• 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.

Inside Ratchet and Clank: Rift Apart on PS5: The Insomniac Technology Breakdown

Corndog

Banned
Did you even read my post before replying to it?

The scenario(s) when something becomes CPU bound is already explained. This is consistent across pretty much all games, hence why CPU benchmarks for gaming are typically conducted under 720p.
Do you read posts? I asked why would THIS GAME be cpu bound? Please explain to me why this game is bound while others are not even on weaker hardware. The only way to find out is from the developers since none of us have access.
 

Corndog

Banned
Zen2 desktop chips aren't bad, they're only like 10% slower than Skylake in terms of single-threaded performance. However, the consoles aren't using the desktop chips, but the mobile ones. Hence, only half the L3 cache per CCX, which is a 15% performance hit right there. Furthermore, they're saddled with GDDR6, which has twice the latency of regular DDR4 (and close to 3x that of hand-tuned Samsung B-die @ 55ns). So not only are they going to have a lot more cache misses than a desktop Zen2 CPU, they're going to be waiting on the RAM much longer each time. And games, with their branch-heavy workloads, produce a lot of those. Lastly, the PS5 CPU's max frequency is 3.5GHz versus somewhere between 4.2 and 4.5GHz for most of Zen2's desktop lineup.

All in, you're looking at roughly half the single-threaded performance of the PS5 CPU (and the Series X) compared to a Zen2 CPU in a gaming rig.
Is it really that severe? If it really is that bad they should have just ate the cost of the cache. Or gone with less cores and more cache.
 

ABnormal

Member
I wonder how much RAM would be needed to run R&C on PC accomplishing the same feats. The entire game in RAM? 3/4 of it with a standard SSD?
Probably having just the worlds that have to be played on a particular "level"(I don't know if it's appropriate to divide the game in levels, though) in RAM should be fine. On ps5 ssd the assets are compressed, so either the pc cpu would have to allocate enough power to decompress it on the fly (equivalent of nine zen 2 cores power on ps5), or the RAM should be big enough to have the space needed for uncompressed assets.
 
Last edited:

Kholinar

Banned
Do you read posts? I asked why would THIS GAME be cpu bound? Please explain to me why this game is bound while others are not even on weaker hardware. The only way to find out is from the developers since none of us have access.
You're not getting it at all. In order to go from 60fps to 120fps, the CPU would need to process instructions 2x faster. The CPU is the bottleneck for achieving 120fps, not the GPU.
 
Sure. Computer motherboards support up to 128GB of DDR4 RAM. Worst case scenario the whole game could be loaded into RAM but I doubt that would need to happen.
How long will it take to load the whole game in ram, I think its possible on pc vut requires tweaking and optimizations
 
Sure. Computer motherboards support up to 128GB of DDR4 RAM. Worst case scenario the whole game could be loaded into RAM but I doubt that would need to happen.
Theoretically 64bit computers support up to billions of gb of ram but pc motgerboards have always been the bottleneck.
 

jroc74

Phone reception is more important to me than human rights
Good to see confirmation that they’re pulling data on the fly from the SSD for the pocket dimensions. There was some argument about that in previous thread.
Yup, I mentioned it was a showcase for the SSD before this DF video was posted and I had several ppl disagree, one damn near upset I could suggest such a thing. On numerous forums, lol.
 

Mister Wolf

Member
Because decompression on pc uses cpu its not fast enough as ps5 or series x because they have hardware decompressors that fo the work of aroind 11 zen 2 cpu cores

The Cpus in the PS4 and Series X are 2 generations behind the 5000 series Ryzens currently for PC. You're telling me those much more powerful and performant Cpus don't have the overhead to take on the task of decompression while handling everything that the cpus inside the consoles do for R&C?
 
