Digital Foundry Article Technical Discussion [2023]

[Shifty Geezer]

i think Richs PC still uses a SSD here. They said in the video that they would come back to a Video with him using a hdd instead..

No, they point out explicitly it's running from a PS4 HDD inside his PC.

The funniest bit is when Rich tried running the game off an actual PS4 hard drive on a machine just above min spec!

Edit: In contrast, the main video is Rich on SSD. This tweet is showing the difference between SSD performance and what an HDD delivers.

 

[Shifty Geezer]

I think John pointed out somewhere that this PS4 launch HDD video is on a PC using only 8GB of ram, so there'll be a lot of page file thrashing making this a lot worse.

Yes, he xed*:

As I note in the video, I had a much better experience with a 7200 RPM HDD in my 12900k PC...but I also have 64gb of ram which helps a lot (though it still pauses much longer than an SSD).

* exed? exited? he x: ??

 

[Globalisateur]

Why compare PC max settings against PS5 RT 60fps mode? Why not use the 40fps fidelity mode? This part I don't understand.

View attachment 9275

Left Alex maxed out PC, right John PS5 Performance mode

This game has some serious texture loading issues on PC...

Like Spider-man running on PC. NIxxes use partial textures loading on PC version as it clearly can't load new assets as fast and with such low latency than PS5 I/O.

 

[Parrish]

I think John pointed out somewhere that this PS4 launch HDD video is on a PC using only 8GB of ram, so there'll (probably) be a lot of page file thrashing making this a lot worse.

16GB of ram and a decent 7200 rpm drive and this should be a lot better, though still well behind even a SATA SSD.

John wrote somewhere on Twitter that he "had a much better experience with a 7200 RPM HDD in my 12900k PC...but I also have 64gb of ram which helps a lot (though it still pauses much longer than an SSD). This low spec system only had 8gb of ram thus it pages the HDD like crazy."
Edit : Sorry, I hadn't see that Shifty Geezer already had cited the post

 

[Dictator]

And even with the flawed comparison PS5 side (right) looks noticeably sharper. I think Insomniac reconstruction solution is here better than DLSS. Aren't some of those blurry surfaces typical DLSS side effects ? I am not talking about the obviously half loaded textures like on the ground.

Maybe they should use higher quality DLSS or native resolution at stable 40fps on PC...which could be now fairly compared against the pristine PS5 40hz mode?

It is over sharpened on PS5 - easy to See. The AF is broken on PC in this Shot.

I purposefully turned the Games shaprening down to 0 and it defaults to 10 with DLSS on

 

[pjbliverpool]

And even with the flawed comparison PS5 side (right) looks noticeably sharper. I think Insomniac reconstruction solution is here better than DLSS. Aren't some of those blurry surfaces typical DLSS side effects ? I am not talking about the obviously half loaded textures like on the ground.

Maybe they should use higher quality DLSS or native resolution at stable 40fps on PC...which could be now fairly compared against the pristine PS5 40hz mode?

Yes as Alex says - on PC the game sharpness is a toggle. I too noticed that it didn;t look as sharp on PC in the video but good to have it confirmed that is down to the sharpness setting.

Certainly the image stability was clearly better on PC in the early close ups they showed in the video image quality comparison section.

But for sure, AF and textures are broken.

@Dictator I think you guys were a bit lenient on the state of the port tbh. Between what you showed there, the AMD RT issues, and what I presume from that short clip of the 2070S is pretty horrible performance on that class of system, I'm not convinced this is in particularly better shape than some other recent big disappointments. I would say seeing how well textures hold up on an 8GB GPU might be a good indicator of that but given how they hold up on a 24GB GPU that may be a moot point!

 

[Dictator]

And even with the flawed comparison PS5 side (right) looks noticeably sharper. I think Insomniac reconstruction solution is here better than DLSS. Aren't some of those blurry surfaces typical DLSS side effects ? I am not talking about the obviously half loaded textures like on the ground.

Maybe they should use higher quality DLSS or native resolution at stable 40fps on PC...which could be now fairly compared against the pristine PS5 40hz mode?

To say the least, DLSS kind of owns every image technique in this game. IGTI does decidedly beat FSR 2.1 here too.

@pjbliverpool - my main video coming on friday is pretty no non-sense. I say it feels unfinished and unpolished within the first few mins there. Sony pushing it out the door too early. Nixxes are very talented and know how to do good things, so many things not right at launch tells me the time table was bad.

 

[Dictator]

Nice but we need comparisons against PS5 40hz mode!

