Digital Foundry Article Technical Discussion [2022]

[Overstander]

so i also reached the 10000 charakters limit

But almost all games last generation were capable of running perfectly well on a dual core CPU which despite having roughly 4x more performance per core vs the console Jaguars, also have 4x fewer cores, so the overall performance is roughly equivalent. So it's not like PC's were throwing enormously greater amounts of CPU power at the problem to overcome this performance edge you speak of. Naturally a little more CPU power will always be needed on the PC side though to overcome the thicker API's.

Thats my Point regarding the Jaguar - it is a Netbook CPU. I think DF used one of these strange Xbox One APUs for Desktop PCs a while ago with Cyberpunk and showed that it would not be a valid gaming CPU on PC.
Only the Fact that due to much more efficient APIs on consoles made it possible to use a netbook CPU in the last Gen. Then there was no multithreaded Engine out there (in a AAA Third Party Game anyways) until Assassins Creed Odyssee i think. If i recal that was the first time the usual Suspect CPUs with 4 or less cores failed to deliver the PS4 Expirience.
Also the IPC of a Jaguar Core is not good - because it was not intended to end up in a Gaming machine. Thats way many 2 Core CPUs could outperform a PS4. It is like a said. That time the PC Community had much better HW already when it the Gen started.
This Time - not so much. Not the majority of PC gamers.

What cost specifically? DirectStorage will remove the majority of the decompression load from the CPU. There is no CPU cost to this, it's a massive CPU saving. The GPU will be doing the work but by all reports the performance impact there will be negligible, or in Nvidia's words - barely measurable. It's not as if the spare GPU power isn't available to absorb this negligible impact.



As above, this is a false comparison, Richard was quite clear about that in his video and I'm sure would be disappointed to see it being twisted in this way.

I meant the Cost of render time on GPUs. I simply doubt Nvidias statement of RTX I/O being so lean on a GPU. Like i wrote obove - i want to see something. At least a Demo of some sorts. Should not be so difficult. And to adress the "2Tf sacrifice are enough to catch up or even overtake PS5s I/O capabilitys" - i think that came out of another Thread wich was linked here some while ago. I absolutly tink that for one - on RTX 20xx nobody has 2Tf to spare when it renders a lets say 2nd wave PS5 Exclusive. I think trying to mimik PS5s I/O throughput is much more stressfull to a GPU then most people think. It is cute that Jen-Hsun Huang thinks that simply flipping a switch on a RTX 20xx / 30xx GPU is sufficient to mimik PS5 I/O Capabilitys without adding up a kinds of latencys in the CPU/ GPU correspondance ( wich is already much higher than on Consoles because UMA instead of hUMA ) .

People seem to forget that PS5s I/O is not only the Kraken Unit. There is more HW there wich is not present on PC.

RTX-IO, or more generally GPU based decompression is there to free up CPU cycles by moving that activity onto the GPU. Nvidia have stated that the impact of that on the GPU is "barely measurable", which stands to reason given that GPU's have orders of magnitude more compute power than CPU's. The RTX 3070 for example has roughly double the compute performance of the PS5 (20TF vs 10TF). That doesn't translate directly into game performance but if the GPU decompression is going to use that compute, then there is more than enough on tap with any Ampere to spare for this.

Now putting aside the IO point and talking about raw performance, a 3070 is roughly equivalent to a 2080Ti. All benchmarks we've seen to date put the PS5 performance at between 2060s and 2080 level depending largely on the use of RT, with more RT pushing PS5 down that stack. RT will obviously increase as the generation goes on and we see more current gen exclusives, so that performance advantage is likely to solidify over time. Spiderman is just the latest in a long line of examples of this with the 3060 (2070 level) offering an equivalent or better experience to the PS5.

Again i doubt the Nvidia Statement. When it would be so much more efficient on GPU ("barely measurable") why did Sony or Cerny not take this aproach when designing the PS5? It would be way smarter then to not develop the whole custom I/O shenanigens and go with a bit bigger GPU and do the Decompression there. Smarter because then , if a game would for some reason not rely much on decompression they could use the GPU then for better IQ and Detail Settings instead. Clearly here is something that Nvidia is not telling. And again - where are the demos?
Logic simply suggests that any major custom HW development that sums up many millions MUST be considered more efficient than using already existing (general purpose) tech, because all comitees and engineers involved in the process would have pointed out or even blocked further investigation in custom tech.

