Sony's Cross-Generation Game Messaging [2021]

[PSman1700]

Maybe you confuse 2020 and 2019 I don't remembre an AMD GPU with raytracing in 2019.

Im talking performance here, rendering performance. The PS5 closest match in raw power is a 5700XT, aside from the weak ray tracing, ofcourse.

For the cost it could not have done better.

Arguebly, true. Time, could have, though. With basically no availability and the pandemic going on, they perhaps could have launched a year or so later and went with better RDNA3 ray tracing/reconstruction, Zen3 etc etc.

I gave the one in the PS6 and next Xbox.

You cant give or have any idea of what and how the next xbox and PS will look like (if there will be ones, ofcourse).

Again I don't care about having 64 GB for a console it is a stupid decison because it cost more than have a low latency and fast SSD and less memory.

Never said it was optimal for costs in a console, i was giving examples, as i stated. And yes, fast main ram offers both much faster and lower latencies then your PS5 will do

I bet they will continue

Maybe, no idea, 7 to 8 years is alot, i for gods sake hope we wont be stuck with this architecture either. Even it is so great were already hitting limitations everywhere.

I know the console will follow this design(low latency and fast storage(SSD), unified memory, and a powerful APU).

Bold statement again You cant look in the future, i doubt even Sony knows lol.

I talk about the same

True and all, but you still made the claim the devs can do whatever they want without limitations. Which is false anyway, even in game design if you want that true next gen leap.

On this site you can learn how videogames are made:
https://www.gamedesigning.org/video-game-development/

This can help you to learn and said things making sense.

Yeah, i suggest you start there

 

[chris1515]

Im talking performance here, rendering performance. The PS5 closest match in raw power is a 5700XT, aside from the weak ray tracing, ofcourse.



Arguebly, true. Time, could have, though. With basically no availability and the pandemic going on, they perhaps could have launched a year or so later and went with better RDNA3 ray tracing/reconstruction, Zen3 etc etc.



You cant give or have any idea of what and how the next xbox and PS will look like (if there will be ones, ofcourse).



Never said it was optimal for costs in a console, i was giving examples, as i stated. And yes, fast main ram offers both much faster and lower latencies then your PS5 will do



Maybe, no idea, 7 to 8 years is alot, i for gods sake hope we wont be stuck with this architecture either. Even it is so great were already hitting limitations everywhere.



Bold statement again You cant look in the future, i doubt even Sony knows lol.



True and all, but you still made the claim the devs can do whatever they want without limitations. Which is false anyway, even in game design if you want that true next gen leap.



Yeah, i suggest you start there

No this is not false this is not because you are ignorant and does not know anything about game developement that this is true.

This is not a bold statement but a clever one. In the next 6 years there is nothing cost efficient looking like it will replace SSD(Reram is too expensive), unified memory is clever I don't think they will go back to stupid design like having system RAM and VRAM, APU is cost efficient. This is not science fiction this is just knowing what are the technology which are available on the market and having a notion of R&D and time to market.

If they release a new hardware, I am sure Sony have a high level idea of the design of the PS6 same for Microsoft. Maybe cloud gaming will replace totally consoles, this is another problem. I don't believe it but we never know.

 

[PSman1700]

No this is not false this is not because you are ignorant and does not know anything about game developement that this is true.

It seems the only you can do is quoting others and sharing random twitter messages and spam the board with GIFs. You dont make any cases with those, you shatter them.

just knowing what are the technology which are available on the market.

This aint the thread for it anyway with your PS6 speculations.

Maybe cloud gaming will replace totally consoles.

Maybe, maybe not.

 

[chris1515]

It seems the only you can do is quoting others and sharing random twitter messages and spam the board with GIFs. You dont make any cases with those, you shatter them.



This aint the thread for it anyway with your PS6 speculations.



Maybe, maybe not.

I am not game dev but I am interested and I read and learn about the subject. I am a developer but not in game developpement. And when I post a tweet I understand it. I make a promise next time I post a tweet quoting one of your message, I will give an explanation. I use gif with an explanation because it seems you don't understand PS5 can fully replace it full memory in less than 2 seconds maybe like little children an image or a video is easier to understand after what it means for game design is another problem but you probably need to read and ask more question of how game developpement works.

