• 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.

Digital Foundry's John: Been talking to developers, people will be pleasantly surprised with PS5's results

ethomaz

Banned
I agree, EVERYONE has bias, I prefer water to any other drink, but I still drink.

I prefer Motorcycles to cars, but I still drive cars also.

I think Sony WWS are at the forefront of game technology and visuals (not the only one) but I love other teams and games equally.

The thing about bias is you have to see past it, mostly it is an emotion. Being a business developer, Architect and Manager over far too many years within the corporate world. You loose this connection and for any scientific analysis you have to remove or see past this "confirmation bias" that can creep in.

Like testing a game performance, test the heavy parts and the light parts and show the balance across all platforms. Where I see this the MOST is in PC benchmarks, I can rarely match up to a major onsite GPU or CPU test and they almost never show the tested section OR even WHAT it actually is.

Is this bias, laziness, lies or all 3?
I worked in review GPU products in the past (R600 / Fermi age).

What I can tell you some reviewers will do a lot of tests with two cards and based in his own bias will choose the best result from one and the worst result from the other to put in the graph... when he should use the same method for both.
That could means a difference of very few fps but that will make the GAP different from what it is actually is (like instead 58 vs 55 it will show 59 vs 52 for example).

Of course there are some reviewers that are extreme methodist.
And others that will show a fair method but will write the article trying to make one product better than other.

There are all kind of reviewers.
 
Last edited:

NXGamer

Member
I think that's less bias, and more preferences. We all have them, but because so many people these days seem to have this Us vs Them tribalism attitude and attachment towards things, they can't see the difference and thus feel compelled to attack and dismiss any opinion they happen to disagree with.
You are correct here IMOP, tribalism is a big issue and the mindset of, if we do not agree then you are my enemy.

Bonkers.
 

Lysandros

Member
Developers weren't saying the XBO would outperform its numbers, were they? To be perfectly crystal clear, I'm not expecting anything to bridge the shader gap, the Series X is more powerful there. But the situation does have significant differences from the base 8th gen. The difference is 17% on peak shader flops, not 40%, and on top of that the PS5's GPU is clocked 23% higher, so while it still comes in with less peak shader flops, everything else on the die is also clocked higher - the command processor, buffers, caches, coprocessors, etc.

It's not unreasonable to think many people alluding to it maybe retaining some advantages outside of shader flops is reasonable.
Well if PS5 cache structure is more robust as evidence suggests (more bandwidth, more L1 and L2 cache available 'per CU', scrubbers) this has the potential to improve CU efficiency since more data is readily available to be processed at lower latency. Can't this also improve 'real world' shader throughput as a result?
 
Last edited:

martino

Member
Like testing a game performance, test the heavy parts and the light parts and show the balance across all platforms. Where I see this the MOST is in PC benchmarks, I can rarely match up to a major onsite GPU or CPU test and they almost never show the tested section OR even WHAT it actually is.

Is this bias, laziness, lies or all 3?
i'm not sure i understand well but and also not sure to what sites you are referring to but the one i consult gives the full hardware used and the testing protocol is documented.
(but we can argue they too often only rely on benchmark when they exist)
for example in a lot of their video DF explain why they choose a specific game and a specific section in them for a specific reason.
i'm less fan of their sponsored content but, so far, it's not hidden when it is.
 
Last edited:

ethomaz

Banned
Developers weren't saying the XBO would outperform its numbers, were they? To be perfectly crystal clear, I'm not expecting anything to bridge the shader gap, the Series X is more powerful there. But the situation does have significant differences from the base 8th gen. The difference is 17% on peak shader flops, not 40%, and on top of that the PS5's GPU is clocked 23% higher, so while it still comes in with less peak shader flops, everything else on the die is also clocked higher - the command processor, buffers, caches, coprocessors, etc.

It's not unreasonable to think many people alluding to it maybe retaining some advantages outside of shader flops is reasonable.
Except for Panello here in GAF I don't remember any developer saying the XB1 will outperform its numbers.
To be fair most were not happy with the performance and the eSRAM complexity.

Activision even demanded MS to upclock the GPU/CPU to not delivery a sub-720p game.
 
Last edited:

LordOfChaos

Member
Well if PS5 cache structure is more robust as evidence suggests (more bandwidth, more L1 and L2 cache available 'per CU', scrubbers) this has the potential to improve CU efficiency since more data is readily available to be processed at lower latency. Can't this also improve 'real world' shader throughput as a result?