That's been addressed many times here. The game is only about 66GB uncompressed. It's inconceivable that it's loading several GB/s worth of data with every camera move or else the game would be over in a hand full of seconds.

i .. I dont want to come across like talking down to you but forgive me when i say that does not make any sense at all. A Game being 66GB does not tell you anything about floating point calculations or texture/asset swaps. There might be a lets say high quality Tree model that weighs in lets say 500Mb copiying that very model 10 times in and out memory already caused 5GB of traffic. Without the Game being "sucked empty" lol. Also why are you oblivious to the fact that any calculations of vectors or basicly any gamedata that goes through the GPU is adding up to several 100GB of internal traffic within the GPU. L1 and L2 caches have hundreds if not thousends of GB/s Bandwith.
https://www.beyond3d.com/content/reviews/55/7 <- This here suggests a mere 1087.74 GB/s Bandwidth for L1 alone on an old GTX 470...
Playing a game for one hour has caused probably 1Million GB worth of internal GPU traffic for all we know...

The game does load new areas very fast when a portal is used but that's simply a loading speed challenge. Something that could well be a little slower on PC at the moment until DirectStorage GPU decompression is available, but if that equates to 2 second portal transitions vs the current 1 second then that's not exactly a deal breaker, and could very likely be mitigated with additional pre-caching into the PC's larger memory pool. Outside of those portal transitions there's no reason to believe the game wouldn't perform similarly to Spiderman insofar as a 3060 class GPU being capable of providing an equivalent experience. I would actually expect the game to be more forgiving on the CPU side due to the much slower world traversal which likely requires more modest streaming and certainly far less frequent/significant BHV updates which are the main cause of the high CPU requirements in Spiderman.

The game swaps out the entire memory usage of a level in a second thats one thing. The other things is that Rift Apart indeed makes use of Cernys forecasted new memory paradigm of only 1sec of gameplay resident in memory.
Memory in high active use.
View attachment 6823
Rift Apart Method of Memory usge is mentioned here :

Interview: How Insomniac made "Ratchet & Clank: Rift Apart" look so good

The veteran studio says the PS5's SSD helped in some unexpected ways.

www.axios.com

We spend as much memory as we can on just the things you can see in front of you at this moment. And the more we can do that and learn how to do that, the more stuff we can cram in there on, like a per pixel by pixel, basis."

also here:

https://twitter.com/x/status/1408036947344384005

So that will be it for right now - sorry for the other people answered to me - i come back
Those big posts are quite exhausting - especially for a non native english speaker hrhrh

 

[PSman1700]

So first of all no intention of any plattform warring on my side. But a honest discussion must be possible without someone immediatly making use of "war" vocabulary , so please in all friendlyness - knock it off.
I did not try to say that a RTX3060/ 3700x Combo cant render the Game in same quality as a PS5. Spiderman on PS5 is more than just a mere patched PS4 Version , People are right here and i was wrong. But it still is a very early into the Gen first party PS5 Game that for sure does not leverage PS5 Intrinsics to an extend 2nd or 3rd wave PS5 Exclusives will do. Even Rift Apart is only tipping the Foot into the water when it comes to PS5 potential.
So the Question we need to ask is what cards will long into the Gen still be able to render PS5 Exclusive Titels in the same quality. I said a RTX 3070 will not be enough . Many disagree and that is fine. It will be intresting to see what comes out of it.

Thats why generally things like 'PS5 outperforms a 3070' and 'pc is in trouble' will be met with equal reactions. Those things are just not going to happen.
A 3060/3700x/nvme will be enough this generation if you can content with equal-to-console levels. Perhaps a stutter here and there. I also believe that we wont be getting too far ahead of Rift Apart, it didnt happen last gen and it probably wont this time around either.

<i believe it when i see it. Nvidia can talk all day long about their RTX I/O . PS5 is here and its concept has proven to be way way more advanced than the old way of doing it on PC. There are games out there that do stuff that is not possible today on PC. - The obvious Answer always is replyd to that point is :,,But PC can go always pure uncompressed and just have 32GB of mainram to make up for large filesizes". My Answer to that is - Where is it? When it is so easy why nobody does it? Why Star Citicen does not try to do it for example? Why none of the High End PC Games like Cyberpunk have a extra Detail Setting called "Hyper Ultra" wich would make use of such a possibility?

Cerny came up with a ingeniuos idea that Sony paid alot of millions for in development. They streamlined for the very first time the entire I/O Process and build a lot of custom HW. Nvidia on the other hand still needs to prove that a GPU busy with a AAA Game doing raytracing at 60 fps still has time left to decompress lets say 5GB/s. Where are the Demos for that? I did not see a single Demo from them proving their concept..

The PS5 concept has not proven itself, neither has the evil pc or the Xbox. Its too early to tell. Look, this Cerny is god and all and NV is dumb, its going to be met with equal reactions etc
Theres nothing on the PS5 neither on the PC doing 5gb/s of streaming from an nvme drive. You cant judge if NV is wrong or not about RTX IO, or that it is less advanced. It has not much to do with advanced, more to do with that the PS5 is using a different architecture. If we can get 14gb/s at the least using DS/GPU decompression, thats quite impressive for PCIE4.

