• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

PS5 Load Time Hype - This is an SSD Thread

ZywyPL

Banned
825GB of the SSD storage is not enough, 100GB will be taken away for system functions which only leaves you with 725GB.

I want to buy 2TB+ of SSD storage, but with the same bells and whistles as the internal one (especially with the I/O). Anyone have any leads on this? It doesn't matter how fast the extra storage is, because the bottleneck will be the I/O. So what external SSD storage can give:

-2TB+ of storage
-Fast or faster speeds of compressed/uncompressed data compared to the internal SSD storage
-Getting rid of the bottleneck I/O

Muchas Gracias.


Sabrent Rocket 4 Plus is your safest bet right now - 2TB space + 7Gbps at both read and write operations. But that will be around 500$.
 

Rikkori

Member
Nope, only fanboys would deny what incredible job Sony has done with their I/O. UE5 demo already proved it's true. Streaming billions of triangles and 8k assets on the fly. Those assets won't fit in the RAM.

Sony's solution can deliver data from SSD straight to GPU caches due to their coherency engines and cache scrubbers as opposed to sending the data first to RAM to perform several steps to the data.

https://d2skuhm0vrry40.cloudfront.n....jpg/EG11/resize/337x-1/quality/75/format/jpg

Again, there's a reason developers are over the moon over Sony's SSD solution.
Like I said - fanboy delusions in action.
Go do the math on PCIe x4 speeds for nvme vs GPU bandwidth then come back when you learn why what you said is total nonsense. And btw - if you supposedly can't hold these things in vram, then how the hell are you going to get them unto the GPU to render? This is graphics rendering 101. Stop drinking marketing kool aid and go learn the basics.

PS: the tech you're harping on about had been demonstrated on much lower SSDs on PC before you even knew the PS logo because it has very little to do with the ssd as much as it does with the GPU.
PPS: that's trillions you're seeing, not puny billions.

 
Last edited:
You're not made for Soulsborne games if you haven't withstood initial unpatched Bloodborne loading times.



First of all, congrats on attributing a industry wide move to just Sony because they merely announced it first? Let's conveniently forget Microsofts Velocity Architecture or NVIDIAs RTX IO.

Cerny built 6 custom chips in the APU which includes embedded Ram to address all the bottlenecks in the I/O. It's a wholistic solution to I/O problem that had developers taken aback on how fast it is. That's why we keep on hearing 2 seconds of loading with no loading screens.

MS built 1 (one) asic chip in the xbox apu which is the decompression chip, and it's not even half as fast the decompression chip in the PS5. Then MS added a bunch of software solution like that DirectStorage that will come to PC in 2021 in beta form with official release in 2022.

The move to solve I/O was industry-wide but it was only Sony who went out of their way to provide something unexpected and built the slew of chips addressing all the bottlenecks.

That's why we see again and again the 11-12 seconds loading of Xsx while devs have been telling us again and again the 2 seconds loading of PS5.

It's not just pure bandwidth advantage. It's the whole slew of customization that Cerny did in the I/O.
 
Wonder what Capcom will do do for the doors in Resident Evil 2 The Remake of the Remake.

Personally I upgraded to SSD a few months ago in my PC. Meh. Improvements, but not by much.

 
Like I said - fanboy delusions in action.
Go do the math on PCIe x4 speeds for nvme vs GPU bandwidth then come back when you learn why what you said is total nonsense.

LOL. Another fanboy with a straw man argument. What about the Unreal Engine 5 demo on PS5 you don't understand?

It was streaming, again streaming, billions of triangles and 8k assets on the fly as the character moves in the game. The GPU does not write on the SSD. The data is streamed to GPU and RAM. The GPU reads and writes to the RAM, not the SSD.

What will the GPU read and render from the RAM if the data needed is in the storage? That's right, don't make a game that will give you that problem. But you won't have a game as texture rich and geometry rich as that UE5 demo on the PS5.
 
Last edited:

Rikkori

Member
LOL. Another fanboy with a straw man argument. What about the Unreal Engine 5 demo on PS5 you don't understand?

It was streaming, again streaming, billions of triangles and 8k assets on the fly as the character moves in the game. The GPU does not write on the SSD. The data is streamed to GPU and RAM. The GPU reads and writes to the RAM, not the SSD.