Could be. Probably not going to completely bridge 17% that way with a single feature, but overall saying it would be wise to wait on results, there are significant differences than the 8th gen re: the bit about "forgetting history".
 

FrankWza

Member
Bickering over low - medium spec hardware, when will it ever end. Hardware wise, neither are anything to brag about.

when PC has its own exclusives that will make you buy one and not feel the need to supplement your gaming options with a console on top of a PC. Or, when a console manufacturer makes an open system.
 

EverydayBeast

thinks Halo Infinite is a new graphical benchmark
PS5 shows SONY is hungry to win another generation and that leaves Microsoft fans MELTING down.
 

Lysandros

Member
Could be. Probably not going to completely bridge 17% that way with a single feature, but overall saying it would be wise to wait on results, there are significant differences than the 8th gen re: the bit about "forgetting history".
There is also the async compute and culling capabilities to consider. The differences between the geometry engines, the number ACE's etc. But yes, we can only speculate about the real world performance right now.
 

Bojanglez

The Amiga Brotherhood
???? We know the rasterization performance. If it's a menu, One guy ordered a Mac Double, and the other a quarter Pounder. We know what they're getting.

What if the quarter pounder has secret sauce that makes it taste better :messenger_grinning_smiling:
 
Last edited:
PS5 shows SONY is hungry to win another generation and that leaves Microsoft fans MELTING down.

214712.gif
 
Well if PS5 cache structure is more robust as evidence suggests (more bandwidth, more L1 and L2 cache available 'per CU', scrubbers) this has the potential to improve CU efficiency since more data is readily available to be processed at lower latency. Can't this also improve 'real world' shader throughput as a result?

What evidence suggests this? Are you using L1$ and L2$ like most of us would, or the way AMD does (they call L1$, L0$, for example)? We already know Series X has an extra MB of L2$ (L3$) on the GPU, and the GPU can also snoop the CPU caches (the inverse is possible too but has to be done via software).

We know PS5 has faster caches but people use speed and bandwidth interchangeably; the total amount of actual cache data a smaller GPU can process on a single cycle will always be less than the total amount of actual cache data a larger GPU can. So a lot of the gains a smaller GPU (with a smaller array of physical cache) gains with faster clocks only kicks in if the graphics task at hand requires enough cycles of rendering in order to make the higher peak clocks of use.

EDIT: I think people are taking John's comments out of context. It's pretty well known (now) that the devkit environment for the Series systems hasn't been as straightforward as it's been for PS5. While the word from most is that PS5 is like working with a "supercharged PS4", utilizing pretty much the same SDK environment but leveraging the PS5 hardware, MS have been restructuring their SDK into the Gamecore package.

That restructuring was causing some profile issues for doing Series S builds. Heck, we had insiders saying Series X devkits were "running behind" almost a year ago and the somewhat more recent stuff with Gamecore (which IIRC is now in place, it just took a while to iron everything out) validated those rumors. But I think people are seeing John's comment and taking it as to mean overall system performances outside of the SDKs, but it's rather erroneous to suddenly pretend the SDKs aren't a factor and may in fact affect some of the early launch-period 3P titles and that could be what he is referring to when saying some folks will be pleasantly surprised.

Because assuming the PS5 SDK environment really has been mature for longer, it would make sense if we see some 3P titles run a bit better on it. People who aren't aware of the devkit SDK history of the two systems, tho (or play coy about it in this context), will probably assume things through some leaps of logic. Or, conflate things in ways they probably shouldn't. After all, launch-period software (for both systems) are not going to be the defining games for these systems WRT their true performance capabilities.
 
Last edited:
What evidence suggests this? Are you using L1$ and L2$ like most of us would, or the way AMD does (they call L1$, L0$, for example)? We already know Series X has an extra MB of L2$ (L3$) on the GPU, and the GPU can also snoop the CPU caches (the inverse is possible too but has to be done via software).

We know PS5 has faster caches but people use speed and bandwidth interchangeably; the total amount of actual cache data a smaller GPU can process on a single cycle will always be less than the total amount of actual cache data a larger GPU can. So a lot of the gains a smaller GPU (with a smaller array of physical cache) gains with faster clocks only kicks in if the graphics task at hand requires enough cycles of rendering in order to make the higher peak clocks of use.
Github data and MS data. We know Oberon has 4MB of L2 cache (for 36 CUs) and XSX 5MB of L2 cache (for 52 CUs).

So clocks aside each CUs on PS5 have 0.111 MB of L2 cache to be compared to 0.096 MB only for XSX. It means PS5 CUs have about 16% more L2 cache for each CUs. XSX has actually less cache by CUs than XBX (0.1 MB).