In that sense this Gen is different than the PS4 one. 2013 you were fine with a little older HW than PS4 almost all of its Gen. Mainly due to Jaguar being used as a CPU and Games not using 8 Core CPU to their potential.

Is there a Chart or diagram that shows the power usage of a Laptop with similar enough HW to PS5 while it renders a game? Best would be the same game and said game should be a modern AAA Game. I think a PS5 should still win this if everything else is the same like Detail settings and resolution ect..

Your going to be fine with a 2070/3060, 6600XT and 3700x or better. Its generally older hw than the PS5 is.

I have a G15 asus with 5800h/3070m (130w gpu), 32gb and nvme. Its drawing below what the PS5 does, yet outperforms it in every game tried (even spiderman). The laptop has a screen, speakers etc aswell. A desktop is going to be much less efficient. Powerdraw from the wall, that is.

 

[PSman1700]

so i also reached the 10000 charakters limit

For me its that i cant seem to seamlessly copy and paste written content to the next page without quotes getting messed up. B3D transited to a new forum not long ago.

So that will be it for right now - sorry for the other people answered to me - i come back
Those big posts are quite exhausting - especially for a non native english speaker hrhrh

I hope your not writing these posts on a phone Il only react to what you quoted me on to keep things abit shorter and let pjb's quotes to himself then

 

[function]

That HUB review is quite illuminating. It seems extra cores do make little difference within the same architecture which is highly surprising given Nixxes statement about dedicating a whole core to decompression. On that basis I can only assume hardware decompression wouldn't do an awful lot for this game unless Nixxes were just simplifying and really meant roughly the equivalent of a core spread across all available cores.

The memory speed comparison is pretty staggering though! Over a third extra performance on a 12900K by going from DDR4 3200 to DDR5 6400!! Holy cow, I know what memory I'll be getting with my next CPU!

Dat Ryzen 2700X performance under raytracing tho.

It's getting murdered by the Ryzen 3 3300X (Zen 2) with only 4 cores.

Clearly the 2700X is the ideal CPU for comparing how RTX GPUs stand up to the mighty raytracing beast that is the RDNA2 based PS5.

That's also pretty illuminating with regards to the PS5's performance given the PS5 has access to very high bandwidth GDDR5 as it's system memory.

Internal BW for memory copies (if even needed) is probably very high on PS5 too.

The quad core Ryzen 3 5300G (Zen 3) loses out to the older quad core Ryzen 3 3300X (Zen 2). Part of this might be due to the double sized L3 on the older Zen 2 3300X, but I suspect a chunk of it is also down to the older Zen 2 3300X actually having full-on PCIe gen 4 while the newer Ryzen 3 5300G is somewhat hobbled with ye-olde PCIe gen 3. So the Zen 2 actually has twice the PCIe BW than this particular Zen 3 chip. As Nixxes said, this game does involve a lot of transfer from CPU memory into GPU memory.

 

[snc]

I think nobody watching this comparison will have doubts its proper remake

 

[function]

Again i doubt the Nvidia Statement. When it would be so much more efficient on GPU ("barely measurable") why did Sony or Cerny not take this aproach when designing the PS5? It would be way smarter then to not develop the whole custom I/O shenanigens and go with a bit bigger GPU and do the Decompression there. Smarter because then , if a game would for some reason not rely much on decompression they could use the GPU then for better IQ and Detail Settings instead. Clearly here is something that Nvidia is not telling. And again - where are the demos?

Sony have a fixed platform, and their decompression block can work for both the CPU and GPU. It removes the burden of implementation from developers. It probably lessens the impact of Sony halving the vector units in the CPU in some circumstances (I'm sure they had their reasons to do so, but the CPU vector units are gimped).

PC has lots of hardware vendors, almost limitless combinations of hardware and software, multiple OSes, at least two major APIs. A single hardware block with a single performance level with a limited selection of hardware accelerated decompression options is a great choice for PS5 (and Xbox), but the PC absolutely requires flexibility and more options than that. To the point where, clearly, no-one thinks it worth the investment to develop asset decompression blocks at this time.

In fact, despite having developed an API, a compression format, and area and power efficient decompression hardware for the Xbox, MS have excluded any mention of dedicated decompression hardware for PC Direct Storage. At first I was a bit disappointed by this, but as I've looked into the various options for software decompression I've quickly come to the conclusion that for the PC this is the right option. At least for the current while. And if there ever is an agreed upon hardware standard it'll have to go way beyond what's in the PS5 to be worthwhile.

Logic simply suggests that any major custom HW development that sums up many millions MUST be considered more efficient than using already existing (general purpose) tech, because all comitees and engineers involved in the process would have pointed out or even blocked further investigation in custom tech.