Last edited:
The Cpus in the PS4 and Series X are 2 generations behind the 5000 series Ryzens currently for PC. You're telling me those much more powerful and performant Cpus don't have the overhead to take on the task of decompression while handling everything that the cpus inside the consoles do for R&C?
Yes they can do decompression but not with the overhead your imagining, theres a reason nvidia put hardware decompression on their gpus and direct storage is supposed to help by throwing data straight from ssd to gpu bypassing ram thats what I think, and mind you the ps5 can decompress 22gb of data per second at max and I think this game is doing that since its 33gb install size compressed but the actuall size is 200gb.

As I said it could be done but not so simple or it could be simple who knows, I even doubt the series x can pull this easily, they are streaming levels during camera shots, streaming on portals, streaming assets that are just behind the player when you turn around, could be done on other hardware but if its not optimised youll see alot of popin.
 

Mister Wolf

Member
Yes they can do decompression but not with the overhead your imagining, theres a reason nvidia put hardware decompression on their gpus and direct storage is supposed to help by throwing data straight from ssd to gpu bypassing ram thats what I think, and mind you the ps5 can decompress 22gb of data per second at max and I think this game is doing that since its 33gb install size compressed but the actuall size is 200gb.

As I said it could be done but not so simple or it could be simple who knows, I even doubt the series x can pull this easily, they are streaming levels during camera shots, streaming on portals, streaming assets that are just behind the player when you turn around, could be done on other hardware but if its not optimised youll see alot of popin.

So there is no way of knowing how fast a current Ryzen or Intel CPU can decompress.
 
So there is no way of knowing how fast a current Ryzen or Intel CPU can decompress.
Well never know until they port this game to pc all I know is hardware decompression is magnitudes faster than cpu because your cpu is already busy when playing a game you dont want to bother it with decompression, even cyberpunk stutters when streaming data from storage during gameplay in pcs and xbox but it didnt on ps5. That shows you something.
 

Mister Wolf

Member
Well never know until they port this game to pc all I know is hardware decompression is magnitudes faster than cpu because your cpu is already busy when playing a game you dont want to bother it with decompression, even cyberpunk stutters when streaming data from storage during gameplay in pcs and xbox but it didnt on ps5. That shows you something.

 
Well its not just cyberpunk the fact that games just load faster on ps5 than pcs with faster storage tells you something because decompression happens in loading screens aswell even in control on pc and xbox frames drop when streaming data from storage in during gameplay.
 

freefornow

Member
You would have to basically remake the whole game.

Jim Carrey Chance GIF
 

sinnergy

Member
So they only load , parts of the new level , in rifts .. as expected, so it doesn’t load 16 gigs . As I talked about in the past here, and stream the other parts in when needed.
 
Last edited:

Pedro Motta

Member
The Cpus in the PS4 and Series X are 2 generations behind the 5000 series Ryzens currently for PC. You're telling me those much more powerful and performant Cpus don't have the overhead to take on the task of decompression while handling everything that the cpus inside the consoles do for R&C?
I have a Ryzen 5950x, 128gb ram, 2x WD SN850 pci-e 4.0. Never seen my PC load anything in one second, not even close.
 
I guess it could be compressed but it would probably require quite a lot of CPU juice to constantly decompress it on the fly, but I guess it possibly could work if you have a very good CPU.
It would work a lot better in upcoming GPUs with dedicated decompressors.
More nonsense I see, look the PS5 is a fine machine but all this it can only be done on PS5 is starting to become tiresome.
 

Dr Bass

Member
The Cpus in the PS4 and Series X are 2 generations behind the 5000 series Ryzens currently for PC. You're telling me those much more powerful and performant Cpus don't have the overhead to take on the task of decompression while handling everything that the cpus inside the consoles do for R&C?
I don’t think you understand how much slower general purpose cpus are vs custom dedicated hardware for … anything. The PS5 is using custom dedicated silicon for its data decompression.

CPUs are good at executing general purpose code but again, don’t hold a candle to dedicated hardware. Why do you think GPUs even exist?
 

