• 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.

DF: The Touryst PS5 - The First 8K 60fps Console Game

Boglin

Member
Call me a dreamer but I personally believe that both The Touryst and Avengers are what we will see in typical performance comparisons going forward but with the differences showing up in the same game. The XSX will have up to a 100% resolution advantage over the PS5 and the PS5 will have a 100% resolution advantage over the XSX.

With both consoles simultaneously outperforming each other, DJ Khaled will finally gather enough spirit energy to perform mitosis and the world will be better for it.
 
Last edited:

azertydu91

Hard to Kill
Call me a dreamer but I personally believe that both The Touryst and Avengers are what we will see in typical performance comparisons going forward, but the differences will show up in the same game. The XSX will have up to a 100% resolution advantage over the PS5 and the PS5 will have a 100% resolution advantage over the XSX.

With both consoles simultaneously outperforming each other, DJ Khaled will finally gather enough spirit energy to perform mitosis and the world will be better for it.
Do you mean Dj Khaled perfect form?No we can't do this .... he will be too strong... Please lord have mercy upon our soul because our doom is coming.
 

Boglin

Member
Do you mean Dj Khaled perfect form?No we can't do this .... he will be too strong... Please lord have mercy upon our soul because our doom is coming.
No, I mean mitosis so he splits into two. The world will be filled with positivity from the infinite well of compliments the DJ Khaleds will have for each other.
 
This thread is full steam going now. Some of these posts are legendary, before the consoles launched types of rants. People feeling like they have a true technical knowledge of the inside and outs of these systems.

Can we all just let it go. There is consistent evidence each few weeks of how these are performing.

Xbox started rough, tools joke was made...xbox improved tools and games like control ended up running better on xbox.

Newer games are launching with the exact differences we expect. Between 10 to 30 percent more pixels being rendered on the same scenes. With sometimes a more steadier fps on ps5 due to the speed the gpu is operating.

Sony ssd is performing faster than xbox and in some freak circumstances...like this 2 year old indie game they are pushing 8k vs 6k once the game has been rebuilt from the ground up for the ps5.

The true test is, we don't really have any true only built for next gen games and with sonys new stance on pc and Jim's dream of not liking consoles, wanting games everywhere we may never see a game built from Cernys dreams of fully utilising the ps5s custom io. I personally find that a shame.

I don't think we will see next gen games ever again. The next gen will still be latched to ps5 and series x. Etc.

So let's just try and enjoy the games?

Like many have said, these machines have never been closer...they are literally a cocks hair between them. 10 vs 12 it's next to nothing as the evidence is showing us.

Because you don't like seeing people speculate why the PS5 runs this game at 8K?

That's all this discussion is about.

And it's an interesting discussion since the XSX tools are fixed and the resolution advantage is supposed to be 20-30% in favor of Microsofts box.

I'm certainly learning a lot from this discussion. It merely proves that flops are not everything and sometimes other factors can lead to a game having a higher resolution.
 
Last edited:
Call me a dreamer but I personally believe that both The Touryst and Avengers are what we will see in typical performance comparisons going forward but with the differences showing up in the same game. The XSX will have up to a 100% resolution advantage over the PS5 and the PS5 will have a 100% resolution advantage over the XSX.

With both consoles simultaneously outperforming each other, DJ Khaled will finally gather enough spirit energy to perform mitosis and the world will be better for it.

Sounds like a joke but it's actually true.
 
The XSX will have up to a 100% resolution advantage over the PS5 and the PS5 will have a 100% resolution advantage over the XSX.

I believe some are considering the outliers to be representative of the typical situation between the two systems. My advice is to look at many comparisons and see what the typical situation is. Cases like The Touryst will be rare as well as situations like The Avengers. Most comparisons tend to be a lot closer that those two from what I've seen.
 

DenchDeckard

Moderated wildly
Do you see @SenjutsuSage That's how you make a subtle console war post.
Praise their proximity in power while casualy mention that one has always between 10 to 30% res advantage, all that while ignoring when it is the other way around.
Consider that a game run better even though it is only in photo mode.
Dismiss subtely the game as a 2 years old indie game that has been redone for ps5 like it was just ad xb1 port the other way arouud.
Make a general statement of the industry all that while criticizing only one manufacturer.
Then finish by blurrig the lines saying they are closer than ever.
That way more subtle than your lies or misunderstanding about tech.
Haha, like genuinely what the funk. The xbox has more CUs at lower clocks just like a 3080 ti vs a 3060 or 3070. The xbox should be able to push more resolution its the nature of its gpu and how its built.

I am no console warrior type and I can guarantee that the post I did was not a console war post. Nice labelling though. Cheers.
 

azertydu91

Hard to Kill
Haha, like genuinely what the funk. The xbox has more CUs at lower clocks just like a 3080 ti vs a 3060 or 3070. The xbox should be able to push more resolution its the nature of its gpu and how its built.

I am no console warrior type and I can guarantee that the post I did was not a console war post. Nice labelling though. Cheers.
Yet there's no labelling or calling a warrior in my post.Just a post that was subtely one sided but nonetheless one sided.But that's okay honestly if I had labelled you as a toxic warrior I'd just put you on ignore like abel (abel empire? I can't remember) or riky.
 
