):
Includes the benchmark commissioned to analyse the impact of SF in their texture space shading demo.

I was overstimplifying, we still need more info
the 23% benefit when using sampler feedback doesn't mean that the pass itself takes longer you only get a 6.5% benefit. But it does give a single hard number about how much faster doing it a certain piece of hardware makes it.

That info could be useful in the future.

 

[Snake29]

Slower SSD works, doesn't mean you won't have problems with it in specific games. The fact that the first party games didn't fully utilize the SSD is only great.

It would be weird recommend slower drives if the internal is faster......

 

[Panajev2001a]

I was overstimplifying, we still need more info
the 23% benefit when using sampler feedback doesn't mean that the pass itself takes longer you only get a 6.5% benefit. But it does give a single hard number about how much faster doing it a certain piece of hardware makes it.

That info could be useful in the future.

Indeed, it was an interesting presentation overall mixing the Pr and technical bits very well. Hence why I wanted to post the full video.

 

[Knightime_X]

None of those games are using engines like unreal engine 5.
It's very likely that the reason even the slowest drive runs well is because there is nothing there that requires that type of speed to begin with.

Let's see how these perform when things like ue5 nanite come into play.

 

[Md Ray]

You know every PS5 comes with a 5.5GB/s drive right?

There won't be, Microsoft just attacked the same issue in a different way with hardware support for Sampler Feedback Streaming, Intel just did a presentation on it and it gives up to three times the performance.
It's going to be a game changer for Xbox consoles.

So when's the Series I/O going to surpass the PS5s?

It doesn't need to, instead of moving lots of data very quickly SFS means you just move a lot less data for the same result, it's fundamental part of VA.

Sounds like what the PS5 is doing.

Xbox Glossary

Exactly, if you try and understand the idea behind SFS then you'll come to know this feature is principally there to avoid loading unwanted data into the RAM. Doing so wastes precious system memory space which is something you needed to do on last-gen consoles due to slow storage (HDD).

Here Mark Cerny explains it better:

PS4 | Contains data for the next 30 seconds of gameplay. Lots of inactive RAM

There's yet one more benefit which is that system memory can be used much more efficiently. On PlayStation 4, game data on the HDD feels very distant and difficult to use. By the time you realize you need a piece of data, it's much too late to go out and load it. So system memory has to contain all of the data that could be used in the next 30 seconds or so of gameplay. That means a lot of the 8GB of system memory is idle, it's just waiting there to be potentially used.

Now we get to next-gen paradigm - Cerny's explanation here suggests this is PS5's version of Sampler Feedback Streaming:

PS5 | Contains data for the next 1 second of gameplay. Most RAM in active use

On PlayStation 5 though, the SSD is very close to being like more RAM. Typically it's fast enough that when you realize you need a piece of data you can just load it from the SSD and use it. There's no need to have lots of data parked in system memory waiting to potentially be used. A different way of saying that is that most of RAM is working on the game's behalf. This is one of the reasons that 16GB of GDDR6 feels right for PS5. The presence of the SSD reduces the need for a massive intergenerational increase in size.

Notice, with next-gen paradigm, they mention that RAM contains data for the next 1 second of gameplay? Meaning PS5 is capable of allowing a constant flow of useful data from SDD -> RAM every second or so. This means you can use only parts of the textures the GPU requires for a scene when it needs to as opposed to loading everything.

Tl;dr:
Basically SFS on PS5.

 

[MasterCornholio]

Xbox Glossary

Exactly, if you try and understand the idea behind SFS then you'll come to know this feature is principally there to avoid loading unwanted data into the RAM. Doing so wastes precious system memory space which is something you would do on last-gen consoles due to slow storage (HDD).

Here Mark Cerny explains it better:

PS4 | Contains data for the next 30 seconds of gameplay. Lots of inactive RAM



Now we get to next-gen paradigm - Cerny's explanation here suggests this is PS5's version of Sampler Feedback Streaming:

PS5 | Contains data for the next 1 second of gameplay. Most RAM in active use


Notice, with next-gen paradigm, they mention that RAM contains data for the next 1 second of gameplay? Meaning PS5 is capable of allowing a constant flow of useful data from SDD -> RAM every second or so. This means you can use only parts of the textures the GPU requires for a scene when it needs to as opposed to loading everything.

Basically SFS on PS5.

Well from my understanding the PS5 also has some capabilities similar to SFS. Developers can choose to render a portion of the textures needed. It's not just limited to the next second of gameplay but it goes beyond just doing that.