What will the GPU read and render from the RAM if the data needed is in the storage? That's right, don't make a game that will give you that problem. But you won't have a game as texture rich and geometry rich as that UE5 demo on the PS5.
You've ignored everything I said as counter-example. Cool. Welcome to my ignore list. :messenger_heart:

PS: trillions are higher than billions
 

IntentionalPun

Ask me about my wife's perfect butthole
SSD is what I'm most excited for next-gen and the near complete removal of load times will have me gaming more on console.. and my console of choice for multi-plats will be PS5 because of it.
 

RaySoft

Member
Microsoft have gone out of their way to explicitly clarify that the 2,4GB/s is the minimum sustained speed at all times.
Have Sony or a developer clarified if the PS5's 5,5 GB/s SSD speed is the theoretical maximum or the sustained performance under load?
5.5 is minimum since that, as the same as 2.4GB/s for MS, is the raw bandwith (i.e. no compressed data)
With compression both of them will be even higher.
 

RaySoft

Member
Like I said - fanboy delusions in action.
Go do the math on PCIe x4 speeds for nvme vs GPU bandwidth then come back when you learn why what you said is total nonsense. And btw - if you supposedly can't hold these things in vram, then how the hell are you going to get them unto the GPU to render? This is graphics rendering 101. Stop drinking marketing kool aid and go learn the basics.

PS: the tech you're harping on about had been demonstrated on much lower SSDs on PC before you even knew the PS logo because it has very little to do with the ssd as much as it does with the GPU.
PPS: that's trillions you're seeing, not puny billions.



L2mGS4r.jpg


5GLVk41.jpg


Pretty self-explanatory.
Source: Road To PS5
 
Last edited:

Rikkori

Member
L2mGS4r.jpg


5GLVk41.jpg


Pretty self-explanatory.
Source: Road To PS5

Yes it is. Notice how it doesn't say any data for any s or ms of gameplay for the package, and how it says in some ways. Not Like RAM. Hint: that's why I'm saying nvme is for bandwidth purposes not utilisation. Again, graphics rendering 101. Should've been obvious by the fact that GDDR6 is a hundred times faster than the nvme, but I guess not.
 

RaySoft

Member
Yes it is. Notice how it doesn't say any data for any s or ms of gameplay for the package, and how it says in some ways. Not Like RAM. Hint: that's why I'm saying nvme is for bandwidth purposes not utilisation. Again, graphics rendering 101. Should've been obvious by the fact that GDDR6 is a hundred times faster than the nvme, but I guess not.
It's not like the SSD is just as fast as the GDDR6 RAM. That's not what ppl mean.
Since the SSD is so much faster than a spindle HDD, you can now use less RAM for asset storage (~1 sec of gametime worth vs 30sec.) You still need to reserve a ram pool for some assets, just not on the same level.
Hence you free up alot of ram to be used for other things.
 

Rikkori

Member
It's not like the SSD is just as fast as the GDDR6 RAM. That's not what ppl mean.
Since the SSD is so much faster than a spindle HDD, you can now use less RAM for asset storage (~1 sec of gametime worth vs 30sec.) You still need to reserve a ram pool for some assets, just not on the same level.
Hence you free up alot of ram to be used for other things.
Yes, but you have to look at the asset quality vs bandwidth/amount relation too. So with improved bandwidth the quality can improve somewhat but you can't escape the baseline requirements for the scene, and the data for a particular time of gameplay depends on the quality of the scene and then whatever the limitations of the GPU bandwidth & storage I/O are. No one is disputing the advantages of an SSD over a HDD, but that doesn't mean SSDs can now act as vram or is anywhere near GDDR6's (or even DDR4's) capabilities. It's a helpful additive but it's not a substitute.
 

RaySoft

Member
Yes, but you have to look at the asset quality vs bandwidth/amount relation too. So with improved bandwidth the quality can improve somewhat but you can't escape the baseline requirements for the scene, and the data for a particular time of gameplay depends on the quality of the scene and then whatever the limitations of the GPU bandwidth & storage I/O are. No one is disputing the advantages of an SSD over a HDD, but that doesn't mean SSDs can now act as vram or is anywhere near GDDR6's (or even DDR4's) capabilities. It's a helpful additive but it's not a substitute.
I'm straight up only talking about the old way of handling i/o (hdd) vs. new way of handling i/o (ssd)
I don't think ppl are comparing ssd bandwith with ram speeds really, except you.
 