Why the 40 FPS Mode and Not the 30 fps one? Also only Rich has the possibility to record 120hz modes (40 FPS) at 4K with a preproduction professional grade capture card. I do not own one, like most humans on earth.

 

[davis.anthony]

A Digital Foundry video of a console game ported to PC and #stutterstruggle wasn't mentioned once

 

[Shifty Geezer]

Primarily, yes. In the beginning. But Sony's innovations will carry gaming tech towards this direction of how data is managed. Which leads me to your follow up question which is a bit misguided - "How many?" is the wrong question to ask. Just know that more and more games will be made around this design philosophy until it becomes the norm. But PlayStation is spearheading all of this.

Spearheading? The same way ID Tech 5 spearheaded streamed textures in 2007, but they still aren't the norm 15 years later? I'm pretty sure by the time and notable number of games get to working with completel memory refreshes every second, PS5 will be long gone.

 

[Shifty Geezer]

is this still possible on PC ?

The photomode showing all the surroundings? If not, data could be optimised on PC to just what's ahead and save a signfiicant amount of PS5's loading.

 

[I'mDudditz!]

Spearheading? The same way ID Tech 5 spearheaded streamed textures in 2007, but they still aren't the norm 15 years later? I'm pretty sure by the time and notable number of games get to working with completel memory refreshes every second, PS5 will be long gone.

This philosophy and application will outlast the PS5.

 

[Cyan]

 

[Overstander]

100TB games apparently.

how big would a 100GB Game be completely uncompressed?
and also - its not like a game has lets say 100gb of data and the Console /PC would read it as a book page by page and then at some point when you finished the game , beat the last boss a hundert GB went through the system...
It is an interactive piece of software. Its all variables That chair, this enemy with this assets around them from this viewpoint is moving towards that direction while player moves backwards. VS The other chair, 3 differerent enemys but player moves through a door. All will create difffrent loads , different mix of assets. Its loaded and unloaded aas needed. That creates all kinds of traffic on the System..
The L2 Cache of GPUs usually have over 1000GB/s bandwidth.. that would mean the game would go through in les than a second. finished and done - according to your logic!
But of course its the myriad of diffrent floating point calculations that produce maybe 500000GB of internal traffic while you play the game.... WTF..Dude why is that so difficult to understand for you? You brought that topic up now several times...

 

[Cyan]

I think John pointed out somewhere that this PS4 launch HDD video is on a PC using only 8GB of ram, so there'll (probably) be a lot of page file thrashing making this a lot worse.

16GB of ram and a decent 7200 rpm drive and this should be a lot better, though still well behind even a SATA SSD.

yup, RAM can play a role. A 64GB RAM PC with a 4090 and a regular HDD.

https://twitter.com/i/web/status/1684231099356946434

is this still possible on PC ?

maybe DF can shed some light into that in a future video.

 

[pjbliverpool]

how big would a 100GB Game be completely uncompressed?

I said 100TB, because that's the kind of size it would take to support the fanciful idea of VRAM being completely refreshed every 1 second of gameplay. And since the average Kraken + RDO compression level is 2:1. 200TB.

and also - its not like a game has lets say 100gb of data and the Console /PC would read it as a book page by page and then at some point when you finished the game , beat the last boss a hundert GB went through the system...
It is an interactive piece of software. Its all variables That chair, this enemy with this assets around them from this viewpoint is moving towards that direction while player moves backwards. VS The other chair, 3 differerent enemys but player moves through a door. All will create difffrent loads , different mix of assets. Its loaded and unloaded aas needed. That creates all kinds of traffic on the System..

PS5 has ~10GB of VRAM. a 100GB game (around the max realistic game size) would be around 200GB worth of assets uncompressed in memory. So you are suggesting that every second of gameplay will result in a the scene needing to be refreshed with 1/20th of the games entire content. It doesn't matter in how many combinations you mix up the assets, you would have seen the entire game content within 20 seconds, or you would be seeing the same assets over and over again (mixed up differently perhaps, but the same models, textures etc... nonetheless) repeatedly for the entire duration of the game.

It's not remotely realistic.

The L2 Cache of GPUs usually have over 1000GB/s bandwidth.. that would mean the game would go through in les than a second. finished and done - according to your logic!

But of course its the myriad of diffrent floating point calculations that produce maybe 500000GB of internal traffic while you play the game.... WTF..Dude why is that so difficult to understand for you? You brought that topic up now several times...

You've just answered your own question as to why data loads from storage cannot be compared to graphics operations taking place in the GPU's caches and registers. Unless you're tying to suggest the game can use the SSD as a render target (lol).