What is game design? What is level design and why it was so technical before having a fast storage? It can help you.

https://www.cgspectrum.com/blog/what-is-game-design

You will see here game designer don't care about native 4k and raytracing

http://level-design.org/

After this it will probably be easier for you to learn why it is a revolution for game and level designer to not work under high streaming constraint.

This two hours Andy Gavin interview can help you too.

EDIT: Another example Mario maker let the player be a level designer. Dreams let the player do some game design because you can do different type of game and the player define the rules of the game.

Super mario maker 2

Dreams

Maybe as a PC player you don't know this two games where the player create some level of Mario a platformer of Nintendo a japanese console platform. And Dreams a game where you can create little games defining the game design made by Media Molecule for Sony Playstation another japanese console platform.

some information about the franchise Mario
https://mario.nintendo.com/es/

 

[PSman1700]

The ssd/io is kinda like having an entire game in a huge amount of memory lets say 64gb vram, you still need to render and process all that.

 

[chris1515]

The ssd/io is kinda like having an entire game in a huge amount of memory lets say 64gb vram, you still need to render and process all that.

And this is exactly like I said the limit will be game size, CPU and GPU power but the graphical fidelity will not be a game design limitation. With an HDD it was worse because graphic fidelity was limited by the streaming. Here it will not be the case only limited by the GPU and the design will not take into account streaming.

A good explanation of streaming.

Another good explanation and the designer will be able to create a city into an open world as they want, not like in Jak 2 or any open world. Many tropes to hide loading will disseapear like elevator, little corridor and so on...

 

[goonergaz]

@chris1515 let it go mate - just enjoy the tech and that you know what it brings...there will always be people who won't understand/agree, it's pretty obvious what the IO brings to the table and that Sony wouldn't have spent the time and money developing it for no good reason, it's just a massive shame Sony didn't fully get behind it with some early games and are dropping the 'PS5 exclusive' aspect so that's bringing it's advantages even more into question.

 

[PSman1700]

And this is exactly like I said the limit will be game size, CPU and GPU power but the graphical fidelity will not be a game design limitation. With an HDD it was worse because graphic fidelity was limited by the streaming. Here it will not be the case only limited by the GPU and the design will not take into account streaming.

A good explanation of streaming.

Another good explanation and the designer will be able to create a city into an open world as they want, not like in Jak 2 or any open world. Many tropes to hide loading will disseapear like elevator, little corridor and so on...

No matter how fast your storage is, even if the SSD/Nvme was as fast as main ram or even GDDR ram, all that information stored still has to be rendered and processed somewhere. What storage does is exactly that, storing information, to later pass it on to example the GPU which then renders this information.
What we have seen is consoles basically having not seen much if any increase in cold storage, being stuck on mechanical spinning drives for a long time. That has been a bottleneck, storage has been playing catching up now with this generation.

Bottlenecks still exist, the perfect architecture aint there yet and wont be for a while to come i think. You'l always need compute power, and the increases are as much needed there, even more so then SSD speeds.
Its almost if your trying to say the SSD/IO is limited by the GPU, if so, they have created a bottleneck and imbalanced system then. Im seeing bottlenecks like the kinda underpowered gpu, at just 10TF's it aint all that much of a compute increase (which still matters, fast feeding of information or not), the lack of meaningfull ray tracing hardware (thus limiting to performance modes again), 16gb of ram (2x increase) which still is the much and much faster solution over the SSD aswell and even more important the latencies.
The CPU should suffice, but that doesnt mean it wont introduce any bottlenecks either, its a cut down zen2 8 core cpu at reduced clock speeds, and with the introduction of 120fps gaming, its going to be a bottleneck somewhere sometime, no matter how fast the SSD is.

Aside from that, yea, the PS5 is already outclassed spec wise, 5700XT with weak RT (2019 mid ranger), 3.5ghz zen2, and an SSD thats already outclassed before compression. After compression things will only improve on the PC side.
Theres a reason why were not really getting that true next generational leap some hyped about pre-launch. We can make excuses and blame it on cross-gen, but what about rift apart and demon souls, those are supposed native PS5 games, which still dont give that leap some said we where going to get.