Clearly the main aim of XSX designers was to reach an arbitrary tflops number at the cost of some efficiency (compared to Navi 10 design).
 

Bojanglez

The Amiga Brotherhood
Sure, you can guess and assume about what sauce might do, but until proven otherwise, we know which one has the most meat to it.
Yes I'm not denying it. I said that I believe XSX is more powerful, all the known attributes point to that. All I am saying is we are yet to see the games that back that theory up.

Also other factors could be in play too like the maturity of APIs.
 

Faithless83

Banned
Right now GAF gaming side has too many console warring posts. 2 factions constantly derailing threads.

Imagine if they just stopped

Or evolved into the better PCMR :messenger_grinning:
I don't have enough gaming budget to evolve. :messenger_grinning_sweat:

I have to go for the cost effective alternative. I have a gaming PC, but next gen I'm going console only.
Besides, instant resume babyyyy!! :messenger_sunglasses:
 

Zathalus

Member
1. Apis interface between the hardware and the game code.

Are you saying the DX apis dont perform as well on XSX using game code from last gen ? I dont buy it.

2. The hardware architectural advantages will be used unless apis dont use them, so no.

Features yes, such as VRS or mesh shading, but it should be performing as expected and was lower than expected.

3. The game is designed around DX apis not the hardware, go to point 1.

4. THis is same for both consoles, performance saving measures such as VRS, does not negate the comparison using the feature set for last gen

Summary, either the abstract apis in XSX are not performant as expected, or the 52 CU / TF scaling is not linear to performance.

The reason DF chose those titles were that they were DX apis and had no caps in frame rate.

This is what a retired Dev said on this subject


YgLhdmw.png

1. No, I'm not saying that at all, I'm stating that the Xbox has to have another layer of emulation so the game recognizes it as a Xbox One or Xbox One X. This goes hand in hand with it emulating GCN as opposed to RDNA 2. The games are not running natively against the hardware, a emulation layer is involved to simulate a Xbox One X that just so happens to have a 3.8 GHZ CPU and 12 TFLOP GPU instead of the 2.3 GHZ and 6 TFLOP of the original console.

2. If it is emulating GCN then no, it won't.

3. The game is designed with the DX API in mind, but it is not necessarily optimized or coded to take advantage of the extra power. The good example is the boost mode of the PS4 Pro, some games scale well with the extra Mhz, while others have zero changes.

4. That retired dev is wrong, PS4 PRO is a perfect example of this, it can run with only half of the CU units in operation when running PS4 games with no patches. That being said, it is very likely that is not what is happening here. This is just another iteration of the BC work Microsoft has been doing for the past few years, emulation on faster base hardware. Software emulation allows you to mimic different hardware, just look at every PC emulator ever made. You can emulate a PS2 game and it will run far faster, but the game itself is not going to take native advantage of modern GPU features.

These games are being emulated with faster base hardware, so yes taking advantage of the SSD speeds, the CPU speed, the VRAM speeds, the GPU speeds, all are being used to increase game performance. But its a emulated environment, so that means performance overhead, and emulating how the previous consoles behaved so mimicking how GCN did things, not how RDNA 2 does things.
 
Last edited:
Github data and MS data. We know Oberon has 4MB of L2 cache (for 36 CUs) and XSX 5MB of L2 cache (for 52 CUs).

So clocks aside each CUs on PS5 have 0.111 MB of L2 cache to be compared to 0.096 MB only for XSX. It means PS5 CUs have about 16% more L2 cache for each CUs. XSX has actually less cache by CUs than XBX (0.1 MB).

Clearly the main aim of XSX designers was to reach an arbitrary tflops number at the cost of some efficiency (compared to Navi 10 design).

That's a...rather mistaken conclusion on your end tbh.

L2$ is still slower than L1$, and AFAIK Series X has both more L0$ and L1$ by degree of having more CUs in the GPU. Having less L2$ per CU, in that context, isn't necessarily a bad thing, especially considering it also can access its main RAM at a rate of up to 112 GB/s faster. That doesn't completely make up for the smaller array of per-CU L2$, but it does make up for a vast amount of it.

Reducing MS's engineering team efforts down to them simply chasing an arbitrary TF total is more than a bit foolish especially considering they're on record saying they could've gone with even more TF if that was simply their only goal.
 