Sony

Nintendo
5.5 is minimum since that, as the same as 2.4GB/s for MS, is the raw bandwith (i.e. no compressed data)
With compression both of them will be even higher.

In the PC world the stated SSD speeds are not sustained. SSD's get warmer and slower in use.
Where is it stated that PS5 SSD achieves 5.5 GB/s at all times/ under load? Did Sony confirm it or are you assuming it?
 

Rikkori

Member
I'm straight up only talking about the old way of handling i/o (hdd) vs. new way of handling i/o (ssd)
I don't think ppl are comparing ssd bandwith with ram speeds really, except you.
They are, that's exactly what I'm responding to.

In the PC world the stated SSD speeds are not sustained. SSD's get warmer and slower in use.
Where is it stated that PS5 SSD achieves 5.5 GB/s at all times/ under load? Did Sony confirm it or are you assuming it?
Exactly, and even those numbers aren't real in the first place because what's really gonna be hit are Q1T1 orders.
 

RaySoft

Member
In the PC world the stated SSD speeds are not sustained. SSD's get warmer and slower in use.
Where is it stated that PS5 SSD achieves 5.5 GB/s at all times/ under load? Did Sony confirm it or are you assuming it?
I know, they always state max numbers for they devices. Once you hook it up though, it becomes a part of a chain that's beholdent to the weakest link in said chain. Sony has worked hard to minimizing all latencies in every layer from software driver down to the hardware itself. No guarantee of the 5.5GB/s, but probably very close at worst.
 
They are, that's exactly what I'm responding to.

Since I was one of those you responded to. I was never implying that the SSD can be used as a Ram. That's stupid.

I was explaining how Sony's solution is a lot better than what MS did with theirs including the software solution they are providing for Nvidia. Sony's solution is wholistic which includes Sram, coherency engines, and cache scrubbers that will allow the SSD to stream directly to the GPU without going through the Ram which cuts down a lot of stall and latency. But of course the data after being read by the GPU will have to be written to the Ram.

We keep hearing from devs the 2 second loading time of the PS5 while what we have seen from XSX/S is 11-12 second loading times. It's not just the pure bandwidth but the whole slew of customized chips that makes this possible on the PS5.

On run-time and in terms of streaming of data, that latency advantage of the PS5 means that more Ram will be available on behalf of the game instead of keeping data that might or might not be used by the game. XSX/S and PC in the future despite the speed allowed by the SSD will still have to account for the latency which means those machines will still have to pre-load more data that are predicted to be used in the next few seconds. PS5 will have to pre-load less data not only becuase of raw speed of the SSD but because of the elimination of all the bottlenecks that causes latency.
 

GODbody

Member
PS: 8k textures. 😉
If you include 8K textures the game file size is going to basically double. That Unreal Demo was probably ~100 GBs and just a demonstration. Besides that only used a 768 GB streaming pool for textures and that is well within the range of most modern SSDs.


6mNa6YZ.png



L2mGS4r.jpg


5GLVk41.jpg


Pretty self-explanatory.
Source: Road To PS5
Alright so if 16GB of ram contained data for the next second of gameplay how large would the gameplay file on storage need to be for that to be meaningful? This is a goal not an actual. We likely won't see this 'next-Gen Paradigm' until we games in excess of 200GBs.

Since I was one of those you responded to. I was never implying that the SSD can be used as a Ram. That's stupid.

I was explaining how Sony's solution is a lot better than what MS did with theirs including the software solution they are providing for Nvidia. Sony's solution is wholistic which includes Sram, coherency engines, and cache scrubbers that will allow the SSD to stream directly to the GPU without going through the Ram which cuts down a lot of stall and latency. But of course the data after being read by the GPU will have to be written to the Ram.

We keep hearing from devs the 2 second loading time of the PS5 while what we have seen from XSX/S is 11-12 second loading times. It's not just the pure bandwidth but the whole slew of customized chips that makes this possible on the PS5.