Mister Wolf

Member
I have a Ryzen 5950x, 128gb ram, 2x WD SN850 pci-e 4.0. Never seen my PC load anything in one second, not even close.

You don't have the majority of any game you've ever played stored in the system RAM. My conversation was specifically about whether the desktop CPU's could take on the task of decompressing swiftly.
 

Mister Wolf

Member
I don’t think you understand how much slower general purpose cpus are vs custom dedicated hardware for … anything. The PS5 is using custom dedicated silicon for its data decompression.

CPUs are good at executing general purpose code but again, don’t hold a candle to dedicated hardware. Why do you think GPUs even exist?

Got any numbers for how fast a CPU can decompress data?
 

Dr Bass

Member
More nonsense I see, look the PS5 is a fine machine but all this it can only be done on PS5 is starting to become tiresome.
Nah it’s totally true. No other hardware in existence right now can do what it does in terms of the SSD. After playing Ratchet it’s obvious Cerny (and Epic) were telling the truth. The dimension shifting in the game is astounding. I’ve never seen multiple entire environments loading in under a second. And go watch the tech interviews with Insomniac.

It has nothing to do with the CPU or GPU and everything to do with the IO and system architecture.
 

Panajev2001a

GAF's Pleasant Genius
You don't have the majority of any game you've ever played stored in the system RAM. My conversation was specifically about whether the desktop CPU's could take on the task of decompressing swiftly.
Depends on which CPU: 16 cores, 2 treads per core, 5 GHz certainly has a better shot dedicating 8+ cores to it and leaving the rest to handle game logic. I think GPU based decompression and/or custom HW decompressor will enter the PC arena at some point too. They will also need OS support and tooling and that will take some time to stabilise.

I think we underestimate the benefits of not having to add HW components and modify the OS while trying to deploy back to billions of devices in a codebase which has necessarily quite a bit of abstractions (thus limiting performance the Desktop CPU’s buy you). Let some for the current software I/O overhead for transferring the data you want to decompress.
 
Last edited:

Dr Bass

Member
Got any numbers for how fast a CPU can decompress data?
Dude don’t be obtuse. I could write some code and instrument it for you but I’m not going to because that’s stupid.

Its like asking me to write a software rasterizer to compare to a GPU. It’s going to get blown out of the water. That’s why dedicated silicon gets made. Just go read a little bit on the topic and quit trying to be clever with dumb gotcha questions that look ridiculous to people who work in the field.

It’s like asking to prove that a plane can beat a bicycle in a race.
 
Nah it’s totally true. No other hardware in existence right now can do what it does in terms of the SSD. After playing Ratchet it’s obvious Cerny (and Epic) were telling the truth. The dimension shifting in the game is astounding. I’ve never seen multiple entire environments loading in under a second. And go watch the tech interviews with Insomniac.

It has nothing to do with the CPU or GPU and everything to do with the IO and system architecture.
It can be done on PC, let's just stop, you're just spouting PR talk.
 

Panajev2001a

GAF's Pleasant Genius
So there is no way of knowing how fast a current Ryzen or Intel CPU can decompress.
We have less than 1 GB/s on a Core i7 running at 3.4 GHz (single threaded):
On the "Silesia" test set, the performance is :
Kraken : 4.05 to 1 compression ratio, 919.8 decode mb/s
zlib9 : 2.74 to 1 compression ratio, 306.9 decode mb/s
lzma : 4.37 to 1 compression ratio, 78.8 decode mb/s
(speed measured on a Core i7-3770 3.4 Ghz , x64, single threaded)
Improving compression ratio of the codec would also slow it down a bit (see Leviathan, but I do not think their HW decompressor support that).

PS5 can push it up to 22 GB/s, XSX pushes BCPACK+zlib decoder to over 6-7GB/s.

Today on PC there is a problem getting data that fast from the disk without also taking a good deal of CPU usage too (no Direct Storage yet too).
 