No, logic doesn't suggest this. That's a faulty logic I'm afraid!

First of all you're making the assumption that what is deemed optimal in one very, very specific set of circumstances is optimal in all circumstances. It's not, certainly not in this case.

Secondly, you're assuming that money = correct.

Third, you're assuming that Sony spent money, and seemingly no-one has spent as much or more (Nvidia's development budget for example dwarfs Sony's by an incredible amount).

There are lots of viable approaches to something like this, and Sony, Xbox, and PC appear to be moving in directions that are better suited for each of them.

P.S. why no love for software based approaches e.g. SFS / PRT+? It's not just Cerny that understands that there's value in loading in only what you need. That idea is as old as the hills!

 

[pjbliverpool]

So the Question we need to ask is what cards will long into the Gen still be able to render PS5 Exclusive Titels in the same quality. I said a RTX 3070 will not be enough . Many disagree and that is fine. It will be intresting to see what comes out of it.

Yes but the question is why do you think this? What's the reasoned basis for the assumption? The 3070 is a significantly more powerful GPU that that of the PS5, as in, if it were in the PS5 instead of the PS5's exiting GPU, the PS5 would be a more powerful console, Hopefully we can agree on that? A simple spec comparison between the 3070's nearest RDNA2 competitor (particularly in RT) and the PS5 should make that abundantly clear. We're looking at a bout a 66% performance difference according to TPU between the PS5 analogous 6600XT and the 6800XT which the 3070 can regularly equal in RT enabled games.

Naturally over time, once a PC architecture falls out of both driver and developer support, it's performance will slip relative to it's potential, but by then such GPU's are considered obsolete anyway, having been replaced by more performant and much cheaper models.

<i believe it when i see it. Nvidia can talk all day long about their RTX I/O .

It's not Nvidia talking about RTX-IO. It's Microsoft talking about Direct Storage with Nvidia and AMD supporting what they say and even several game developers providing commentary around it. Direct Storage GPU decompression is clearly on it's way. There's no reasonable basis upon which to doubt that at present.

PS5 is here and its concept has proven to be way way more advanced than the old way of doing it on PC.

Way way more advanced than the old way? Hardware based decompression has been around since at least the PS4 generation. "Way more advanced" is a pretty inaccurate term. It's more efficient in terms of power usage. More complex in terms of design. Requires additional silicon, but ultimately free's up CPU cycles that can be used elsewhere to potentially increase performance depending on the IO requirements and CPU capabilities.

Having dedicated silicon to perform this task vs doing it on an already present GPU that can do it with negligible performance impact doesn't sounds like either a particularly positive or negative trade off to me.

There are games out there that do stuff that is not possible today on PC.

Unless you have evidence to back up a statement like this (which you don't) then you should state it as your opinion rather than as a fact (which it isn't).

- The obvious Answer always is replyd to that point is :,,But PC can go always pure uncompressed and just have 32GB of mainram to make up for large filesizes". My Answer to that is - Where is it? When it is so easy why nobody does it? Why Star Citicen does not try to do it for example? Why none of the High End PC Games like Cyberpunk have a extra Detail Setting called "Hyper Ultra" wich would make use of such a possibility?

No offense but this is mostly nonsense. You haven't even stated what "stuff" it is that you think a PC can't do which would require the confusing "solution".

There's nothing about hardware decompression that makes anything impossible on the PC, even without DS based GPU decompression, all the console hardware decompression is doing is freeing up CPU cycles. i.e. it's doing literally nothing that a more powerful CPU couldn't do. Sure, that would actually require a more powerful CPU on the PC side - so not ideal, but it's not like they don't exist in abundance now.

The raw IO throughput in hardware terms is already there, and Direct Storage can now also largely equalize the file IO overhead if the developer feels that's necessary/a bottleneck. Of course there is still a way to go on the PC for things to get quite as efficient as the PS5 in every area (i.e. offloading the CPU based decompression, and eliminating unnecessary system memory copies) but those things will simply reduce the CPU requirement to match PS5 streaming capabilities.

NB: PC's will likely always have a bit more to do at loading times as well like shader compilation and static BHV generation that consoles don't have to deal with.

Cerny came up with a ingeniuos idea that Sony paid alot of millions for in development. They streamlined for the very first time the entire I/O Process and build a lot of custom HW. Nvidia on the other hand still needs to prove that a GPU busy with a AAA Game doing raytracing at 60 fps still has time left to decompress lets say 5GB/s. Where are the Demos for that? I did not see a single Demo from them proving their concept..

Well that's because the API literally isn't ready yet. But we have Microsoft, Nvidia, AMD and game developers all talking about working with it so it's quite silly to assume it's just a fabrication at this point. But yeah, Sony and Microsoft absolutely got there first with a non-CPU based decompression solution - on the PS4 and XBO.