 

[Panajev2001a]

Xbox Glossary

Exactly, if you try and understand the idea behind SFS then you'll come to know this feature is principally there to avoid loading unwanted data into the RAM. Doing so wastes precious system memory space which is something you would do on last-gen consoles due to slow storage (HDD).

Here Mark Cerny explains it better:

PS4 | Contains data for the next 30 seconds of gameplay. Lots of inactive RAM



Now we get to next-gen paradigm - Cerny's explanation here suggests this is PS5's version of Sampler Feedback Streaming:

PS5 | Contains data for the next 1 second of gameplay. Most RAM in active use


Notice, with next-gen paradigm, they mention that RAM contains data for the next 1 second of gameplay? Meaning PS5 is capable of allowing a constant flow of useful data from SDD -> RAM every second or so. This means you can use only parts of the textures the GPU requires for a scene when it needs to as opposed to loading everything.

Basically SFS on PS5.

I think we are comparing apples to oranges a bit: Cerny is basically saying that thanks to the faster storage they use main RAM less wastefully as they need smaller buffers to hide external storage latency.
Reducing the amount of memory used to store ahead of time data, to hide the high latency and low throughput of HDD’s and old file I/O API’s (which meant that even super fast storage solutions would have destroyed your CPU on overhead alone), does bring savings to both systems.
The higher the throughput and the lower the latency the less RAM you waste to store data not needed for the current scene or the current frame.

PRT and SFS benefit from that (much harder to take advantage of if you feed yourself from a super slow external storage system), but are a bit of a different beast: it allows you to implement granular texture streaming (with GPU support) so that we manage and load in and out of GPU memory only the portions of a texture that the GPU actually needs (not entire textures chains or entire mip levels). For this to work well you need to be able to have a very fast and low latency access to what you consider external to GPU memory (the SSD in this case).

People had streaming engines before in the older HDD and DVD days but the slow speed of external storage meant that your movement in the world for example was constrained and memory had to be wasted holding larger streaming buffers. Think PS2 and the following chain: DVD —> Direct RDRAM —> GS’s e-DRAM. GS could only texture from its eDRAM, but programmers would manually stream new texture data in as needed from main RAM and also update the main RAM texture data with new data from the DVD (storage space <-, memory speed -> in that order in the chain).

SFS takes the idea of tiled resources/partial resident textures where the system is able to keep only certain portions of a texture in GPU memory and help you page data in and out (you would use this to accelerate your virtual texturing implementation as the HW helps keep track of which portion of the texture is where).
Beyond PRT, Samplers Feedback allows you to read from the GPU directly what portion of the overall texture is being sampled and helps you predict what data will need to be loaded next (while doing some additional bookkeeping for you): SFS adds instructions to help with texture streaming including one to blend texture data from different mip levels (different levels of detail) without showing artefacts.

MS tech docs describe (in depth coder friendly docs) Sampler Feedback applied to texture streaming in full here: https://microsoft.github.io/DirectX-Specs/d3d/SamplerFeedback.html
SFS is defined as PRT+

 

[Andodalf]

None of those games are using engines like unreal engine 5.
It's very likely that the reason even the slowest drive runs well is because there is nothing there that requires that type of speed to begin with.

Let's see how these perform when things like ue5 nanite come into play.

We’ve already seen ue5 run well with far, far slower drives

 

[Blond]

Head in the sand as usual. You don't have a "hypothesis", you just don't seem willing to understand the basics of the technology.

It's been said over and over again, it's not the SSD speed itself in the PS5 that is the game changer, it's the IO/decompression path. And if you've actually played R&C, it's impossible to deny.

I really wonder how many years this discussion will persist with people either refusing to understand or purposefully misrepresenting what the PS5 is actually doing in this area.

I played R and C and I disagree. 99% of the gameplay that was promised in trailer’s wasn’t there in anything but Uncharted style set pieces.

 

[Knightime_X]

We’ve already seen ue5 run well with far, far slower drives

What games on the market are using us5 right now available for purchase?

 

[Md Ray]

I think we are comparing apples to oranges a bit: Cerny is basically saying that thanks to the faster storage they use main RAM less waste fully as they have smaller buffers to hide external storage latency.
Reducing the amount of memory used to store ahead of time data, to hide the high latency and low throughput of HDD’s and old file I/O API’s (which meant that even super fast storage solutions would have destroyed your CPU on overhead alone), does bring savings to both systems.
The higher the throughput and the lower the latency the less RAM you waste to store data not needed for the current scene or the current frame.