Last edited:

Mister Wolf

Member
Dude don’t be obtuse. I could write some code and instrument it for you but I’m not going to because that’s stupid.

Its like asking me to write a software rasterizer to compare to a GPU. It’s going to get blown out of the water. That’s why dedicated silicon gets made. Just go read a little bit on the topic and quit trying to be clever with dumb gotcha questions that look ridiculous to people who work in the field.

It’s like asking to prove that a plane can beat a bicycle in a race.

The only thing I've read is Kraken is equivalent to 9 Zen 2 cores. The 5000 series is not Zen 2 its Zen 3. I have no idea of knowing how much better it is at decompression. Gamers Nexus uses 7 Zip to test CPUs but I don't have much understanding on the topic. If you think my question was some kind of gotcha so be it. Don't bother me.
 

kyliethicc

Member
I don't think compression would be an issue for a Ryzen 3700X let alone a Ryzen 4000 or 5000 series CPU.

The Cpus in the PS4 and Series X are 2 generations behind the 5000 series Ryzens currently for PC. You're telling me those much more powerful and performant Cpus don't have the overhead to take on the task of decompression while handling everything that the cpus inside the consoles do for R&C?

So there is no way of knowing how fast a current Ryzen or Intel CPU can decompress.

Got any numbers for how fast a CPU can decompress data?

The PS5 has a customized version of the 3700X. 8 cores 16 threads Zen 2 CPU.

Cerny then said the dedicated I/O silicon in the PS5's SoC are the equivalent of over 10 more Zen 2 CPU cores in performance.

So its as if the PS5 has like 18 Zen 2 CPU cores worth of performance. Good luck doing all that on a PC with only 8 cores.

Now that the PS5 and XBS have Zen 2 CPUs, PC gamers will not be able to run new next gen only games at crazy high framerates. Because these games won't be designed for the XBO & PS4's weak old Jaguar CPU. Lots of PC gamers only have 6 core CPUs. And no dedicated hardware for I/O either.

To play R&C RA on PC at same quality as PS5, I'd bet more than 8 Zen 2 cores would be required.
 
Last edited:

Dr Bass

Member
Found this as reference on the custom hardware.

“The PS5 uses a derivative of RAD Tool's potent Kraken decompression technology, which offers 10% better compression--that's about 10% more installations or data on a Blu-ray disc. Without the dedicated decompressor, it would take 9 Zen 2 CPU cores to decompress Kraken-level data.”

So there you go. Assuming this is correct a CPU core is nearly ten times slower than the dedicated decompressor. And the you’re not factoring in the difficulty of writing multithreaded code for processing serialized data, not to mention robbing the game of being able to process game code if you wanted to even attempt it. It’s a complete non starter to try and program your game this way.

It’s a real advantage for the PS5, there is just no getting around it.
 
Last edited:

Three

Member
It's the same asset with smart technology seamlessly transitioning between the high and low quality pieces that make up the asset, depending on cinematic vs gameplay.

Was VFXVeteran claiming otherwise? That there were LITERALLY two separate assets? I think it's common knowledge that they're not going to render things at LOD0 when you're pulled so far back... even if it's the same 'asset'.
So what is the point of trying to mention that there is a difference between cinematic and gameplay? You would see the exact same detail if you got close in gameplay.

Why the distinction of cinematic vs gameplay? Why not just they use the same assets but the camera isn't close enough to see it.
 

Pedro Motta

Member
You don't have the majority of any game you've ever played stored in the system RAM. My conversation was specifically about whether the desktop CPU's could take on the task of decompressing swiftly.
Why are we taking fairy tales and unrealistic scenarios? The flexing has a limit man. Why buy 800 $ worth of RAM to alocate and run games like a 400$ console with only 16GB RAM.

The Ryzen CPU would be able to decompress swiftly but sacrificing a lot of cores to do so (about 10 cores, as stated by one of the creators of Kraken).