I think I've said this before, but I find it interesting how vocal MS are about having the most powerful console, you would assume they have something to back that up, maybe they have seen/heard how the 3rd Party titles compare on both platforms otherwise if the PS5 run's them better it would turn out quite embarrassing for them if the "so called" weaker console performs better.
 
Last edited:

Lysandros

Member
That's a...rather mistaken conclusion on your end tbh.

L2$ is still slower than L1$, and AFAIK Series X has both more L0$ and L1$ by degree of having more CUs in the GPU. Having less L2$ per CU, in that context, isn't necessarily a bad thing, especially considering it also can access its main RAM at a rate of up to 112 GB/s faster. That doesn't completely make up for the smaller array of per-CU L2$, but it does make up for a vast amount of it.

Reducing MS's engineering team efforts down to them simply chasing an arbitrary TF total is more than a bit foolish especially considering they're on record saying they could've gone with even more TF if that was simply their only goal.
There is one block of L1 cache per shader array, XSX has four shader arrays like PS5, the amount of L1 cache isn't tied to the number of CU's, see the RDNA diagram. The difference is XSX's L1 cache has to feed 4 more CU's per array compared to PS5 at 20% lower bandwidth.
 
There is one block of L1 cache per shader array, XSX has four shader arrays like PS5, the amount of L1 cache isn't tied to the number of CU's, see the RDNA diagram. The difference is XSX's L1 cache has to feed 4 more CU's per array compared to PS5 at 20% lower bandwidth.

And you're assuming they wouldn't have increased the size of the L1$ cache in order to accommodate this? Given the other smart engineering choices they've made, it'd be rather dumb of them not to and I doubt that to be the case.

Estimates put the Series X's GPU at about 45% of the total die size. That's pretty substantial, and if some of that already accounts for a larger L2$ then it's likely a slightly larger L1$ is also present. Even if not, it still has more L0$ which is the absolute fastest cache of the GPU, so you're still kind of taking to bad conclusions in terms of insinuating some crippled performance here.

I think I've said this before, but I find it interesting how vocal MS are about having the most powerful console, you would assume they have something to back that up, maybe they have seen/heard how the 3rd Party titles compare on both platforms otherwise if the PS5 run's them better it would turn out quite embarrassing for them if the "so called" weaker console performs better.

Read my post above regarding the Gamecore situation. If certain people who were professing for the better part of the past year that Series devkits were "running behind" don't suddenly play coy and give no-substance wink-wink answers (i.e people like Matt over on Era) pretending that's suddenly not a thing/factor, then yes it's possible a select range of 3P games using PS5 as the base platform may run a bit better there.

At least, for launch-period titles. Since the Gamecore stuff has been fixed up more or less, that will probably be less of an occurrence rolling into the latter half of 2021, assuming if a game's using PS5 as the base and isn't getting lazily ported to other platforms. Aside from that there will probably still be an area or two where we see some 3P games do a bit better on one system than the other. For example, I'd generally expect at least slightly faster raw load times for 3P games on PS5.
 
Last edited:

Faithless83

Banned
No need to evolve then. Just don’t devolve into the lower life console warrior.
I was just joking, I got the point. I think people are just bored out of lack of news. So everything that is thrown at us, turns to ammunition "against the other side". Everyone has preferences. I do enjoy throwing a few jabs at PCMR due to most people there being in a high horse act.

That said, it's fun to watch, definitely not fun to moderate.
 
Last edited:
I was just joking, I got the point. I think people are just bored out of lack of news. So everything that is thrown at us, turns to ammunition "against the other side". Everyone has preferences. I do enjoy throwing a few jabs at PCMR due to most people there being in a high horse act.

That said, it's fun to watch, definitely not fun to moderate.
High horse? Is it the fault of the PCMR that our horses are taller? More majestic? :messenger_sunglasses:
 

Faithless83

Banned
High horse? Is it the fault of the PCMR that our horses are taller? More majestic? :messenger_sunglasses:
This. :messenger_grinning_sweat:


I had a blast in this thread though. People couldn't straight up reply "You can't", as soon as they realized it's a feature consoles have but they didn't. People went mental there, cries of horror left and right. Some lost the notion of time and space. It was glorious. :messenger_tears_of_joy:

But I'm derailing the thread, sorry. I must not do that again. :messenger_anxious:
 
Last edited:
Sony fanboys are simply too obsessed, they want everybody to catter to Sony at the extreme and all the time. There's no room for criticism, that's a sin and turns you into a sinner.

Whenever a comparison or opinion disfavours Sony in whatever minimal aspect, then their full extremist mentality is unloaded unto those who dared wrong Mommy Sony.