PRT and SFS benefit from that (much harder to take advantage of if you feed yourself from a super slow external storage system), but are a bit of a different beast: it allows you to implement granular texture streaming (with GPU support) so that we manage and load in and out of GPU memory only the portions of a texture that the GPU actually needs (not entire textures chains or entire mip levels). For this to work well you need to be able to have a very fast and low latency access to what you consider external to GPU memory (the SSD in this case).

People had streaming engines before in the older HDD and DVD days but the slow speed of external storage meant that your movement in the world for example was constrained and memory had to be wasted holding larger streaming buffers. Think PS2 and the following chain: DVD —> Direct RDRAM —> GS’s e-DRAM. GS could only texture from its eDRAM, but programmers would manually stream new texture data in as needed from main RAM and also update the main RAM texture data with new data from the DVD (storage space <-, memory speed -> in that order in the chain).

SFS takes the idea of tiled resources/partial resident textures where the system is able to keep only certain portions of a texture in GPU memory and help you page data in and out (you would use this to accelerate your virtual texturing implementation as the HW helps keep track of which portion of the texture is where).
Beyond PRT, Samplers Feedback allows you to read from the GPU directly what portion of the overall texture is being sampled and helps you predict what data will need to be loaded next (while doing some additional bookkeeping for you): SFS adds instructions to help with texture streaming including one to blend texture data from different mip levels (different levels of detail) without showing artefacts.

MS tech docs describe (in depth coder friendly docs) Sampler Feedback applied to texture streaming in full here: https://microsoft.github.io/DirectX-Specs/d3d/SamplerFeedback.html
SFS is defined as PRT+

Awesome, informative post, as usual, man. I need to read up on PRT stuff more.

 

[graywolf323]

NASCompares tried a Gen3 NVME and the Gen4 FireCuda 520 which while being below the recommended speed actually worked at least right now during the beta (though he doesn't recommend it because there's a decent chance that when the firmware releases to the public that Sony could block drives that speed test as too slow)

the Gen3 NVME Samsung Evo+ 970 did not work at all

 

[MasterCornholio]

NASCompares tried some Gen3 NVMEs and the FireCuda 520 actually worked at least right now during the beta

the Samsung Evo+ 970 did not work at all

Gen3

 

[Md Ray]

NASCompares tried some Gen3 NVMEs and the FireCuda 520 actually worked at least right now during the beta

the Samsung Evo+ 970 did not work at all

The FireCuda 520 is Gen4 though.

 

[phil_t98]

What games on the market are using us5 right now available for purchase?

fortnite will be using it from next month on all versions

 

[graywolf323]

The FireCuda 520 is Gen4 though.

oops yeah you're right, I misphrased it, I went ahead & fixed my post

the 520 was just because it's under the recommended speeds while still being Gen4

 

[Knightime_X]

fortnite will be using it from next month on all versions

Will fortnite use nanite?
A lot of games will start using ue5 but how many will take full advantage of the latest abilities?

 

[ethomaz]

Every comparison I've seen show that using a 7000 MB/s SDD on the PS5 makes games load faster than using PS5's internal storage, which proves than even its super fast SDD is the bottleneck in PS5's I/O system and it's capable of taking advantage of even higher speeds.




I guess it's the same the other way around: if you use a slower SDD than PS5's internal storage games and asset streaming will work a bit slower, as Insomniac said and you'd expect.

The opposite every single test until now shows little to no gain in loading times with faster SSD.

That means either PS5 games are not stressing 5.5GB/s or the I/O Complex is limited to 5.5GB/s.

 

[ethomaz]

NASCompares tried a Gen3 NVME and the Gen4 FireCuda 520 which while being below the recommended speed actually worked at least right now during the beta (though he doesn't recommend it because there's a decent chance that when the firmware releases to the public that Sony could block drives that speed test as too slow)

the Gen3 NVME Samsung Evo+ 970 did not work at all

Because that Firecuda 530 (they tested that one… not 520) is PCI-E 4.0.