Still not sure about the latency from the system RAM to the GPU RAM tho.
 

Mister Wolf

Member
Why are we taking fairy tales and unrealistic scenarios? The flexing has a limit man. Why buy 800 $ worth of RAM to alocate and run games like a 400$ console with only 16GB RAM.

The Ryzen CPU would be able to decompress swiftly but sacrificing a lot of cores to do so (about 10 cores, as stated by one of the creators of Kraken).

Still not sure about the latency from the system RAM to the GPU RAM tho.

This wasn't about If you could afford it or if it was worth it. We all make different amounts. The question was could it be done?
 

hoplie

Member
Also, on PC you have to copy the data several times afaik, which is not needed on PS5. You have to take that into account, too.
 

Md Ray

Member
I don't think compression would be an issue for a Ryzen 3700X let alone a Ryzen 4000 or 5000 series CPU.
What? You mean decompression? They require over 10 cores, and according to NVIDIA, it was 24 cores... You need a Threadripper worth of CPU cores just for decompression work. On top of that, gaming has its own CPU requirements, gameplay, physics, handling draw calls to name a few.
 
Last edited:

Md Ray

Member
Got any numbers for how fast a CPU can decompress data?
Yeah.

From Ampere video presentation:
ksXSyzD.png


"Over 20 CPU cores."
OsszvXy.png


From Ampere whitepaper:
Look at the CPU requirement when decompressing from 7GB/s NVMe according to NVIDIA:
Pbo7rYh.png

Iv7KJor.png


Level load time using 24 cores CPU + 7GB/s NVMe: 5 seconds
which is still significant compared to 1.5 sec with non-CPU decompression, a -70% reduction:
U0SesRJ.png


Here's a level load time on PS5 with HW decompression + 5GB/s NVMe for reference:
Y5Sq1gK.jpg

I don't think compression would be an issue for a Ryzen 3700X let alone a Ryzen 4000 or 5000 series CPU.
You can see how intensive of a workload this is with a conventional CPU. When decompressing alone takes up all of Threadripper's 24 cores - what do you think will be left for the game itself? And an 8 core 3700X is far, far from 3960X. Not even the 5000 series is enough.
 
Last edited:

Panajev2001a

GAF's Pleasant Genius
Yeah.

From Ampere video presentation:
ksXSyzD.png


"Over 20 CPU cores."
OsszvXy.png


From Ampere whitepaper:
Look at the CPU requirement when decompressing from 7GB/s NVMe according to NVIDIA:
Pbo7rYh.png

Iv7KJor.png


Level load time using 24 cores CPU + 7GB/s NVMe: 5 seconds
which is still significant compared to 1.5 sec with non-CPU decompression, a -70% reduction:
U0SesRJ.png


You can see how intensive of a workload this is with a conventional CPU. When decompressing alone takes up all of Threadripper's 24 cores - what do you think will be left for the game itself? And an 8 core 3700X is far, far from 3960X. Not even the 5000 series is enough.
The nVIDIA presentation includes I/O overhead on top of decompression (see the numbers for Kraken software decompression above), but it is quite interesting and both give a good understanding on the amount of work these next generation consoles are doing with dedicated I/O HW.
 

Jack Uzi

Banned
He posts on resetera right? So yeah... you could fill 10 football fields with completely insane posts from there. If you seen his twitter mental break downs, the guy has some issues.

He doesn’t, only Alex does. He got chased from there too.

Dark sent a DM to KingThrash about how the Xbox fans were a million times worse than the PS ones. They’re all nutballs those these fanboys.
 
Last edited:

Md Ray

Member
He posts on resetera right? So yeah... you could fill 10 football fields with completely insane posts from there. If you seen his twitter mental break downs, the guy has some issues.
Any sane person would have mental break downs when it gets to the point of shit getting sent to your home by these rabid fanboys and chasing you each time you didn't praise their favorite console.
 
Top Bottom