On run-time and in terms of streaming of data, that latency advantage of the PS5 means that more Ram will be available on behalf of the game instead of keeping data that might or might not be used by the game. XSX/S and PC in the future despite the speed allowed by the SSD will still have to account for the latency which means those machines will still have to pre-load more data that are predicted to be used in the next few seconds. PS5 will have to pre-load less data not only becuase of raw speed of the SSD but because of the elimination of all the bottlenecks that causes latency.
PS5 streams data to system memory (RAM), not directly to the GPU.

qlmfVsC.jpg


Xbox is capable of streaming data directly to the GPU for use through direct storage a side effect of this is that it reduces CPU overhead as well as reduces the need for an I/O complex due to the miniscule impact on CPU resources.

PS5 has more raw bandwidth, but Xbox has built in efficiencies that reduce the need to transfer significant amounts of data. Such as Sampler Feedback Streaming which acts as a multiplier due to how reduced the data is (The baseline of which is PRT which is present on current gen, but Microsoft's implementation here is PRT+ and such an improvement that it's really not comparable.)

mFlKLM6.jpg


After that Xbox has other improvements such as DirectStorage, VRS, Mesh Shaders that further increase performance and reduce overhead.

The end result is that the I/O solutions of both end up just as capable as the other.

I/O is not that important though as neither system will be fully utilizing their massive transfer bandwidth in anything except bursts until games are allowed to be hundreds of GBs in size. It's unrealistic to fully use even a 2.4 GB/s transfer speed on a 100 GB game. The entire game could be read in 42 seconds. Sounds like a boring game right? That would be 20 seconds with a 5.5 GB/s speed. It's unrealistic to expect games to constantly be able to saturate that as there is just not enough data to fill 5.5 GB/s in anything other than short bursts. A game that constantly needed 5.5 GB/s would have a massive data footprint on storage.
 
Last edited:

Boglin

Member
They are, that's exactly what I'm responding to.

You're misunderstanding people.
If you have to store 30 seconds worth of assets in memory because of a slow HDD and only 1 second with the SSD then you freed up more memory for what your rendering RIGHT NOW.

The GPU can only render what's already in memory, but you have less available memory if it's tied up for assets you're not using yet, right?
I'm really surprised you're unable to understand this but since you've been pretty adamant that an SSD can't help then I'll break it down further for you.

If I have 10GB of memory for my GPU and 5GB of it is being reserved for future scenes that need rendered because my HDD is too slow to stream it in then I only have 5GB to render my current scene.

If I only need to reserve 1GB of memory because my SSD is fast enough to stream in more assets as needed, then I have 9GB for my current scene.

Surely you can agree that 9GB is better than 5GB of available memory for rendering the current scene, so you know what that means?

There ARE super magical SSD powers that could actually help graphics fidelity!

I know, it's crazy.
You probably thought all these stupid Sony fanboys actually believe that the SSD can actually render more, like it was some second GPU or something. At the same time you probably forgot that more GDDR6 memory doesn't render anything either yet it can still help your performance and it needs to get fed its data from somewhere.

Don't be so dense and assume people are dumb fanboys just because you can't grasp a simple concept that will benefit both consoles and PC moving forward.

I/O is not that important though as neither system will be fully utilizing their massive transfer bandwidth in anything except bursts until games are allowed to be hundreds of GBs in size. It's unrealistic to fully use even a 2.4 GB/s transfer speed on a 100 GB game. The entire game could be read in 42 seconds. Sounds like a boring game right? That would be 20 seconds with a 5.5 GB/s speed. It's unrealistic to expect games to constantly be able to saturate that as there is just not enough data to fill 5.5 GB/s in anything other than short bursts. A game that constantly needed 5.5 GB/s would have a massive data footprint on storage.

Games do not need to saturate the SSD bandwidth 100% percent of the time to be useful, just like not every scene needs to utilize the CPU or GPU at 100% all the time.
You're looking at the usefulness of this badwidth from too large a scale. Scenes don't have to be loaded at a per second basis and those short bursts you mentioned are all that's needed to be massively useful. When playing at 60FPS with the PS5, you could stream in over 90MB per frame if the latency is low enough.