 

[chris1515]

No matter how fast your storage is, even if the SSD/Nvme was as fast as main ram or even GDDR ram, all that information stored still has to be rendered and processed somewhere. What storage does is exactly that, storing information, to later pass it on to example the GPU which then renders this information.
What we have seen is consoles basically having not seen much if any increase in cold storage, being stuck on mechanical spinning drives for a long time. That has been a bottleneck, storage has been playing catching up now with this generation.

Bottlenecks still exist, the perfect architecture aint there yet and wont be for a while to come i think. You'l always need compute power, and the increases are as much needed there, even more so then SSD speeds.
Its almost if your trying to say the SSD/IO is limited by the GPU, if so, they have created a bottleneck and imbalanced system then. Im seeing bottlenecks like the kinda underpowered gpu, at just 10TF's it aint all that much of a compute increase (which still matters, fast feeding of information or not), the lack of meaningfull ray tracing hardware (thus limiting to performance modes again), 16gb of ram (2x increase) which still is the much and much faster solution over the SSD aswell and even more important the latencies.
The CPU should suffice, but that doesnt mean it wont introduce any bottlenecks either, its a cut down zen2 8 core cpu at reduced clock speeds, and with the introduction of 120fps gaming, its going to be a bottleneck somewhere sometime, no matter how fast the SSD is.

Aside from that, yea, the PS5 is already outclassed spec wise, 5700XT with weak RT (2019 mid ranger), 3.5ghz zen2, and an SSD thats already outclassed before compression. After compression things will only improve on the PC side.
Theres a reason why were not really getting that true next generational leap some hyped about pre-launch. We can make excuses and blame it on cross-gen, but what about rift apart and demon souls, those are supposed native PS5 games, which still dont give that leap some said we where going to get.

First RDNA is more performant than GCN pure flop We have a GPU around 6 to 7 times more powerful this is good. This is like comparing Kepler and Maxwell on Nvida GPU.

The GPU is around two times slower than the best GPU in rasterization in multiplatform game and two and a half to three times slower in hybrid rasterization/raytracing multiplatform game compared to Nvidia best offer. This is a good level of performance for something inside a 400 or 500 dollars/euros consoles. And pure triangle RT will not be in every engine.

And on CPU side this is difficult to find a 8 cores 16 threads two time faster than PS5 and Xbox Series X CPU. I suppose the multithreading and games will be tailored around this configuration. After I am not a PC gamer and maybe an Intel or AMD offer 8 cores 16 threads CPU going two times or better than Zen 2 CPU inside consoles. Desktop have four times more cache and runs at higher clock but not two times better.

The raytracing is not as powerful as on Nvidia side but it is more flexible on consoles than PC because of DXR limitation and we will see some innovation on this side and this is valid for Xbox Series too.

Where did I say GPU slowing down SSD learn to read again? I just said streaming is not the limitation anymore for game design only the CPU, GPU and in the future game size because the slowest part storage is always the bottleneck and this time the storage is fast enough to not be a bottleneck. Before HDD was not able to load the data the GPU was able to render. This is not the case anymore, game and level designer just need to stay inside the rendering budget and CPU power and this is ok.

RAM size and streaming capabilites are linked. This is why virtualized geometry and texturing like UE5 needs less RAM but a low latency and fast storage. UE 5 virtualisation of geometry and texturing needs less RAM than old technology not using virtualisation.

Here it is a good explanation, listen what Mark Cerny about RAM usage on old gen where they store everything in RAM for the next 30 seconds even some assets we will never render because this is an open world and the player can go anywhere.

http://www.xcg.cs.pitt.edu/papers/conn-ems12.pdf

study about virtual texturing and SSD in 2012

https://publik.tuwien.ac.at/files/PubDat_193670.pdf

Study about virtual texturing

https://fabiensanglard.net/ssd/

Like I said learn, you think to know but make a games and tune a PC and see benchmark is not the same thing.

Director of R&D John Carmack said a few words about how SSD performances "reboot" the old dichotomy: "RAM" Vs "Storage Drive".