The xbox should be able to push more resolution its the nature of its gpu and how its built.

Not exactly according to this developer. It really depends on the engine and what it's requirements are. The developer isn't putting the blame to tools or lazy development practices. He's merely stating that they were able to achieve 8K on the PS5 due to the higher clocks and memory setup.

Now your smart enough to know it's not just one sided where advantages are concerned. In some cases the XSX will pull ahead and in others the PS5 will pull ahead. It might not be often when the PS5 does but due to how the hardware is it's definitely a possibility instead of an impossibility. Really not trying to console war here though.
 

S0ULZB0URNE

Member
Haha, like genuinely what the funk. The xbox has more CUs at lower clocks just like a 3080 ti vs a 3060 or 3070. The xbox should be able to push more resolution its the nature of its gpu and how its built.

I am no console warrior type and I can guarantee that the post I did was not a console war post. Nice labelling though. Cheers.
Xbox Series X has split ram and a slower I/O so sometimes these are the results.
 

FrankWza

Member
.
I am no console warrior type and I can guarantee that the post I did was not a console war post.
fox GIF by Proven Innocent
 
While both machines have the same number of color ROPs, PS5 has twice more Depth / stencil ROPs than XSX. PS5 ROPs take almost twice more space on the APU than XSX ROPs.

Locuza wondered if that would bring some benefits in some games. Well I think we have found one of those games with The Touryst. A game heavily limited by pixel throughput performing almost twice better on a machine having almost twice more ROPs?

Area footprint for ROPs [of XSX] appears nearly half in comparison to the PS5. (Not sure if the latter can benefit from more Depth/Stencil ROPs in practise)

 
Last edited:

StreetsofBeige

Gold Member
This game is 8K on PS5 but 6K on Xbox Series X it's a simple game but because it is simple it's not compute limited so it's putting the fix function pipeline to the test.


It's rendered at almost 2X higher than it is on Xbox Series X .
Well, I hope it looks better on PS5.

It came out a year later and the dev said he rewrote the game engine from scratch to take into account PS5 specs and API.
 
Last edited:
The dev was simply saying that it's a PS5 game & not just a PS4 game running on PS5. Now y'all want to act like they made a whole new game for PS5 while running the Switch version on Xbox Series X 😂

Obviously the Series X got the Series version and not the Switch version of the game. If that wasn't the case I believe the developer would have stated as much. Something along the lines of this.

That was just the Series version running through BC and we didn't have the best tools at the time. It's why after making a Native PS5 version we were able to hit a higher internal resolution.

Im sure the clocks were a factor but I struggle to understand the memory situation. Especially since the PC version at 8K doesn't surpass 10GBs. But maybe it's a console thing.
 
Last edited:

StreetsofBeige

Gold Member
The dev was simply saying that it's a PS5 game & not just a PS4 game running on PS5. Now y'all want to act like they made a whole new game for PS5 while running the Switch version on Xbox Series X 😂
Not at all.

But when devs take advantage of certain rewriting code, it can work wonders.

That's why BC games can have fasting loading times on Series X than PS5 because the dev didn't want to do anything special with the PS5 game. But when the dev does some PS5 API stuff to it to make the system recognize it as a PS5 game and not just a brute force a PS4 BC game, the PS5 version loads faster than Series X.

Games can work wonders when a dev puts effort into reworking code.
 
Last edited:
Not at all.

But when devs take advantage of certain rewriting code, it can work wonders.

That's why BC games can have fasting loading times on Series X than PS5 because the dev didn't want to do anything special with the PS5 game. But when the dev does some PS5 API stuff to it to make the system recognize it as a PS5 game and not just a brute force a PS4 BC game, the PS5 version loads faster than Series X.

Games can work wonders when a dev puts effort into reworking code.

Do we have evidence that the Series X version was a half ass port or running through BC?

The developer certainly didn't mention it. All they said were the clocks and the memory setup.
 

01011001

Banned
Do we have evidence that the Series X version was a half ass port or running through BC?

The developer certainly didn't mention it. All they said were the clocks and the memory setup.

no it wasn't but it was clearly an earlier version... given that it was released, well, way earlier. so the people working on the game might have refined it while porting it to PS5.

Not at all.

But when devs take advantage of certain rewriting code, it can work wonders.

That's why BC games can have fasting loading times on Series X than PS5 because the dev didn't want to do anything special with the PS5 game. But when the dev does some PS5 API stuff to it to make the system recognize it as a PS5 game and not just a brute force a PS4 BC game, the PS5 version loads faster than Series X.

Games can work wonders when a dev puts effort into reworking code.

small correction here, back compat games on PS5 will never load as fast as back compat titles on Series X|S. the back compat games on PS5 have an artificial speed limit to how fast they can load. this is most likely done to be safe when it comes to compatibility. so the PS5 in BC mode is locked to the max storage bandwidth of the PS4 Pro

so you need a native PS5 port to use the additional bandwidth, and that is way more than just "using PS5 API stuff" it's literally a completely different port of the game.
 