So when a team like DF tries to do objective analysis, that inevitably mess with the fanboys who in return try to destroy the credibility of DF to invalidate their objective analysis.


One hunnit %100

This should be on the front page for generations to come. Straight up legit mentality of delusional, fanatic Sony fanboy (there's a few of them, but most of the majority are cool open minded folks)

Try to ask a question that doesn't favor them in a Sony thread, they'll deport/ban you on the spot to F train downtown Brooklyn...living proof right here 🤭😁
 

Lysandros

Member
And you're assuming they wouldn't have increased the size of the L1$ cache in order to accommodate this? Given the other smart engineering choices they've made, it'd be rather dumb of them not to and I doubt that to be the case.

Estimates put the Series X's GPU at about 45% of the total die size. That's pretty substantial, and if some of that already accounts for a larger L2$ then it's likely a slightly larger L1$ is also present. Even if not, it still has more L0$ which is the absolute fastest cache of the GPU, so you're still kind of taking to bad conclusions in terms of insinuating some crippled performance here.
The fact of XSX's CU's having L0 caches for each one of them is very nice since they wouldn't really function without it. By the way i know that they are the fastest caches, PS5's are even faster. If we are discussing per CU efficiency (like we were before your reply) i can't see how it helps your cause. Secondly, i didn't claim to know XSX's L1 cache size did i? I only said it has to feed 4 more CU's at 20% lower bandwidth. I used 5700 xt as a reference, it has 128kb. I don't know if it's really possible to increase that amount by 20% like the L2 cache. We'll see.
 
Last edited:

onQ123

Member
Not reading through this thread because all I see is people making this about the people & not the issue at hand but what I will say is that PS5 is showing games while Xbox Series X is still hiding with only a month to go & if you don't think it's a reason for this you are most likely delusional or just going off of blind faith.


My opinion of the situation is that Microsoft plan was to use the hardware to enhance the software across multiple devices like what we see in phones , tablets & PC & not really dig into the Xbox Series X until it sell enough units that it would be worth making software for. I think the result is that the games don't look as good as what we are seeing from PS5 unless devs make the choice to dig into Xbox Series X on their own without doing things the way MS is laying things out for them.
 

Neo_game

Member
There is one block of L1 cache per shader array, XSX has four shader arrays like PS5, the amount of L1 cache isn't tied to the number of CU's, see the RDNA diagram. The difference is XSX's L1 cache has to feed 4 more CU's per array compared to PS5 at 20% lower bandwidth.

I think that is what they meant increasing clocks will also give better BW. Though 16gbps RAM would have definitely helped. They also admitted it.
 

NickFire

Member
Some people overlook the amount of extra dedicated hardware that's in the PS5 to relieve the CPU from some duties plus making the SSD speed even faster, and you get a system that punches well above it's TF weight. But it's done specifically from additional dedicated hardware.
I'm not trying to knock MS at all by suggesting they don't have brilliant people working for them because they do. But some people really seem to forget just how well Sony engineers things. So far in 4 generations their only screw up, if you want to call it that, was making the PS3 too complex for third party devs. Which was a lesson they clearly took to heart when designing PS4. I have no doubt, none at all, that PS5 will be revolutionary for gaming despite a little less brute force capability.
 
Last edited:

Lysandros

Member
I think that is what they meant increasing clocks will also give better BW. Though 16gbps RAM would have definitely helped. They also admitted it.

The RAM bandwidth is the only aspect of PS5 hardware that i don't really like, 512 GB/s would be much preferable. Maybe cache bandwidth will help this stuation a bit, we'll see.
 
Last edited:

Panajev2001a

GAF's Pleasant Genius
And you're assuming they wouldn't have increased the size of the L1$ cache in order to accommodate this? Given the other smart engineering choices they've made, it'd be rather dumb of them not to and I doubt that to be the case.

Estimates put the Series X's GPU at about 45% of the total die size. That's pretty substantial, and if some of that already accounts for a larger L2$ then it's likely a slightly larger L1$ is also present. Even if not, it still has more L0$ which is the absolute fastest cache of the GPU, so you're still kind of taking to bad conclusions in terms of insinuating some crippled performance here.

It is not about being smart or stupid. It is about costs.