It can be used for removing low quality LOD models in certain areas of the games, which saves storage space, and to avoid pop in or just load in something quickly on demand that is used occasionally.

For example in a first person shooter game you can switch weapons weapons on the fly. That gun model does not need to be kept in memory at all times anymore, just waiting for you to press the button, because now you can stream it in fast enough from storage.

Even on a larger scale the game would not necessarily need to be massive if it's loading assets that aren't entirely unique at every corner of the map. It could still take advantage of the memory speed by streaming in common assets.

A functional example would be a racing game where parts of the course and environment are streamed in as you're driving around.
Those assets can be reused throughout every course in the game and the same ones can even be unloaded and reloaded multiple times as needed within the same race as you're driving around.

Obviously not all games will benefit, but I just thought of these examples off the top of my head and they would be hindered by any speed reduction. The faster the SSD, the better!
 

LordOfChaos

Member
According to the "Road to PS5" video it will have an expansion bay for an M.2 SSD drive which I very much prefer over Microsoft's proprietary solution.

Do we know if it's pure expansion (825GB + X), or you'd be removing the previous NAND (not necessarily attached to the controller, see the Mac Pro/iMac Pro, the sticks are dumb NAND)?

I forget if this was just a dream I had lol but making sure it's not replaceable, rather than expandable. They used the word expansion, so unless it was very loosely used it should be.
 

Boglin

Member
Do we know if it's pure expansion (825GB + X), or you'd be removing the previous NAND (not necessarily attached to the controller, see the Mac Pro/iMac Pro, the sticks are dumb NAND)?

I forget if this was just a dream I had lol but making sure it's not replaceable, rather than expandable. They used the word expansion, so unless it was very loosely used it should be.

We don't know for sure but Mark Cerny has been fairly candid in the past so I think it's safe to assume 825GB + X. If Jim Ryan said it I would be less sure of things.
 

LordOfChaos

Member
C) having consumers who can barely boot windows by themselves rip open a ps5 is a bad idea

The PS4 and PS3 before it had replicable hard drives. Yeesh, it'll be fine, consumers aren't total chimps, and the least technically inclined will largely not even think of it having expandable storage.
 

RaySoft

Member
If you include 8K textures the game file size is going to basically double. That Unreal Demo was probably ~100 GBs and just a demonstration. Besides that only used a 768 GB streaming pool for textures and that is well within the range of most modern SSDs.

6mNa6YZ.png


Alright so if 16GB of ram contained data for the next second of gameplay how large would the gameplay file on storage need to be for that to be meaningful? This is a goal not an actual. We likely won't see this 'next-Gen Paradigm' until we games in excess of 200GBs.
First off, that 1 sec. of gametime doesn't use 16GB. That's the whole point, that with faster i/o you can refresh your data in ram much faster, thus not needing as big of a pool like current-gen requires.
And yes, the UE5 demo used only 768MB of RAM as it's streaming pool. But they don't tell us how often that pool was updated.
They could update that many times a second based on what's currently on screen. The faster you can refresh that pool, the smaller the actual pool needs to be. If your i/o was slower, you'd need to allocate a bigger pool for the drive to keep up with gpu/frametime demand.

I/O is not that important though as neither system will be fully utilizing their massive transfer bandwidth in anything except bursts until games are allowed to be hundreds of GBs in size. It's unrealistic to fully use even a 2.4 GB/s transfer speed on a 100 GB game. The entire game could be read in 42 seconds. Sounds like a boring game right? That would be 20 seconds with a 5.5 GB/s speed. It's unrealistic to expect games to constantly be able to saturate that as there is just not enough data to fill 5.5 GB/s in anything other than short bursts. A game that constantly needed 5.5 GB/s would have a massive data footprint on storage.
I'm not sure what you are talking about here, but a game isn't stored on a storagemedium like a linear movie, wich it seems you are thinking based on your comment. It could be thousands of induvidual files wich the game engine "cut's & edits" together into a "movie" to use your own flawed analogy.
 