This is not sure the 7GB/s goes faster than the SSD in the PS5 in real world. Mark Cerny told because of customization in the controller and priority queue they need faster SSD when they will be able to deliver internal SSD expansion for PS5. But they will soon go faster than the PS5 SSD when in 2022 PCIE 5 SSD will arrive. The PS5 SSD speed is enough for a console with 16 GB of RAM.

That's your opinon not a general one this two games and Horizon Forbidden West PS5 looks like a generational leap compared to the following PS4 game Ratchet 2016, Bloodborne and Horizon Zero Dawn. Digitalfoundry think it will be hard to hide the difference when HFW will be on PS4.

And seeing Unreal Engine 5 we have more impressive things coming later during the generation.

After I think the improvement this generation will take more time to arrive due to cross gen and having the same X86 and ISA compatible GPU. For example Unreal Engine 5 can improve a lot on Nanite side and on Lumen side too. Epic are pioneer and have a few years advantage compared to other engine.

 

[PSman1700]

First RDNA is more performant than GCN pure flop We have a GPU aroung 6 to 7 times more powerful this is good.

That applies to PS3 to PS4 aswell, i would say, an even bigger IPC improvement going from 7800GTX arch to GCN1.1 arch. Its the smalles increase in compute power ever going from PS4 to PS5.

The GPU is around two times slower than the best GPU in rasterization and three times in raytracing compared to Nvidia best offer.

Nope, were at over 23+TF's on AMD's side, which already more than twice the performance as opposed to the PS5's at 10TF. That already is a huge difference in just compute, those AMD gpus also have infinity cache and the full feature set that RDNA2+ has to offer. They also aint bandwith limited as the PS5 is, which does and will limit performance. Also much stronger RT performance on those due to the mentioned capabilities.

In pure compute, on NV's side we are at 36TF's of raw power, well over twice the amount what the PS5 has to offer, close to three times its performance all things considered like TF scaling etc.
In ray tracing three times, but more so when things get hammered, heck, even Turing does a better job.

Oh and thats before taking DLSS (AMDs equalevant) into consideration which the PS5 lacks.

The raytracing is not as powerful as on Nvidia side but it is more flexible on consoles than PC because of DXR limitation and we will see some innovation on this side and this is valid for Xbox Series too.

Dunno about that, nothing that we have seen yet atleast. Contrary to some beliefs, the ray tracing on Ampere isnt all that 'limited' and quite flexible aswell.

RAM size and streaming capabilites are linked. This is why virtualized geometry and texturing like UE5 needs less RAM but a low latency and fast storage. UE 5 virtualisation of geometry and texturing needs less RAM than old technology not using virtualisation.

The RAM size has alot to do with cost, they had to find the best balance between cost and performance ratio, and they found it at 16GB. The SSD is also, to an extend, making up somewhat for this.

This is not sure the 7GB/s goes faster than the SSD in the PS5 in real world. Mark Cerny told because of customization in the controller and priority queue they need faster SSD when they will be able to deliver internal SSD expansion for PS5. But they will soon go faster than the SP5 SSD when in 2022 PCIE 5 SSD will arrive.

We will perhaps never know, but what we do know is that on raw speeds, the current crop of top end SSD's already have faster transfer speeds and quite considerably so.
The optimizations needed are largely covered by GPUs taking care of compression and DirectStorage on windows side improving the software stacks. Im certainly sure this will be atleast enough to keep up with the PS5's new IO subsystem and then some.

That's your opinon not a general one this two games and Horizon looks like a generational leap compared to the following PS4 game Ratchet 2016, Bloodborne and Horizon Zero Dawn.

I think rift apart looks amazingly nice, but it aint the leap we have had from previous generations. And thats all OK to me, it doesnt really matter, its the fact that some have been hyping up things way too much IMO.
UE5 tech demos still stand on top for the wow-factor.

After I think the improvement this generation will take more time to arrive due to cross gen and having the same X86 and ISA compatible GPU. For example Unreal Engine 5 can improve a lot on Nanite side and on Lumen side too.

Yes maybe where seeing Apple hardware in next generation consoles and pc's, who knows. I hope not, but, whatever if it happens, if its faster i dont care whats powering my gaming systems.