Were they stupid because they have only 10 GB of full speed RAM and the rest travelling at a significant lower speed? Were they stupid to allow Sony an advantage on shared resources on the GPU where clock is the only differentiator across consoles (could have gone for an extra Shader Engine and thus overall smaller Shader Arrays)? Was Sony stupid for not adding more CU’s? Nope... it is possible that they believe the current solution fits their targets, regardless of it it gives them an absolutely dominating position or not (12 TFLOPS tagli be is enough to have).
 
Last edited:

LordOfChaos

Member
My opinion of the situation is that Microsoft plan was to use the hardware to enhance the software across multiple devices like what we see in phones , tablets & PC & not really dig into the Xbox Series X until it sell enough units that it would be worth making software for. I think the result is that the games don't look as good as what we are seeing from PS5 unless devs make the choice to dig into Xbox Series X on their own without doing things the way MS is laying things out for them.

This is also reflected in them moving to the GDK, what was previously the XDK was specific for the Xbox while GDK which is a more general environment for PC, Xbox, even Xbox One, so it might not get as many titles that use every bit of its bespoke customization.
 
The fact of XSX's CU's having L0 caches for each one of them is very nice since they wouldn't really function without it. By the way i know that they are the fastest caches, PS5's are even faster. If we are discussing per CU efficiency (like we were before your reply) i can't see how it helps your cause. Secondly, i didn't claim to know XSX's L1 cache size did i? I only said it has to feed 4 more CU's at 20% lower bandwidth. I used 5700 xt as a reference, it has 128kb. I don't know if it's really possible to increase that amount by 20% like the L2 cache. We'll see.

Man, I can almost taste the facetiousness in this post...

Anyway, it's like I said earlier about the PS5 caches being "faster"; that's something that begins to factor in if the graphical tasks in question need a certain duration of cycles to where it actually becomes a factor. Otherwise, on a cycle-for-cycle basis, the larger GPU with the larger physical amount of cache is going to be able to crunch more data in parallel than the smaller GPU with a smaller physical array of cache.

I don't know what was being discussed before my reply, but everything I'm bringing up fits neatly into that discussion. It's contingent to it, it's pertinent. You don't get to determine something that fits relatively close in with what you were discussing prior (does not physical cache allocation on the L0$ level affect CU efficiency? I surely would think it does) just because it brings up a point you either didn't consider or, in light of being indicated, don't like.

Why would it not be possible to increase the L1$ size? These systems are at least on 7nm DUV Enhanced; even a few slight architectural changes here and there would allow for more budget to cache sizes. I'm not saying it's 100% a lock they did increase the L1$ size, just that it's premature to assume they did not, when they've already increased the L2$ amount.

Otherwise yes, it's true if the L1$ sizes are the same for both then Series X feeds more at a 20% reduced speed. But you won't be needing to access the L1$ frequently in the first place if you have more physical L0$ allowing for a higher amount of unique data to be retained in the absolute fastest cache pool available. And that's where, on same architectures, the larger GPU has a very clear advantage in; always have and always will (unless we're talking about GPUs of two different architectures where the smaller one has a much larger L1$, but that's not what we're dealing with here regards RDNA2. Only discrete GPUs I can think of doing this are some upcoming Intel Xe ones that are very L0$-happy).

It'd be really nice if we stopped confusing cache speed with cache bandwidth. IMO the former should pertain to overall data throughput measured in overall time (cycle) duration. The latter should pertain to single-cycle throughput, which is dependent on actual cache sizes. Assuming L1$ is the same, their bandwidth is the same and the speed advantage for a given graphical task getting crunched on the caches only starts showing a perceptible difference in favor for that with the faster clocks if a certain threshold of data processing for that task in the caches is done. We can apply this to the L0$ as well.

It is bot about being smart or stupid. It is about costs.

Were they stupid because they have only 10 GB of full speed RAM and the rest travelling at a significant lower speed?

No, because if PC GPU benchmarks are anything to go by many, MANY reserve a chunk of the VRAM as just-in-case cache, even if the game isn't actually occupying the cache in that moment of time.

So you'd think smarter utilization of the VRAM by cutting down on the use of chunks of it as a cache would make better use of it...thankfully MS have developed things into XvA like SFS to enable that type of smarter utilization of a smaller VRAM budget. Sony has a great solution too; it's different to MS, but both are valid and make a few tradeoffs to hit their marks. At least regarding MS's, I don't think those tradeoffs are what you're highlighting here, going by extensive research into this.

Were they stupid to allow Sony an advantage on shared resources on the GPU where clock is the only differentiator across consoles (could have gone for an extra Shader Engine and thus overall smaller Shader Arrays)?