Rikkori

Member
If you have to store 30 seconds worth of assets in memory because of a slow HDD and only 1 second with the SSD then you freed up more memory for what your rendering RIGHT NOW.
I have acknowledged that but there's a nuanced point that I'm making there and which you have missed. That point being that the SSD allows you to use more of your memory for rendering a scene than with an HDD because you have to keep less stored in it from moment to moment BUT that doesn't mean it suddenly acts as a vram! You are still constrained to whatever the memory is as the limit for the scene, and the SSD no matter how fast cannot help overcome that.

Again, I said it over and over again and I'll say it once more, for you: it will help with bandwidth management but it cannot substitute for memory size. What is in question isn't whether or not a console with an SSD can help memory usage more than a HDD, that's a non-sequitur. What is in question is to what extent can a faster SSD make up for less memory.
 

Closer

Member
I don't believe a single word coming from Sony, but I do want everything to go fast. I'm already furious with loading times these days.
 

Neo_game

Member
In the PC world the stated SSD speeds are not sustained. SSD's get warmer and slower in use.
Where is it stated that PS5 SSD achieves 5.5 GB/s at all times/ under load? Did Sony confirm it or are you assuming it?

Pretty sure 5.5gb/sec is min because they have customized solution to get that speed and compatible addon SSD has to be 7gb/sec to match it
 
Last edited:

Boglin

Member
I have acknowledged that but there's a nuanced point that I'm making there and which you have missed. That point being that the SSD allows you to use more of your memory for rendering a scene than with an HDD because you have to keep less stored in it from moment to moment BUT that doesn't mean it suddenly acts as a vram! You are still constrained to whatever the memory is as the limit for the scene, and the SSD no matter how fast cannot help overcome that.

Again, I said it over and over again and I'll say it once more, for you: it will help with bandwidth management but it cannot substitute for memory size. What is in question isn't whether or not a console with an SSD can help memory usage more than a HDD, that's a non-sequitur. What is in question is to what extent can a faster SSD make up for less memory.

To what extent it helps is indeed a great question, but "it does help" is the only argument I have seen made by people championing SSDs.
You know an SSD helps to free up vram that was previously reserved moment to moment versus an HDD. This, in practice, increases your available and actual vram per scene versus an HDD, correct?

So why are you inferring that people are saying the SSD is acting like vram? It's disingenuous to say that it is and it's annoying seeing the same strawman over and over.

I have not seen someone make the argument that SSDs will act as vram, except for only a tiny handful bringing up Xbox's Velocity Architecture's 100GB instantly available for games.
 
Last edited:

GODbody

Member
First off, that 1 sec. of gametime doesn't use 16GB. That's the whole point, that with faster i/o you can refresh your data in ram much faster, thus not needing as big of a pool like current-gen requires.
And yes, the UE5 demo used only 768MB of RAM as it's streaming pool. But they don't tell us how often that pool was updated.
They could update that many times a second based on what's currently on screen. The faster you can refresh that pool, the smaller the actual pool needs to be. If your i/o was slower, you'd need to allocate a bigger pool for the drive to keep up with gpu/frametime demand.


I'm not sure what you are talking about here, but a game isn't stored on a storagemedium like a linear movie, wich it seems you are thinking based on your comment. It could be thousands of induvidual files wich the game engine "cut's & edits" together into a "movie" to use your own flawed analogy.
That's what the image I was responding to was implying. Streaming pool size is mostly irrelevant I was just using it to illustrate how much of a reachable goal that is for most SSD's when configured with at least 16 GB's of RAM.

Games are not configured like movies but an increase in quality demands an increase in data size. Going from 4k assets > 8k assets will double texture sizes of a game's files because 8k is double 4k in data size. Likewise an increase in speed demands an increase in data to maintain that high speed. Demon souls remake for example is 66 GBs in size. If you have a 5.5 GB/s raw bandwidth speed, the entirety of Demon souls remake could be read in 11 seconds if the bandwidth is fully saturated. This is not a likely, or a realistic scenario so realistically, the bandwidth used /s is going to be much lower than that. That 5.5 GB/s would likely be used in bursts during level transitions but not much anywhere else if the games size is small.

If you want to full saturate a 5.5 GB/s bandwidth speed you would need enough data (i.e. 5.5 GB/s worth) to throw at it to require it to run at that speed like my previous post says. 5.5 GB/s would be good in bursts when you're transitioning between levels but for the vast majority of gameplay time the game is not going to be swapping out 5.5 GB/s of RAM every second, unless you expect a game with a significant amount of reused assets to save storage space or you expect a game that is large in size.