 

[chris1515]

That applies to PS3 to PS4 aswell, i would say, an even bigger IPC improvement going from 7800GTX arch to GCN1.1 arch. Its the smalles increase in compute power ever going from PS4 to PS5.



Nope, were at over 23+TF's on AMD's side, which already more than twice the performance as opposed to the PS5's at 10TF. That already is a huge difference in just compute, those AMD gpus also have infinity cache and the full feature set that RDNA2+ has to offer. They also aint bandwith limited as the PS5 is, which does and will limit performance. Also much stronger RT performance on those due to the mentioned capabilities.

In pure compute, on NV's side we are at 36TF's of raw power, well over twice the amount what the PS5 has to offer, close to three times its performance all things considered like TF scaling etc.
In ray tracing three times, but more so when things get hammered, heck, even Turing does a better job.

Oh and thats before taking DLSS (AMDs equalevant) into consideration which the PS5 lacks.



Dunno about that, nothing that we have seen yet atleast. Contrary to some beliefs, the ray tracing on Ampere isnt all that 'limited' and quite flexible aswell.



The RAM size has alot to do with cost, they had to find the best balance between cost and performance ratio, and they found it at 16GB. The SSD is also, to an extend, making up somewhat for this.



We will perhaps never know, but what we do know is that on raw speeds, the current crop of top end SSD's already have faster transfer speeds and quite considerably so.
The optimizations needed are largely covered by GPUs taking care of compression and DirectStorage on windows side improving the software stacks. Im certainly sure this will be atleast enough to keep up with the PS5's new IO subsystem and then some.



I think rift apart looks amazingly nice, but it aint the leap we have had from previous generations. And thats all OK to me, it doesnt really matter, its the fact that some have been hyping up things way too much IMO.
UE5 tech demos still stand on top for the wow-factor.



Yes maybe where seeing Apple hardware in next generation consoles and pc's, who knows. I hope not, but, whatever if it happens, if its faster i dont care whats powering my gaming systems.

For DLSS wait a little Sony have a new reconstruction method too.

I don't compare GPU with specs sheet. I compare GPU in realworld performance and Ampere GPU aren't three times faster than PS5 GPU in games non using raytracing. Ampere flop are less efficient than Turing flop for example. a 36 Tflops Turing GPU would be faster.

For decompression we need to wait the only study I saw using CUDA for decompressing a 1GB/s it needs a full 980 to follow the SSD speed decompressing some compressed data using LZW algorithm but I suppose the sofware will be much better on DirectStorage side.

https://www.researchgate.net/publication/300414932_Fast_LZW_Compression_Using_a_GPU

We will know for the SSD when Sony will release the spec they need to use inside the bay expansion. It is rumored to arrive into a summer uptdate maybe with COVID19 it will be again late and arrive this winter. But we will know and being able to compare maybe it is only 6 GB/s or maybe it is 7GB/s.

 

[iroboto]

I appreciate a civil disagreement.

I just think having seen the length and speed of these responses, that I could perhaps give you both back time with your lives to enjoy the weather, your family, some games, etc. If you both recognize that this discussion is likely not going anywhere but is sucking up all your time

 

[manux]

I appreciate a civil disagreement.

I just think having seen the length and speed of these responses, that I could perhaps give you both back time with your lives to enjoy the weather, your family, some games, etc. If you both recognize that this discussion is likely not going anywhere but is sucking up all your time

I would add that based on experiment I did it's unlikely anyone would change their opinion. Pointless to discuss/debate if outcome is decided before any argument is made/evidence is provided.

 

[BRiT]

Pointless to discuss/debate if outcome is decided before any argument is made/evidence is provided.

Indeed.

 

[chris1515]

I would add that based on experiment I did it's unlikely anyone would change their opinion. Pointless to discuss/debate if outcome is decided before any argument is made/evidence is provided.

I will not change of opinion knowing mine have a base, he just need to read some documents. I posted some link. And for performance goes follow sites comparing PC version to PS5 and Xbox Series performance. And some benchmark of games running Turing GPU against Ampere GPU.

2080 Ti have less flops than a 3070 but the performance are nearly the same between the two GPU in games benchmark.

I hope he choose his GPU seeing benchmark and not reading spec sheet.

