General Next Generation Rumors and Discussions [Post GDC 2020]

Together with a much better CPU especially vs a lower relative amount of RAM I would like to see figures that show that is or will be a considerable amount of overhead?
Well if we look at the pieces and compare to this gen, and PC running the same game on faster CPUs, it should be possible to get a ball-park estimation. In short though, let's say the SSD removes that bottleneck and we just have CPU to worry about. Something like GTA is doing a lot of CPU work and taking minutes to load; we know that's not all storage reads because other games fill the RAM in much shorter time. So 300 seconds for GTA V, say. If PS5's CPU is 10x faster in that workload, that'll be 30 seconds on PS5, loading the same content and world complexity as GTA V.

I can't see any realistic way to think all games, even those taking minutes to load on faster CPU PCs, will get down to 2 seconds on consoles.

This table from DF shows load times for games where the data was already cached in RAM

upload_2020-4-4_13-55-52.png

That must be CPU workload on an i5 6600K.

The only wild-card is how steaming-friendly load structures develop, if partial loads are implemented across all games so 1-2 load gets you up and running, and then in the background the rest of the game is loaded. But that's more about game design than storage speed.
 
Last edited:
But we don’t know enough about GTA V. Let’s say the game fills up the full 6.5GB before starting, which is not unlikely for ending up in a random location in a big open world. That is 130 seconds x 50MB to fill already, even if the data is well placed and easily streamed and not in multiple files.

It is hard to grasp sometimes just how much faster 5.5GB/s is than we have now, between 55-110 times that of the PS4!

So to be clear, I do expect most work to be actually getting the data. We can do a whole lot of processing even in a single frame after all.
 
Well if we look at the pieces and compare to this gen, and PC running the same game on faster CPUs, it should be possible to get a ball-park estimation. In short though, let's say the SSD removes that bottleneck and we just have CPU to worry about. Something like GTA is doing a lot of CPU work and taking minutes to load; we know that's not all storage reads because other games fill the RAM in much shorter time. So 300 seconds for GTA V, say. If PS5's CPU is 10x faster in that workload, that'll be 30 seconds on PS5, loading the same content and world complexity as GTA V.

I can't see any realistic way to think all games, even those taking minutes to load on faster CPU PCs, will get down to 2 seconds on consoles.

This table from DF shows load times for games where the data was already cached in RAM

View attachment 3746

That must be CPU workload in an i5 6600K.

The only wild-card is how steaming-friendly load structures develop, if partial loads are implemented across all games so 1-2 load gets you up and running, and then in the background the rest of the game is loaded. But that's more about game design than storage speed.

Just trying to load grand theft auto V in pc. When loading story mode(very slow) it seems only 1 cpu is utilized. I have 4 core cpu and load is 20-25%. To me it looks like at least gtav leaves a lot to table to be optimized in console env. use all 8 cores and use kraken. Could make loading a lot faster. I have gtav in nvme ssd that can read about 2GB/s. The loading time of gtav didn't really improve in observable way when I moved gtav from sata ssd into nvme ssd. See attached picture for sadness caused by very poor optimization :/

loading_gtav.png
 
I was wondering about that. I'm not sure you can parallelise game world creation that easily. You need to create lots of memory structures and you run the risk when creating them in parallel in either overwriting memory structures from another thread or fragmenting your memory. That's my crude "just thought about it" understanding and I might be way off!
 
I was wondering about that. I'm not sure you can parallelise game world creation that easily. You need to create lots of memory structures and you run the risk when creating them in parallel in either overwriting memory structures from another thread or fragmenting your memory. That's my crude "just thought about it" understanding and I might be way off!

One naive idea would be to create the structure into ram without content, i.e. read "metadata" only. Then once structure is in ram go about in parallel way to allocating memory and reading in data. Maybe even metadata has to be thought of in structures that can be parsed parallel, i.e. different nodes of trees and such could go parallel.

But fundamentally I think it's just not possible to retrofit into old games. Sony/MS forcing those instant load times will give developers trouble initially and will force rethinking how data/memory is managed. Hopefully developers have known about this for a good while so they have had time to think and re-engineer.
 
Well you might be more right than I thought though ... thought shame on you Rockstar ;)
 
Other than speed one aspect of making life easier is minimising clicks to get to where you want to go. in the XSX suspend scenario I can pick up exactly where I left of in several games in a couple of clicks - it’s an ideal that I really hope is in PS5 because I am not the only user of my console.

Even with quick loading you’d still need to load the game, then the save and then it won’t be exactly where you left off (unless they revamp the save system in games).
You can have several save states/suspend instances per multiple users, per console?
 
You can have several save states/suspend instances per multiple users, per console?

The Xbox Series X has a fixed size portion of the SSD used for instant-resume of the last N items. It depends on the size of the game as to how many you can fit in there. The lowest number is 3 next-gen Series X games. If you're running XBox One or X360 games that could become far more.

I don't know how exactly it's managed across multiple profiles -- it could be a per-user allocated amount of space or it could be allocated just once on the console.

Edit: The floor is 3, not 4 games. From DF Article;

The Velocity Architecture also facilitates another feature that sounds impressive on paper but is even more remarkable when you actually see it play out on the actual console. Quick Resume effectively allows users to cycle between saved game states, with just a few seconds' loading - you can see it in action in the video above. When you leave a game, system RAM is cached off to SSD and when you access another title, its cache is then restored. From the perspective of the game itself, it has no real idea what is happening in the background - it simply thinks that the user has pressed the guide button and the game can resume as per normal.