You can read the article too:

Seagate Firecuda 530 PS5 SSD Expansion Test

Testing the Seagate Firecuda 530 SSD on the PS5 Time for another PS5 SSD Expansion Test and this time it is one of the hottest pick for the best M.2 NVMe upgrade drives, the Seagate Firecuda 530. Now that Sony has enabled the ability to expand the storage of the Playstation 5 in the latest software

nascompares.com

 

[graywolf323]

Because that Firecuda 530 (they tested that one… not 520) is PCI-E 4.0.

You can read the article too:

Seagate Firecuda 530 PS5 SSD Expansion Test

Testing the Seagate Firecuda 530 SSD on the PS5 Time for another PS5 SSD Expansion Test and this time it is one of the hottest pick for the best M.2 NVMe upgrade drives, the Seagate Firecuda 530. Now that Sony has enabled the ability to expand the storage of the Playstation 5 in the latest software

nascompares.com

yeah they tested the 530 before, but in the video I linked to they test the 520

 

[phil_t98]

Will fortnite use nanite?
A lot of games will start using ue5 but how many will take full advantage of the latest abilities?

Well seeing that fortnite is epics own game and one of the biggest on the planet I suspect they will try to show UE5 off as much as possible.

 

[The_Mike]

According to Mike you should expect slower results from it. That's if you go with one that's below the recommended specs BTW.

But some comparisons in here shows that the internal ssd ain't that fast as cerny and many others in here have stated.

even RC works perfectly on 'slow' ssd, pretty much proves my hypothesis about the ssd i/o hype that never really was.

going from hdd to nvme is a huge jump and that's that

big numbers big multiplier, but the real bottleneck is not IO for next gen games

Weren't we told ratchet and clank couldn't even be played on other hardware because of the i/o?

 

[TheDreadLord]

Currently, it seems that the PS5 internal storage spec is overdesigned. I think Sony should clarify this as soon as possible. It smells like "the power of the Cell CPU" all over again.

 

[twilo99]

Weren't we told ratchet and clank couldn't even be played on other hardware because of the i/o?

based on testing so far, it seems like it would play just fine on a series s

 

[The_Mike]

based on testing so far, it seems like it would play just fine on a series s

Said that since the beginning, but was flamed a lot for it in here.

 

[ZywyPL]

I think people are forgetting that R&C loads assets in&out on the fly, as you rotate the camera, which is an extremely demanding task since you have literally single milliseconds to do it, if not less. So if the slowest available drive on the market wouldn't be up to the task it would've been immediately visible on the screen with missing textures or entire objects, no benchmarks are needed for that.

Currently, it seems that the PS5 internal storage spec is overdesigned. I think Sony should clarify this as soon as possible. It smells like "the power of the Cell CPU" all over again.

I think the PS5 reveal would've been vastly different if MS didn't show its cards first, but it is what it is, Sony had to switch the narrative to highlight their strong points like the SSD and Tempest audio, but the very moment Cerny said "it's dangerous to rely on TFlops and CU count should be avoided as well" you knew he's doing the damage control, instead of just comparing PS5 to their previous console like they always used to, saying it has almost 6x the GPU power then PS4, before accounting any gains from architecture advancements, 4x the CPU power, 3x the usable RAM etc. But needless to say, they did spin the narrative very effectively.

 

[onesvenus]

I don't think it has been posted but this Twitter thread from Fabian Giesen, one of the creators of Oodle textures and Kraken, is really interesting and on topic:



Basically he states that almost everyone who claims to be IO bound is CPU bound instead. This points to two interesting things: the SSD speed not mattering as much as some believe and the secret sauce on PS5's I/O complex being the descompressor hardware.
With that in mind, it makes sense to believe that variation in SSD speeds won't affect as much as it would not having the descompressor hardware.

 

[skit_data]

I don't think it has been posted but this Twitter thread from Fabian Giesen, one of the creators of Oodle textures and Kraken, is really interesting and on topic:



Basically he states that almost everyone who claims to be IO bound is CPU bound instead. This points to two interesting things: the SSD speed not mattering as much as some believe and the secret sauce on PS5's I/O complex being the descompressor hardware.
With that in mind, it makes sense to believe that variation in SSD speeds won't affect as much as it would not having the descompressor hardware.

Personally I have always been under the assumption that the dedicated decompressor is the bigger deal rather than the speed of the SSD. Thats where the actually meaningful multiplier lies to get to the end result in data throughput.

Edit: and of course performance gains thanks to the CPU being offloaded to do more meaningful work

 

[reksveks]

I think people are forgetting that R&C loads assets in&out on the fly, as you rotate the camera, which is an extremely demanding task since you have literally single milliseconds to do it, if not less. So if the slowest available drive on the market wouldn't be up to the task it would've been immediately visible on the screen with missing textures or entire objects, no benchmarks are needed for that.