Games do not need to saturate the SSD bandwidth 100% percent of the time to be useful, just like not every scene needs to utilize the CPU or GPU at 100% all the time.
You're looking at the usefulness of this badwidth from too large a scale. Scenes don't have to be loaded at a per second basis and those short bursts you mentioned are all that's needed to be massively useful. When playing at 60FPS with the PS5, you could stream in over 90MB per frame if the latency is low enough.

It can be used for removing low quality LOD models in certain areas of the games, which saves storage space, and to avoid pop in or just load in something quickly on demand that is used occasionally.

For example in a first person shooter game you can switch weapons weapons on the fly. That gun model does not need to be kept in memory at all times anymore, just waiting for you to press the button, because now you can stream it in fast enough from storage.

Even on a larger scale the game would not necessarily need to be massive if it's loading assets that aren't entirely unique at every corner of the map. It could still take advantage of the memory speed by streaming in common assets.

A functional example would be a racing game where parts of the course and environment are streamed in as you're driving around.
Those assets can be reused throughout every course in the game and the same ones can even be unloaded and reloaded multiple times as needed within the same race as you're driving around.

Obviously not all games will benefit, but I just thought of these examples off the top of my head and they would be hindered by any speed reduction. The faster the SSD, the better!

None of that makes full use of a 5.5 GB/s speed though without the quality of the underlying asset increasing over current gen. The speed of those transitions can only get so fast before you experience diminishing returns. Most to all of that could be achieved with intelligent data management, which would improve all areas of a game and not just in single scenes. My post was mostly in response to the dream of the 'Next-Gen Paradigm' that Cerny spoke of in the Road to PS5, the ideal of being able to dump and refill RAM every couple seconds. Even then. I saw someone in this thread pointing towards 8k assets as becoming normalized due to the SSD speed which is unrealistic while games are only allowed to occupy so much drive space. Sure an increase in speed would decrease the loading times of things like what you've described but if you need to increase the quality of those assets that are being transferred to keep in line with next-gen transitions game sizes will also need to increase.
 

Boglin

Member
None of that makes full use of a 5.5 GB/s speed though without the quality of the underlying asset increasing over current gen.

I disagree with the quoted part just because any speed increase lowers the amount of vram that has to be reserved for upcoming scenes.
I'm going to be using data rate per frame rather than per second because you can see just how granular you can get while saturating such high bandwidths. Assuming no other factors, if you halve the PS5's SSD's fully saturated bandwidth at 30FPS, you go from being able to stream 180MB+ per frame to 90MB+ per frame. That's a lot of memory to work with and I believe that will be used in scenarios similar to my "gun swapping" example I mentioned earlier.
You can even use it for simpler things like dumping unused assets and using the freed up memory to lower the amount of fluctuations with features like dynamic resolutions and then reloading the asset as needed.

This gen will be very memory limited and I would wager certain first party developers will show substantial benefits of the higher bandwidth in real world scenarios but I have no concrete evidence.
I should probably be clear on my stance that I don't think these benefits will be widely used. I think a good portion of exclusive games and all multiplatform games will just use the increased bandwidth for quicker loading times, but I just can't imagine Sony wasting money on silicon space for I/O to reach those speeds if it was superfluous when the same space could have been used for more CUs.

I fully agree with you about dumping and refilling the entirety of RAM ,though. As far as I'm concerned, the new Ratchet and Clank is the closest we'll get to a scenario even approaching the need to do something like that which is practically a tech demo. Or maybe something like simultaneous worlds of The Medium with higher quality assets but that's a very niche game mechanic.
 

GODbody

Member
I disagree with the quoted part just because any speed increase lowers the amount of vram that has to be reserved for upcoming scenes.
I'm going to be using data rate per frame rather than per second because you can see just how granular you can get while saturating such high bandwidths. Assuming no other factors, if you halve the PS5's SSD's fully saturated bandwidth at 30FPS, you go from being able to stream 180MB+ per frame to 90MB+ per frame. That's a lot of memory to work with and I believe that will be used in scenarios similar to my "gun swapping" example I mentioned earlier.
You can even use it for simpler things like dumping unused assets and using the freed up memory to lower the amount of fluctuations with features like dynamic resolutions and then reloading the asset as needed.