I see that the phrase " PC is in trouble" caused some unrest in People hehe. But maybe my english was of a bit. What i was trying to say was that not that PC as a Plattform is in trouble but rather the people withhin its community. Because lets face it - the absolut majority is still gaming on machines that are way below PS5 in terms of render Capabilitiys. Those People need to invest alot in either a new PC altogheter or a serious upgrade.
In that sense this Gen is different than the PS4 one. 2013 you were fine with a little older HW than PS4 almost all of its Gen. Mainly due to Jaguar being used as a CPU and Games not using 8 Core CPU to their potential.

You need to specify whether you're talking about CPU or GPU because it feels like you're confusing the topics here. In CPU terms I do actually agree with you. This gen has brought the console CPU standard up much higher relative to the PC vs last gen. So yes, many gamers on an older systems will facing a several hundred $ upgrade to regain parity or better with the current gen. That's new wrt the PS4 gen but nothing really new compared with generations prior and therefore nothing really new from a PC gamer stand point. In GPU terms the situation is quite comparable to last gen with older (than the current gen consoles) GPU's still being very competitive. Albeit prices have gone up across the board over the last few years.

Still as a PC gamer, outside of temporarily high prices caused by COVID and the mining craze, I actually think there has never been a better time to be a PC gamer. The platform is far more gamer friendly and optimised than it's ever been while still maintaining all of that historical flexibility and backwards compatibility. And while it no longer gets those super high end timed exclusives like Crysis (which were always very few and far between anyway), it instead gets by far the widest selection of games available on any platform thanks to being the new, and only common home for all 3rd party titles, all Microsoft titles, and an ever growing catalogue of Sony titles (plus Nintendo via emulation).

 

[pjbliverpool]

Thats my Point regarding the Jaguar - it is a Netbook CPU. I think DF used one of these strange Xbox One APUs for Desktop PCs a while ago with Cyberpunk and showed that it would not be a valid gaming CPU on PC.
Only the Fact that due to much more efficient APIs on consoles made it possible to use a netbook CPU in the last Gen. Then there was no multithreaded Engine out there (in a AAA Third Party Game anyways) until Assassins Creed Odyssee i think. If i recal that was the first time the usual Suspect CPUs with 4 or less cores failed to deliver the PS4 Expirience.
Also the IPC of a Jaguar Core is not good - because it was not intended to end up in a Gaming machine. Thats way many 2 Core CPUs could outperform a PS4. It is like a said. That time the PC Community had much better HW already when it the Gen started.
This Time - not so much. Not the majority of PC gamers.

It seems like you've just ignored my previous response to this since I already addressed all of the above. PC versions of games would have been re-architecture to utilise high IPC with a small number of cores / threads. You seemingly concur with this above with your "no multithreaded Engine out there". Did games last generation scale linearly with core counts up to 8 cores/threads? Nope. But you can bet the same games on consoles were making near full use of all 8 Jaguar cores.

So there wasn't some magic console efficiency making a Jaguar perform like a CPU that was 4x faster. It was the fact of the games being re-architected to take advantage of high IPC/low core count CPU's that meant dual core PC CPU's could perform similarly to their 8 core console counterparts where the total potential was roughly equivalent.

This generation is different though because the CPU's are relatively better in consoles. So where last gen most people already had a quad core, let alone a dual core when the generation began (and thus more than enough CPU power to not have to worry about it), this gen, more people will have to upgrade. As noted earlier though, this isn't a new thing. The aberration was actually last gen with it's crazy weak CPU's.

I meant the Cost of render time on GPUs. I simply doubt Nvidias statement of RTX I/O being so lean on a GPU. Like i wrote obove - i want to see something. At least a Demo of some sorts. Should not be so difficult. And to adress the "2Tf sacrifice are enough to catch up or even overtake PS5s I/O capabilitys" - i think that came out of another Thread wich was linked here some while ago. I absolutly tink that for one - on RTX 20xx nobody has 2Tf to spare when it renders a lets say 2nd wave PS5 Exclusive. I think trying to mimik PS5s I/O throughput is much more stressfull to a GPU then most people think. It is cute that Jen-Hsun Huang thinks that simply flipping a switch on a RTX 20xx / 30xx GPU is sufficient to mimik PS5 I/O Capabilitys without adding up a kinds of latencys in the CPU/ GPU correspondance ( wich is already much higher than on Consoles because UMA instead of hUMA ) .

The 2Tf info came from @BRiT if I recall. I know there was a link to back that up so perhaps he'd be able to supply it. That said, it was in a different algorithm to that which DS will be using - possibly BC7Prep? So doesn't necessarily tell us much about the performance hit of DS GPU decompression. All we have for that is Nvidia's statement that it's barely measurably. With no evidence to the contrary at this stage I see little reason to doubt them.