You do realize the RAM still needs to hold the OS, CPU-bound tasks and audio data, correct? Realistically we're looking at 14 GB for everything outside of the OS reserve for PS5 (NX Gamer's brought up the whole idea of caching the data to the SSD before; not that it's a realistic option IMHO outside of some tertiary OS utilities seeing as how the vast bulk of critical OS tasks expect the speed and byte-level addressable granularity of volatile memory to work with), and if we're talking games with similar CPU and audio budgets on both platforms, at most you have 1 extra GB for the GPU on PS5 vs. Series X, but you sacrifice half a gig of RAM for CPU and audio-bound data.

Yes it does have a faster SSD but there's still a lot of aspects of the I/O data pathway that are apparently CPU-bound once the data is actually in RAM.

Was Sony stupid for not adding more CU’s? Nope...

Glad we agree on this part.

it is possible that they believe the current solution fits their targets, regardless of it it gives them an absolutely dominating position or not (12 TFLOPS tagli be is enough to have).

Just because MS happens to have more TF performance doesn't mean they didn't aim for a balanced design target, either. This is a common misconception and comes from a binary mode of thinking, where everything's either a hard either/or. Console design is much more complicated than that.
 
Last edited:

onQ123

Member
This is also reflected in them moving to the GDK, what was previously the XDK was specific for the Xbox while GDK which is a more general environment for PC, Xbox, even Xbox One, so it might not get as many titles that use every bit of its bespoke customization.

The most telling thing is that they are showing you Xbox Series S games with no problem while keeping Xbox Series X gameplay in the background so that people will have this false assumption that Xbox Series X games will look 3X better. Also letting people think that whatever they see from PS5 games is going to look 20% better on Xbox Series X all while not showing us anything yet.
 

AgentP

Thinks mods influence posters politics. Promoted to QAnon Editor.
The RAM bandwidth is the only aspect of PS5 hardware that i don't really like, 512 GB/s would be much preferable. Maybe cache bandwidth will help this stuation a bit, we'll see.

I'm more worried about the XBSX having split memory design. So if they want more than the 10GB (can't recall the exact number), they are going to have two memory bandwidths to contend with. At least the PS5 can dynamically allocate memory away from the OS as time goes by and the performance is constant.
 
This is also reflected in them moving to the GDK, what was previously the XDK was specific for the Xbox while GDK which is a more general environment for PC, Xbox, even Xbox One, so it might not get as many titles that use every bit of its bespoke customization.

Would they not just simply roll the XDK features into GDK and provide the API environment for targeting Series platform hardware performance?

Granted even doing that would complicate the integration process, but that would seem to also explain why Gamecore has taken a bit long to fully lock down. No company actually regresses in API functions and ease of abstraction for developers to target the underlying hardware of a given platform going from one generation to the next, why this is suddenly entertained as a possibility here is beyond me.

I'm more worried about the XBSX having split memory design. So if they want more than the 10GB (can't recall the exact number), they are going to have two memory bandwidths to contend with. At least the PS5 can dynamically allocate memory away from the OS as time goes by and the performance is constant.

Man, the FUD never stops :rolleyes:

Series X isn't using split memory. There are two memory pools of the fundamentally same GDDR6 volatile memory. The mixture comes from the different chip densities. Six of the chips are 2 GB, 4 are 1 GB. If the GPU wants to access memory, it does so through the 4x 1 GB chips and the first 1 GB portion of the 2 GB chips. That gives 560 GB/s for the GPU. If the CPU and audio want data, they access it through the second 1 GB portion o the 2 GB chips. That gives them 336 GB/s.

People like Lady Gaia (not the only one, but a name that stands out to me) on Era were erroneously leading people to think that there would be massive bandwidth drops by assuming worst-case scenarios and interleaved memory channel sharing. The Hot Chips presentation dispelled this by denoting 20 channels for the GDDR6. Any developer stupid enough to spool data for the GPU outside of the 10 GB fast pool will see a performance hit, but that depends on the amount of time (in cycles) their code spends in the slower pool, then swapping back to the faster pool.

This ISN'T an average, because going between the two pools would be a cycle at most, or only a scan few cycles to do. Trying to weigh it as an average would also assume 50% of the GPU time spent in both pools (similar to how when people try calculating the Series X system bandwidth average they try doing the same, assuming game code would ever spend even duration on GPU-bound data and CPU-bound data over any period of time), or for any massive amount of cycle time, is disingenuous.