This gen will be very memory limited and I would wager certain first party developers will show substantial benefits of the higher bandwidth in real world scenarios but I have no concrete evidence.
I should probably be clear on my stance that I don't think these benefits will be widely used. I think a good portion of exclusive games and all multiplatform games will just use the increased bandwidth for quicker loading times, but I just can't imagine Sony wasting money on silicon space for I/O to reach those speeds if it was superfluous when the same space could have been used for more CUs.

I fully agree with you about dumping and refilling the entirety of RAM ,though. As far as I'm concerned, the new Ratchet and Clank is the closest we'll get to a scenario even approaching the need to do something like that which is practically a tech demo. Or maybe something like simultaneous worlds of The Medium with higher quality assets but that's a very niche game mechanic.
Yes indeed much more data per frame but the entire scene will likely still be tied to a budget due to system memory constraints. 5.5 GB/s is fast but it's still almost 100x slower than the 448 GB/s of system memory.

My point still stands though, increased data per frame requires increased data and we won't see the full potential of that SSD speed utilized until we have relatively large games that have large pools of assets. Smaller games won't require much data to be shifted from storage to system memory.

Sony confirmed it.
images


22GB/s was the maximum
22GB/s is the maximum speed for lossy compressed data. 8-9 GB/s is the average for lossless compressed data given by Sony
 
Last edited:

Darklor01

Might need to stop sniffing glue
I posted this elsewhere, but, these can now be pre-ordered;


$229.99+tax.

Not sure if this is better than the Sabrent Rocket 4

 
Last edited:
During the Reveal the PS5 showed that it had a thunderbolt port on the front. Do you think we can safely assume that we can connect a thunderbolt SSD here ? Or do we think this will be reserved for PSVR or another external device.

I Know they have some Thunderbolt 3 NVME drive where we can get at least 2500MB/S and My guess is it can go higher .. I am wondering how fast a drive we can get to work on this port.
 

Saberus

Member
I posted this elsewhere, but, these can now be pre-ordered;


$229.99+tax.

Not sure if this is better than the Sabrent Rocket 4

Sabrent is to slow.. (speeds can reach up to 5000 MB/s (read) and 4400 MB/s (write) Needs to be faster to make up because these are only two channels, you need higher speed 7000 MB/s to make up for the channel difference.
 

Tarin02543

Member
During the Reveal the PS5 showed that it had a thunderbolt port on the front. Do you think we can safely assume that we can connect a thunderbolt SSD here ? Or do we think this will be reserved for PSVR or another external device.

I Know they have some Thunderbolt 3 NVME drive where we can get at least 2500MB/S and My guess is it can go higher .. I am wondering how fast a drive we can get to work on this port.

sorry that is usb c.
 

Darklor01

Might need to stop sniffing glue
Sabrent is to slow.. (speeds can reach up to 5000 MB/s (read) and 4400 MB/s (write) Needs to be faster to make up because these are only two channels, you need higher speed 7000 MB/s to make up for the channel difference.

I guess I'll wait on Sony to reveal recommended ones. Thanks!
 

Bitmap Frogs

Mr. Community
Besides that only used a 768 GB streaming pool for textures and that is well within the range of most modern SSDs.

Those peak speeds are only for sequental reads. Random reads which are more inline to your actual usage don't hit those speed...

Here's a comparison random 4k read (wd blue sn550 chosen to represent XboxX storage, based on sequential reading speeds):

Storage-Review-WD-Blue-SN550-Rnd-Read-4-K.png


Here is that same benchmark on the 980 pro

Storage-Review-Samsung-980-Pro-1-TB-Rnd-Read-4-K.png


Notice that the WD blue peaks at 390,000 iops at 800 miliseconds while the 980 hits 390,000 iops at about 100 miliseconds and tops out at about 550,000 iops at 250 miliseconds.

Personally, I don't expect this to have any impact whatsoever on graphics, but the quality of life of almost instant storage is just magical.
 
Last edited:
Top Bottom