Also, you're making a false assumption that there is a requirement to have a constant decompression stream of 5+ GB/s for in game streaming which will never be the case. In reality you're likely to be looking at around 1/10th of that at the high end. UE5 games for example are known to use even less. So the decompression requirements are no where near that stressful. You're only likely to need the full speed of the SSD when loading which is when the full power of the GPU is available anyway.

People seem to forget that PS5s I/O is not only the Kraken Unit. There is more HW there wich is not present on PC.

Not much. Most of that diagram which Cerny sells to you as custom hardware is actually just standard components of a regular SSD controller. The difference in PS5 is that it's running a custom firmware better suited to a fixed platform.

The additional hardware elements in the PS5 are basically the hardware decompression unit and the cache scrubbers.

Again i doubt the Nvidia Statement. When it would be so much more efficient on GPU ("barely measurable") why did Sony or Cerny not take this aproach when designing the PS5?

Because a dedicated hardware ASIC is a pretty simple addition to a fixed console where every last scrap of GPU power is like gold dust. Also, it's not clear if there were any viable GPU based GPU decompression solutions available at the time of these consoles designs. It's actually a fairly recent research topic to make it scaleable to GPU architectures since decompression is traditionally quite single threaded.

It would be way smarter then to not develop the whole custom I/O shenanigens and go with a bit bigger GPU and do the Decompression there. Smarter because then , if a game would for some reason not rely much on decompression they could use the GPU then for better IQ and Detail Settings instead.

It depends entirely on the cost, and as above, whether such a solution was even considered feasible at the time of the consoles design. There's a reason why it's still not available on PC - they're still developing the solution.

Also the consoles have precedent with this kind of design so in many ways this is just an evolution of the PS4's design which also had a hardware decompression block (albeit a far less useful one).

i .. I dont want to come across like talking down to you but forgive me when i say that does not make any sense at all. A Game being 66GB does not tell you anything about floating point calculations or texture/asset swaps. There might be a lets say high quality Tree model that weighs in lets say 500Mb copiying that very model 10 times in and out memory already caused 5GB of traffic. Without the Game being "sucked empty" lol.

The point is that with say 10GB of available graphics memory, you can have 1/6th of the entire game resident in memory at any given time. For most games that's multiple levels or an entire environment. Why would your stream the same tree in and out of memory every second when you can just keep it in memory to be re-used? Unless you have to clear out that memory to make space for new assets it would be senseless to do so. And why would you need to make space for new assets from just a single 180 degree turn? The only reason for that would be to have literally more that 1/6th of your entire game content present within viewing range at any given moment. That would make for an extremely unvaried game.

To use the Cerny analogy of the current viewport filling the memory, or perhaps just the next 1 second of gameplay. That's like having a tree in front of you and behind you, but rather than keeping it in local memory while you turn around, you unload it when its leaves your viewport and then re-load it from disk half a second later when the second version of it re-enters your view. That's pure madness and what have you gained from it unless your memory was insufficient to store everything that was currently within your viewing range? i.e. more than 1/6th of your game content makes up your immediate surroundings.

Also why are you oblivious to the fact that any calculations of vectors or basicly any gamedata that goes through the GPU is adding up to several 100GB of internal traffic within the GPU. L1 and L2 caches have hundreds if not thousends of GB/s Bandwith.
https://www.beyond3d.com/content/reviews/55/7 <- This here suggests a mere 1087.74 GB/s Bandwidth for L1 alone on an old GTX 470...
Playing a game for one hour has caused probably 1Million GB worth of internal GPU traffic for all we know...

The cache bandwidth consumed by the GPU actually rendering is entirely different to the bandwidth consumed by the games assets being moved from disk to memory and back. Obviously so, or else how do you square those cache bandwidths with every game that's ever run off an HDD?

The game swaps out the entire memory usage of a level in a second thats one thing. The other things is that Rift Apart indeed makes use of Cernys forecasted new memory paradigm of only 1sec of gameplay resident in memory.
Memory in high active use.
View attachment 6823
Rift Apart Method of Memory usge is mentioned here :

Interview: How Insomniac made "Ratchet & Clank: Rift Apart" look so good

The veteran studio says the PS5's SSD helped in some unexpected ways.

www.axios.com

also here:

https://twitter.com/x/status/1408036947344384005

Some of that is just marketing spiel, as above, it simply doesn't make sense to flush your entire memory of all the textures, models, and other assets for a 180 degree turn when by necessity of game design and storage capacity limitations those same assets are being used behind you as well. Absolutely the portal transitions between different environments described in that first tweet will be making heavy use of the IO. But a simple turn within the same environment will not.

 

[Flappy Pannus]

Is it the trees, or is it the lack of buildings meaning the draw distance is much, much greater?