Other reasons it's not a split memory pool is because there's no need for data shadow-copying between the two pools like it'd be on PC with DDR system RAM and GPU VRAM. Additionally, no specific processor component has its own exclusive reserve of physically separate RAM, like you'd get in older systems like NES, MegaDrive, or even more recent systems like PS3.
 
Last edited:

geordiemp

Member
1. No, I'm not saying that at all, I'm stating that the Xbox has to have another layer of emulation so the game recognizes it as a Xbox One or Xbox One X. This goes hand in hand with it emulating GCN as opposed to RDNA 2. The games are not running natively against the hardware, a emulation layer is involved to simulate a Xbox One X that just so happens to have a 3.8 GHZ CPU and 12 TFLOP GPU instead of the 2.3 GHZ and 6 TFLOP of the original console.

2. If it is emulating GCN then no, it won't.

3. The game is designed with the DX API in mind, but it is not necessarily optimized or coded to take advantage of the extra power. The good example is the boost mode of the PS4 Pro, some games scale well with the extra Mhz, while others have zero changes.

4. That retired dev is wrong, PS4 PRO is a perfect example of this, it can run with only half of the CU units in operation when running PS4 games with no patches. That being said, it is very likely that is not what is happening here. This is just another iteration of the BC work Microsoft has been doing for the past few years, emulation on faster base hardware. Software emulation allows you to mimic different hardware, just look at every PC emulator ever made. You can emulate a PS2 game and it will run far faster, but the game itself is not going to take native advantage of modern GPU features.

These games are being emulated with faster base hardware, so yes taking advantage of the SSD speeds, the CPU speed, the VRAM speeds, the GPU speeds, all are being used to increase game performance. But its a emulated environment, so that means performance overhead, and emulating how the previous consoles behaved so mimicking how GCN did things, not how RDNA 2 does things.

We disagree, thats fine, ps4 pro had nearly everything in boost mode Sony were just covering the odd one, I cant think of any of top of my head.

XSX was not cut down in CU, and it was running DX code, emulating or abstract api is a fine line of definition as its the MS way of operation.

Emulating is just wrong, its X86 and DX12 so no, just no.

Saying CGN game code is different to RDNA game code is why that dev mocked, and so do I.
 

DonMigs85

Member
I think the Series X has more ROPs and TMUs to compensate for its clock deficit right? Heck even the One X had half the ROPs of PS4 Pro but the memory bandwidth and shader count more than made up for it
 

Elog

Member
There are some strange statements in the last few posts.

Every additional CU to a GPU adds less output than the previously added CU. This is a mathematical fact unless a task can be parallelised to 100% (which no task can).

Secondly, under heavy load the bottle-neck for a GPU is most often the unified cache (L2 in AMDs language) which results in CUs either idling or doing redundant work (i.e. performs a task that is either no longer required or is done by another CU). This is a real problem and if you look around at solutions where people try to increase cache efficiency on GPUs you can see crazy numbers such as +50% in actual output in terms of computational performance.

Since we lack data on the PS5's cache system it is hard to make a comparison. All we know is that Sony has spent some serious work in trying to increase cache efficiency. We will see if they have succeeded. However, as seen in the +50% number above good cache management can yield much better results if done right than just adding CUs. Please note that I am not claiming that the PS5 is getting +50% output from the CUs - it is just an example to make clear that proper cache management can yield results that are much more impressive than a few % in increased actual performance.
 
Last edited:

AgentP

Thinks mods influence posters politics. Promoted to QAnon Editor.
Other reasons it's not a split memory pool is because there's no need for data shadow-copying between the two pools like it'd be on PC with DDR system RAM and GPU VRAM. Additionally, no specific processor component has its own exclusive reserve of physically separate RAM, like you'd get in older systems like NES, MegaDrive, or even more recent systems like PS3.

Yes, you basically just summarized the PS4 design that MS moved to with the XBX. The bottom line is they have two memory bandwidths and if they are ever going to dip into the slower pool, they are going to create a bottleneck. Still better than the XBSS single low bandwidth solution.
 
The sound cut out before he finished, but as a professional lip reader I was able to get everything he said.

The full sentence was

"We haven't actually had hands on with the PS5 yet, but just talking to developers... I think people will be surprised in a good way. [I have been] hearing some good things about that PS5[running at 1440p and dynamic 4K]. :messenger_beaming:
Resolution doesn't matter nearly as much to me like it did last gen when most PS4 games ran at 1080p, because anything above that is already in full HD so I don't really get the point of running games in 4k as demanding as it is.

I'd rather games be 1440p/60FPS but we can't all get our way.
 
Top Bottom