I don't even understand why there is all this focus on the PS5 IO speed. It's already been established that IO speed isn't a bottleneck on PC. The reason we're not seeing significantly faster portal transitions on PC is much more likely either a CPU limit because the PC is doing more work on the CPU side, or a hard coded game engine limit.

 

[Overstander]

I said 100TB, because that's the kind of size it would take to support the fanciful idea of VRAM being completely refreshed every 1 second of gameplay. And since the average Kraken + RDO compression level is 2:1. 200TB.



PS5 has ~10GB of VRAM. a 100GB game (around the max realistic game size) would be around 200GB worth of assets uncompressed in memory. So you are suggesting that every second of gameplay will result in a the scene needing to be refreshed with 1/20th of the games entire content. It doesn't matter in how many combinations you mix up the assets, you would have seen the entire game content within 20 seconds, or you would be seeing the same assets over and over again (mixed up differently perhaps, but the same models, textures etc... nonetheless) repeatedly for the entire duration of the game.

It's not remotely realistic.

you seem not to understand ... lets stay in your example - lets say from a raw uncomporessed 200gb game lets say one level has 35gb worth of data in assets textures and such. Its not that all of that 35gb is needed from the first second you start the level. Loaded in memory is what you see and maybe some additional 10% more around the screen edge to give the I/O system a bit of reaction time to accomodate for player movement. so lets say that viewpoint makes up for 8-10gb of data . If you turn around and other assets that are durrently missing are loaded and those who are not needed anymore go out. you turn back and the it starts form beginning again. So , if you would run through this level for an hour or so. The swapping of assets in and out would have caused maybe already 200gb worth of traffic.
Cerny did not lie! The tech talk was a honest presentation of mainly how they perfected their I/O Block. The slides made sense! But you seem to think that PS5s vram ( of wich there is a higher number available to the GPU than your mere 10GB btw but hey ) needs to full swap in and out 16GB at all times just because cernys slides implied that it is technicaly possible. And it still is. It stil stands - if Cerny the mein System architekt from now 3 succesfull consoles says it in his presentation slides - then it is honest and real and no marketing bullshit you hear??

You've just answered your own question as to why data loads from storage cannot be compared to graphics operations taking place in the GPU's caches and registers. Unless you're tying to suggest the game can use the SSD as a render target (lol).

I don't even understand why there is all this focus on the PS5 IO speed. It's already been established that IO speed isn't a bottleneck on PC. The reason we're not seeing significantly faster portal transitions on PC is much more likely either a CPU limit because the PC is doing more work on the CPU side, or a hard coded game engine limit.

It was an example how numbers can produce alot more traffic than what they look like. And oh yeah? Nothing was established - for now it is established only that what i was saying and many others - that RTX 20xx cards are no match for PS5 in Games that REALY are programmed with only PS5 in Mind. All your fancy games so far from third party devs that probably for money sakes used only the high Level API on PS5 and that performed better on a RTX 2060 - their times are over. I here say it again even a RTX 2080ti will not be able to achieve similar to PS5 results when it is taxed on the same time with 5-9Gb/s GPU decompression on comming PS5 Exclusive Ports.

The infamous DF experiment wich revealed that Ratchet & Clank on PS5 cannot use more than 2Gb/s streaming because they taped of most of a M2 SSD contacts tolimit its bandwith was a big boomerang at the end. Yeah sure it was fun to dunk on Cernys console "oh look the fancy SSD is not actually needed yadayada" And look now - what havok that mere 2GB/s streaming causes already even with DS GPU decompression enabled - or should is say because it is enabled? Only good modern GPUs can handle the Game right now on PS5s settings.
Your arrogancy will make you eat crow later on. Or you eating it already i dont know.

 

[Flappy Pannus]

I don't even understand why there is all this focus on the PS5 IO speed. It's already been established that IO speed isn't a bottleneck on PC. The reason we're not seeing significantly faster portal transitions on PC is much more likely either a CPU limit because the PC is doing more work on the CPU side, or a hard coded game engine limit.

Hell, the difference from an NVME to a HDD in this is usually just a little over twice as fast (if even that) with only the last world showing a larger difference. I don't know what texture settings they were using in this video, but the user has a 12GB 4070ti so I would assume very high, but not confirmed.

The game is just clearly not I/O bound in any meaningful way, at least from storage. There's another bottleneck.

 

[I'mDudditz!]

I don't even understand why there is all this focus on the PS5 IO speed. It's already been established that IO speed isn't a bottleneck on PC.

No it hasn't. All of your confusion will be cleared once you realize SSD speed /= I/O speed.

 

[Karamazov]

He got those times with an HDD because he could compensate with higher amount of ram, and ram is faster than even the fastest SSD.