Even in a small enclosed areas, like a parkette with a few trees surrounded by buildings will cause GPU usage to spike significantly I find.

 

[Flappy Pannus]

I think nobody watching this comparison will have doubts its proper remake

Definitely a huge upgrade. Quite surprised though to still see the 1440p/4k split from Naughty Dog - no reconstruction tech, and the 'dynamic' res seems to be far less dynamic than hoped.

Edit: 50 minutes, good lord

 

[snc]

Definitely a huge upgrade. Quite surprised though to still see the 1440p/4k split from Naughty Dog - no reconstruction tech, and the 'dynamic' res seems to be far less dynamic than hoped.

Edit: 50 minutes, good lord

At least there is 40fps 4k mode from start tough sadly they didnt implemented lfc when its uncapped (70-80fps 1440p mode uncapped)

 

[PSman1700]

Clearly the 2700X is the ideal CPU for comparing how RTX GPUs stand up to the mighty raytracing beast that is the RDNA2 based PS5.

2700x has never been that performant, Zen for gaming began to get much better starting with the 3xxx series.

uspect a chunk of it is also down to the older Zen 2 3300X actually having full-on PCIe gen 4 while the newer Ryzen 3 5300G is somewhat hobbled with ye-olde PCIe gen 3

Isnt the CPU still supporting PCIE4, just the m2 slots being PCIE3? Its like that on my 5800h laptop. Doesnt hinder performance in Spiderman.

PC has lots of hardware vendors, almost limitless combinations of hardware and software, multiple OSes, at least two major APIs. A single hardware block with a single performance level with a limited selection of hardware accelerated decompression options is a great choice for PS5 (and Xbox), but the PC absolutely requires flexibility and more options than that. To the point where, clearly, no-one thinks it worth the investment to develop asset decompression blocks at this time.

In fact, despite having developed an API, a compression format, and area and power efficient decompression hardware for the Xbox, MS have excluded any mention of dedicated decompression hardware for PC Direct Storage. At first I was a bit disappointed by this, but as I've looked into the various options for software decompression I've quickly come to the conclusion that for the PC this is the right option. At least for the current while. And if there ever is an agreed upon hardware standard it'll have to go way beyond what's in the PS5 to be worthwhile.

Indeed. Different platfroms and architectures fit for different solutions. The on-gpu decompression for the PC makes perfect sense, while for the consoles a hardware decompressor does. I see no disadvantages anyway, GPU's are both much faster and more flexible for these types of operation (with neglible impact).

There are lots of viable approaches to something like this, and Sony, Xbox, and PC appear to be moving in directions that are better suited for each of them.

P.S. why no love for software based approaches e.g. SFS / PRT+? It's not just Cerny that understands that there's value in loading in only what you need. That idea is as old as the hills!

Glad that some can see things like this.

Still as a PC gamer, outside of temporarily high prices caused by COVID and the mining craze, I actually think there has never been a better time to be a PC gamer. The platform is far more gamer friendly and optimised than it's ever been while still maintaining all of that historical flexibility and backwards compatibility. And while it no longer gets those super high end timed exclusives like Crysis (which were always very few and far between anyway), it instead gets by far the widest selection of games available on any platform thanks to being the new, and only common home for all 3rd party titles, all Microsoft titles, and an ever growing catalogue of Sony titles (plus Nintendo via emulation).

You get all MS/xbox games, ALL pc games obviously going back decades and all enjoy the 'remaster/remake' all automatically, often community and modders releasing patchs and total conversions aswell (Q2RTX for example). You also get Sony games, probably all of them going forward and have the ability emulate almost every console that matters. On top of that, you get, if you have the hardware, the technically best versions of games across all platforms.
And... you can do so much more on a pc. Also you'd be able to upgrade if you want to, without the need to hunt for a total mid-gen upgrade. On pc you usually dont concur with delayed 'console exclusives' between platforms.
Also we shouldnt forget that Indie titles and tech demos can be quite stunning for that graphical experience, same for certain mods that raise the bar quite alot. In special with ray tracing these days.
You can game 'on the go' with a gaming laptop, small, compact and easy to take with you. The power of a console and more.

Absolutely the portal transitions between different environments described in that first tweet will be making heavy use of the IO.

And even if so, it wouldnt be a problem for a system equipped with fast CPU, GPU and nvme drive with DS support.

 

[Flappy Pannus]

At least there is 40fps 4k mode from start tough sadly they didnt implemented lfc when its uncapped (70-80fps 1440p mode uncapped)

John mentions that the color light bounce from your flashlight can cause dips below 40fps in that mode, so I'm wondering if the dynamic res is even working properly at this point - may be a bug.

 

[snc]

John mentions that the color light bounce from your flashlight can cause dips below 40fps in that mode, so I'm wondering if the dynamic res is even working properly at this point - may be a bug.