Do we have a recording of someone just spinning the camera around constantly on R&C?

 

[Crimson_Fate]

Assuming that the Verge tested things properly using a sup-spec drive with read speeds of 3900 (71% of the recomended 5500) that drive is still 63% faster then the drive in the XSX or S (ignoring compression). I assume we will have some better indepth testing from DF or NXG but regardless , nothing shown here states that R&C was possible on a GEN 3 Drive.

 

[martino]

I don't think it has been posted but this Twitter thread from Fabian Giesen, one of the creators of Oodle textures and Kraken, is really interesting and on topic:



Basically he states that almost everyone who claims to be IO bound is CPU bound instead. uce on PS5's I/O complex being the descompressor hardware.This points to two interesting things: the SSD speed not mattering as much as some believe and the secret sa
With that in mind, it makes sense to believe that variation in SSD speeds won't affect as much as it would not having the descompressor hardware.

add to this on the i/o front all solution so far target 100x over previous gen.
but what will happen when they use the new render pipeline. it's a long road but it should lead to more less draw calls.

I think people are forgetting that R&C loads assets in&out on the fly, as you rotate the camera, which is an extremely demanding task since you have literally single milliseconds to do it, if not less. So if the slowest available drive on the market wouldn't be up to the task it would've been immediately visible on the screen with missing textures or entire objects, no benchmarks are needed for that.

 

[Werewolfgrandma]

I think people are forgetting that R&C loads assets in&out on the fly, as you rotate the camera, which is an extremely demanding task since you have literally single milliseconds to do it, if not less. So if the slowest available drive on the market wouldn't be up to the task it would've been immediately visible on the screen with missing textures or entire objects, no benchmarks are needed for that.

Um I don't think it does. At least there is no actual confirmation of that being the case.

Assuming that the Verge tested things properly using a sup-spec drive with read speeds of 3900 (71% of the recomended 5500) that drive is still 63% faster then the drive in the XSX or S (ignoring compression). I assume we will have some better indepth testing from DF or NXG but regardless , nothing shown here states that R&C was possible on a GEN 3 Drive.

Gen 3 Max speed is 3.9gbs, so the same speed that was tested. Nothing shown proves Ratchet isn't possible on a 2.4gbs drive. Obviously nothing proves it does either, but there is zero confirmation of anything really.

 

[ZywyPL]

Um I don't think it does. At least there is no actual confirmation of that being the case.

It's been confirmed by Insomniac's community manager via Twitter:

Ratchet and Clank loads only what you can see in the moment

Mark Cerny talked about it, Unreal 5 demo’d it and now Insomniac have managed to put it into a full game. This is what will seperate current gen from the last, no cross-gen games will be able to achieve this feat, this is an exclusive only feature and we’re only at the beginning. With this...

www.neogaf.com

 

[Werewolfgrandma]

It's been confirmed by Insomniac's community manager via Twitter:

Ratchet and Clank loads only what you can see in the moment

Mark Cerny talked about it, Unreal 5 demo’d it and now Insomniac have managed to put it into a full game. This is what will seperate current gen from the last, no cross-gen games will be able to achieve this feat, this is an exclusive only feature and we’re only at the beginning. With this...

www.neogaf.com

From the same interview.
"Spider-Man is an open-world title. We built all of this tech to stream that open world as you go through it. When you’re downtown, there’s not much Midtown in memory. You can see it from a distance, but then as you go farther north, we pull in those areas. No Ratchet game has ever been constructed that way. They’ve always been: here’s a level, load the level, now you’re in that level and you play it. But by switching over the Ratchet world to use that same streaming architecture, we can pack more and more density and content and quality in every corner of a Ratchet & Clank world, because we’re happy to ditch the west side of Nefarious City when you go to the east side, and that type of thing."

I could be wrong, but this is saying it's loaded in zones. Not on a frame by frame basis.

 

[CuteFaceJay]

youdontsay.jpg

Any modern SSD should be good enough.

 

[Lysandros]

But some comparisons in here shows that the internal ssd ain't that fast as cerny and many others in here have stated.



Weren't we told ratchet and clank couldn't even be played on other hardware because of the i/o?

I/O... PS5's processor heavy I/O complex isn't its SSD...