 

[PSman1700]

For DLSS wait a little Sony have a new reconstruction method too.

Like the posts below, i'd want to see evidence of that, its all just 'on paper' untill it materializes

I don't compare GPU with specs sheet. I compare GPU in realworld performance and Ampere GPU aren't three times faster than PS5 GPU in games non using raytracing. Ampere flop are less efficient than Turing flop for example. a 36 Tflops Turing GPU would be faster.

And thats the thing, just like the PS5 SSD, we can only go after what the specs say, since were not actually seeing much of games using the next generation capabilities. I have no trouble believing 5.5gb/s and 36TFs could both be used anyway, somewhere. The 3090 could very well be three times faster in situations where compute matters.

For decompression we need to wait the only study I saw using CUDA for decompressing a 1GB/s it needs a full 980 to follow the SSD decompressing some LZW but I suppose the sofware will be much better

https://www.researchgate.net/publication/300414932_Fast_LZW_Compression_Using_a_GPU

We will know for the SSD when Sony will release the spec they need to use inside the bay expansion. It is rumored to arrive into a summer uptdate maybe with COVID19 it will be again late and arrive this winter. But we will know and being able to compare maybe it is only 6 GB/s or maybe it is 7GB/s.

Yes well, NV was claiming quite high numbers with their RTX IO presentation, again this is largely 'on paper' again but we dont really have much else to go on right now.

I appreciate a civil disagreement.

I just think having seen the length and speed of these responses, that I could perhaps give you both back time with your lives to enjoy the weather, your family, some games, etc. If you both recognize that this discussion is likely not going anywhere but is sucking up all your time

Oh absolutely, i do appreciate civil conversations, its top priority almost i think.

Its not all that time consuming for me, im using a laptop to write my comments mostly (atleast the longer ones), the longer posts take around five to ten minutes to compose, and i write fast, so it aint a big deal. No idea about chris but i hope not all his time goes into posting or something

I would add that based on experiment I did it's unlikely anyone would change their opinion. Pointless to discuss/debate if outcome is decided before any argument is made/evidence is provided.

Yes and thats the thing, theres not much to go after, most is tech specs discussions etc. Its alot of numbers throwing, presentations, spec sheets etc and what we personally think.

I will not change of opinion knowing mine have a base, he just need to read some documents. I posted some link. And for performance goes follow sites comparing PC version to PS5 and Xbox Series performance. And some benchmark of games running Turing GPU against Ampere GPU.

2080 Ti have less flops than a 3070 but the performance are nearly the same between the two GPU in games benchmark.

I hope he choose his GPU seeing benchmark and not reading spec sheet.

You certainly have your base, but so do i. The SSD/IO is very important, but so is the GPU/compute power. Id guess both arguments have their bases.
TF's do scale and usually improve IPC from gen to gen if thats what you mean, that accounts all generations of consoles. Im on a 2080Ti now, i didnt have to research all that much to know the 2080Ti would be the fastest and most capable gpu to get back in 2018.
Between a 3070 and 2080Ti, if i could choose between those two i'd take the 3070 in a heartbeat, they might be performing close now, but that might and probably will change in the future. And thats just talking normal rendering performance.

To say, the PS5 is an amazing machine and they found the best belance between cost, performance and size/heat, just like the XSX and XSS. They currently offer a bang for the buck thats hard to beat on the pc side. You get graphics out the box for 500 dollars (or less) like rift apart that slot in on the top end of graphics so far. The IO enables fast (or total lack of) loading and the RT, even though subtle, is there alongside other improvements like massive worlds full of detailed assets.

I just am of the idea that were not ready to abandon GPU (and CPU) prowess yet, nor its feature sets and all, the increases there still matter as much as in previous generations.

 

[Karamazov]

i guess we can see another effect of faster IO speeds, other than loading times, this gen compared to last gen with ratchet and HZFW where the character lighting and details stays the same between gameplay/cutscene, before as you could not fit these in ram you had to turn of some effects, so it might at last be the end of cutscene VS gameplay graphics debate(that "hero lighting" mentionned by DF).

Gameplay getting to the same quality as cutscenes or tech demos, as a lot of ram is freed for additionnal details/effects during gameplay.