There is no dynamic res, according to John

 

[function]

Isnt the CPU still supporting PCIE4, just the m2 slots being PCIE3? Its like that on my 5800h laptop. Doesnt hinder performance in Spiderman.

As far as I can tell online, both the R5 5300G and the R5 5800H both only support PCIe gen 3. The 5xxx series desktop parts are PCIe Gen 4.

If so (and the AMD website also seems to confirm this) then that could represent and unusually large impediment for this game while using PCIe 3, and transfers onto the GPU will be slower, and perhaps cause some kind of stall on the CPU and/or GPU while using RT.

Whapping an old desktop Zen 2 part down to PCIe gen 3 and seeing if it causes performance degradation might be an interesting test. I don't think PCIe 3 will make a difference with NVMe in terms of RT performance, but at high RT settings (well above PS5) an old PCIe 3 connection might start causing some issues somewhere along the CPU -> PCIe -> GPU trainline.

And even if so, it wouldnt be a problem for a system equipped with fast CPU, GPU and nvme drive with DS support.

The Forespoken devs were demoing ~ 5 GB/s after decompression with an early DS implementation and CPU decompression. That's pretty much up there with what R&C developers were talking about getting from the current R&C implementation.

PC is only going to get faster SSDs, faster CPUs, better DS implementations, better CPU decompressions and insanely powerful GPU decompression. PS5 is great, but PC is going to sail past it at the top end.

Big thing with PC isn't what the platform can do .... it's what the baseline is going to be, for which games, and when.

 

[Flappy Pannus]

There is no dynamic res, according to John

Yes I know, I'm referring to the fact the game describes it as having it though. I'm saying there's a possibility it's just not functioning correctly atm. If there isn't any dynamic res in the 40fps mode (albeit John couldn't find it in the performance mode either) and it routinely drops fps with a flashlight, adding dynamic res to that mode would seem like a sensible patch upgrade.

 

[snc]

Yes I know, I'm referring to the fact the game describes it as having it though. I'm saying there's a possibility it's just not functioning correctly atm. If there isn't any dynamic res in the 40fps mode (albeit John couldn't find it in the performance mode either) and it routinely drops fps with a flashlight, adding dynamic res to that mode would seem like a sensible patch upgrade.

Dynamic res is desribed for 60fps mode not 40

 

[Flappy Pannus]

Logic simply suggests that any major custom HW development that sums up many millions MUST be considered more efficient than using already existing (general purpose) tech, because all comitees and engineers involved in the process would have pointed out or even blocked further investigation in custom tech.

Insinuating that anything designed by a large committee backed by huge corporations will naturally bring about the most efficient implementation possible is ridiculously naive enough (hello "not invented here" syndrome?), but you can literally use recent console development examples to demonstrate this is decidedly not always the case - even in the narrow context of this thread it's clearly false. PS3? Xbox One?

There are obviously more decisions involved in making a mass-market gaming device than simply "What gives us the most power most efficiently?" - other areas of the business want their say in what's prioritized. The fuck-up of the Xbox One is a prime example, the included SRAM for the APU was only 'more efficient' than not having any, but it was also required because they decided to go with DDR3 because the rest of the business wanted to devote component costs to making it the ultimate cable TV/Kinect device and couldn't fully devote those costs to just making the best gaming machine possible. This incompetent prioritization was overseen by many qualified engineers and well paid executives, but it still hit the market in its deeply flawed state.

Then there's the PS3 - An incredibly complex, highly specialized design that was also no doubt the work of countless engineering hours and committee reviews - and ultimately proved to be an albatross around the neck of all but a few very select developers that could devote the significant resources needed to actually get a handle on it. The far more straightforward design of the 360 with it's more 'standard' CPU meant that ultimately, it was was more efficient than the over-engineered, expensive beast than the PS3 was. A lot of considerable talent and experience went into it, and again - it hit the market despite these flaws.

i .. I dont want to come across like talking down to you

My dude, I can assure you that is absolutely not a problem here.

 

[Flappy Pannus]

Dynamic res is desribed for 60fps mode not 40

If there isn't any dynamic res in the 40fps mode (albeit John couldn't find it in the performance mode either) and it routinely drops fps with a flashlight, adding dynamic res to that mode would seem like a sensible patch upgrade.

 

[snc]

@Flappy Pannus

John mentions that the color light bounce from your flashlight can cause dips below 40fps in that mode, so I'm wondering if the dynamic res is even working properly at this point - may be a bug.

Yes I know, I'm referring to the fact the game describes it as having it though

From this messages I thought you think dynamic res was described for 40fps mode, which wasn't. Edit: btw yeah dynamic res would help but even more welcome would be lfc implementation like in insomniac games and then we could use uncapped mode for 40fps saving some performance impact of vsync.