Microsoft wasn't sharing the actual size of the SSD cache used for Quick Resume, but saying that the feature supports a minimum of three Series X games. Bearing in mind the 13.5GB available to titles, that's a notional maximum of around 40GB of SSD space, but assuming that the Velocity Architecture has hardware compression features as well as decompression, the actual footprint may be smaller. Regardless, titles that use less memory - like the games we saw demonstrated - should have a lower footprint, allowing more to be cached.​
 
Just trying to load grand theft auto V in pc. When loading story mode(very slow) it seems only 1 cpu is utilized. I have 4 core cpu and load is 20-25%. To me it looks like at least gtav leaves a lot to table to be optimized in console env. use all 8 cores and use kraken. Could make loading a lot faster. I have gtav in nvme ssd that can read about 2GB/s. The loading time of gtav didn't really improve in observable way when I moved gtav from sata ssd into nvme ssd. See attached picture for sadness caused by very poor optimization :/

View attachment 3748

This made the Spiderman demo much more impressive.
 
But fundamentally I think it's just not possible to retrofit into old games. Sony/MS forcing those instant load times will give developers trouble initially and will force rethinking how data/memory is managed. Hopefully developers have known about this for a good while so they have had time to think and re-engineer.
Hmmm...nothing has stopped devs trying to create multi-threaded load structures up to now, with the potential to massively reduce them. Has no-one even bothered to look at it before?
 
Out of the games I play GTAV has the absolute worst loadtime. And still it probably is one of the most sold games ever. I doubt developers would prioritize load times above minimum threshold that consumers accept. Latest game I finished was half life alyx. It has disturbing load times, it's so bad in vr waiting in load screen.

I don't think anything but sony/ms/platform owner setting and enforcing minimum bar making the situation better. Perhaps pc gamers benefits as a side effect.

Would be interesting if there was load times of games collected somewhere. Not sure if my memory works proper but I seem to remember uncharted thiefs end and some other ps4 games loaded pretty fast? Or perhaps I just remember wrong.
 
I don't think anything but sony/ms/platform owner setting and enforcing minimum bar making the situation better. Perhaps pc gamers benefits as a side effect.
I can't see them doing that TBH. "Want your multiplatform game on our console? Well we demand you rewrite the entire thing to incorporate multithreaded loading."

Cue a very empty library and the other machines getting lots and lots of games with at worst 30-60 second load times that users are quite happy to tolerate if it means they actually have something to play. ;)

Console games haven't had properly enforced QA for two generations now. XB360 was going to have every game at 720p with 2x MSAA minimum, right? Every single one or it wouldn't pass TRCs... I think competition has meant they've all given up with quality TRCs as they can't afford to lose access to the games people want, even if those games are buggy, low res, low framerate, whatever they wish their console wasn't associated with.
 
I can't see them doing that TBH. "Want your multiplatform game on our console? Well we demand you rewrite the entire thing to incorporate multithreaded loading."

Cue a very empty library and the other machines getting lots and lots of games with at worst 30-60 second load times that users are quite happy to tolerate if it means they actually have something to play. ;)

Console games haven't had properly enforced QA for two generations now. XB360 was going to have every game at 720p with 2x MSAA minimum, right? Every single one or it wouldn't pass TRCs... I think competition has meant they've all given up with quality TRCs as they can't afford to lose access to the games people want, even if those games are buggy, low res, low framerate, whatever they wish their console wasn't associated with.

Would be sad if hardware goes unused. On the other hand that would let platform exclusives shine a lot more.
 
It wouldn't go unused and load times will be much faster. I just don't think 1-2 seconds for every single game is going to happen. 30 second load times is still going to be a huge improvement, and instant restore will provide '1-2 second load times' for games you are currently playing, satisfying that marketing feature.

In short, I don't think that '1-2 second load time' suggestion should be taken as a baseline from which we extrapolate back to how that'll be achieved for every game. It's probably better to look at the IO specs and extrapolate forwards from that what the load times might be.
 
Well if we look at the pieces and compare to this gen, and PC running the same game on faster CPUs, it should be possible to get a ball-park estimation. In short though, let's say the SSD removes that bottleneck and we just have CPU to worry about. Something like GTA is doing a lot of CPU work and taking minutes to load; we know that's not all storage reads because other games fill the RAM in much shorter time. So 300 seconds for GTA V, say. If PS5's CPU is 10x faster in that workload, that'll be 30 seconds on PS5, loading the same content and world complexity as GTA V.

I can't see any realistic way to think all games, even those taking minutes to load on faster CPU PCs, will get down to 2 seconds on consoles.

This table from DF shows load times for games where the data was already cached in RAM

View attachment 3746

That must be CPU workload on an i5 6600K.

The only wild-card is how steaming-friendly load structures develop, if partial loads are implemented across all games so 1-2 load gets you up and running, and then in the background the rest of the game is loaded. But that's more about game design than storage speed.

Why not? If you can quickly resume in a few seconds. Why should the initial load of a console game be any different?

If games can quickly load from a save state then why not have a default save state as part of the install? The default save state is the beginning of the game.

This is not as practical on PCs but as of now consoles only come in one configuration per platform.
 
Last edited:
Load times are notoriously unoptimized. It’s a thing I’ve seen devs readily admit to. I expect there are a lot of optimizations left on the software side, and titles that load slowly will leave a bad taste for users, so devs will spend more time on it.
 
Load times are notoriously unoptimized. It’s a thing I’ve seen devs readily admit to. I expect there are a lot of optimizations left on the software side, and titles that load slowly will leave a bad taste for users, so devs will spend more time on it.

Potentially with all games having long load times current gen, especially open world games maybe it didn't matter if your game took 90 seconds and another took 70 seconds so not much effort was put into it.

If next gen a game loads in 3 seconds and another loads in 30 that would be much more noticeable to the user so maybe loading will get more effort put into optimising it.
 
Back
Top