 

[see colon]

I don't want to get too far into the weeds with this storage speed argument but I do think history is important. N64 carts have a read speed something like 6MB/s, and the console had 4MB of system ram. PS2's DVD drive has a read speed that could fill it's system memory in 6 seconds. There are N64 games with load times. There are, to my knowledge, no PS2 games that load in 6 seconds. I/O speeds are important, but just because you can move data into RAM doesn't mean that it's in it's useable state. And I don't mean just things like compression. That's handled by the on board compression hardware. Think about a game like Ratchet and Clank where you are swapping between worlds and unloading data for the worlds completely, for example. What if you add in massive amount of physics based destruction, worldwide, like Red Faction (the first one on PS2). Now imagine you destroy parts of the world, swap to world 2. Do some stuff and swap back. If you are unloading world 1 from memory, and you want that destruction to stay, you will likely have to recalculate the destruction. That takes time. Probably not too much time on today's hardware, but if you add in other things you may need to process while loading between worlds, you can see how this will add up and give you load times. Is there a solution to this? Sure, you bake all the destruction you have and save the world state by dumping the level data stored in RAM back to your super fast SSD so when you swap back there is no need to calculate. But... You are already saturating the I/O loading in your new level, so writing back an equal amount of data before you read in your new level data is going to double your load time. Or maybe make it longer, assuming you are loading in compressed data and writing uncompressed data.

Obviously there are optimizations that could be made. You could, in the right circumstances, write only the data you need for the changes made to the level, assuming that works for your game and engine. And perhaps your two levels have many shared assets that wouldn't need to be swapped. But you can see how a situation like this can be limited by I/O even with the fast speeds we have now. It's less of a bottleneck, of course. But you can still have it limit your performance in the right circumstance.

 

[PSman1700]

But you can see how a situation like this can be limited by I/O even with the fast speeds we have now. It's less of a bottleneck, of course. But you can still have it limit your performance in the right circumstance.

True, and everything thats being done on screen still has to be output and rendered/processed by the GPU and CPU somewhere, the faster feeding of data only increases the need for more bandwith and more powerfull compute performance.

 

[chris1515]

I don't want to get too far into the weeds with this storage speed argument but I do think history is important. N64 carts have a read speed something like 6MB/s, and the console had 4MB of system ram. PS2's DVD drive has a read speed that could fill it's system memory in 6 seconds. There are N64 games with load times. There are, to my knowledge, no PS2 games that load in 6 seconds. I/O speeds are important, but just because you can move data into RAM doesn't mean that it's in it's useable state. And I don't mean just things like compression. That's handled by the on board compression hardware. Think about a game like Ratchet and Clank where you are swapping between worlds and unloading data for the worlds completely, for example. What if you add in massive amount of physics based destruction, worldwide, like Red Faction (the first one on PS2). Now imagine you destroy parts of the world, swap to world 2. Do some stuff and swap back. If you are unloading world 1 from memory, and you want that destruction to stay, you will likely have to recalculate the destruction. That takes time. Probably not too much time on today's hardware, but if you add in other things you may need to process while loading between worlds, you can see how this will add up and give you load times. Is there a solution to this? Sure, you bake all the destruction you have and save the world state by dumping the level data stored in RAM back to your super fast SSD so when you swap back there is no need to calculate. But... You are already saturating the I/O loading in your new level, so writing back an equal amount of data before you read in your new level data is going to double your load time. Or maybe make it longer, assuming you are loading in compressed data and writing uncompressed data.

Obviously there are optimizations that could be made. You could, in the right circumstances, write only the data you need for the changes made to the level, assuming that works for your game and engine. And perhaps your two levels have many shared assets that wouldn't need to be swapped. But you can see how a situation like this can be limited by I/O even with the fast speeds we have now. It's less of a bottleneck, of course. But you can still have it limit your performance in the right circumstance.

Why do you want to write the state of the world at the moment you change of world? Each time something is destroyed into the world it is the updated on the SSD, this is fast and out of extreme case you don't saturate the I/O during normal streaming and if you switch world you read the updated state of the world.

And on Ps2 you read from the disk, the theoretical speed is far from realworld speed. You have the same problem than on HDD.