Last edited:
no it wasn't but it was clearly an earlier version... given that it was released, well, way earlier. so the people working on the game might have refined it while porting it to PS5

Maybe the Series X version didn't work on the PS5 with a few tweaks so they had to work on the code. I don't see a reason why they didn't state that if it was the issue.
 

01011001

Banned
Maybe the Series X version didn't work on the PS5 with a few tweaks so they had to work on the code. I don't see a reason why they didn't state that if it was the issue.

well they did state that this port is a completely new port made from the ground up for PS5, they didn't port the Xbox version.
not because it didn't work, but most likely because it's cleaner that way, instead of doing a derivative version of a derivative version.

people have to remember that Shinen has its roots in the german graphics demo scene.
 
Last edited:

onQ123

Member
Not at all.

But when devs take advantage of certain rewriting code, it can work wonders.

That's why BC games can have fasting loading times on Series X than PS5 because the dev didn't want to do anything special with the PS5 game. But when the dev does some PS5 API stuff to it to make the system recognize it as a PS5 game and not just a brute force a PS4 BC game, the PS5 version loads faster than Series X.

Games can work wonders when a dev puts effort into reworking code.

And you think these devs advanced so much in the last 10 months that they could get double the resolution ?



I will say this if it's the ROPS holding Xbox Series X back in a game like this that's not compute heavy they could just go around the ROPS & do it with compute.
 
well they did state that this port is a completely new port made from the ground up for PS5, they didn't port the Xbox version.
not because it didn't work, but most likely because it's cleaner that way, instead of doing a derivative version of a derivative version.

people have to remember that Shinen has its roots in the german graphics demo scene.

So why are people assuming the XSX version is dirty?
 

StreetsofBeige

Gold Member
no it wasn't but it was clearly an earlier version... given that it was released, well, way earlier. so the people working on the game might have refined it while porting it to PS5.



small correction here, back compat games on PS5 will never load as fast as back compat titles on Series X|S. the back compat games on PS5 have an artificial speed limit to how fast they can load. this is most likely done to be safe when it comes to compatibility. so the PS5 in BC mode is locked to the max storage bandwidth of the PS4 Pro

so you need a native PS5 port to use the additional bandwidth, and that is way more than just "using PS5 API stuff" it's literally a completely different port of the game.
Ya thats what I meant.

Just to show how big of a difference it can be, a native next gen game can load a game in a couple seconds. It'll look something like this:

Old game on last gen system: 1 minute loading
Next gen system running game in BC mode: 20 seconds
Next gen game running native: 2 seconds

On paper it doesn't even make sense, since if a native next gen game can load in 2 seconds, a BC game should be able to do it even faster at maybe half a second or 1 second. But an old game can take 10x longer to load.

And that doesn't even take into account a next gen native game getting extra perks like RT or whatever over an old BC game. So not only does it look and run better. But also load 10x faster.
 

01011001

Banned
So why are people assuming the XSX version is dirty?

it isn't, but if you work on the same game again and again, you will most likely find ways to improve it over time.
the question is if they could improve the Xbox version if they wanted to, which is hard to say.

this could very well be a game where the PS5 simply has an advantage due to the fact that the way the renderer works in this game is very unusual compared to most modern games.

what could also be the case is that they early on had 6K as a goal on the Xbox version
it's a nice and round resolution to aim for after all. and there is most likely headroom above that on Xbox even as it is now (would be cool if they had an FPS unlock setting so we could see just how much headroom there is)
so they might have hit that target and didn't even try to push for 7K or 8K, it's only used for Antialiasing after all.

it's hard to say without a dev speaking in detail about the differences in play here. and the devs themselves might also not really know how far they could push the Xbox version if they really targeted 6K from the start and were satisfied with that once they had it running locked at 60fps
 

01011001

Banned
Ya thats what I meant.

Just to show how big of a difference it can be, a native next gen game can load a game in a couple seconds. It'll look something like this:

Old game on last gen system: 1 minute loading
Next gen system running game in BC mode: 20 seconds
Next gen game running native: 2 seconds

On paper it doesn't even make sense, since if a native next gen game can load in 2 seconds, a BC game should be able to do it even faster at maybe half a second or 1 second. But an old game can take 10x longer to load.

And that doesn't even take into account a next gen native game getting extra perks like RT or whatever over an old BC game. So not only does it look and run better. But also load 10x faster.

btw, the PS5 doesn't actually load ps4 games faster than a PS4 Pro, if you put a reasonably fast SSD in the PS4 Pro it can actually load faster than on PS5 in rare cases, and is mostly on par in most other cases.
hence my suspicion that they locked the BC mode to PS4 Pro bandwidth for compatibility reasons. which is honestly most likely not even really necessary but unlocking it would take too much testing on their side maybe.
 
Last edited:
Nothing about it is contradictory if you read carefully what I actually said. What I am speaking on are hardware advantages that are well documented in favor of Xbox Series X. They aren't up to feelings or opinion on if they are actual hardware advantages or not. Saying those hardware advantages will more readily come into play further into the gen when game engines and techniques that are designed to get the most from it actually arrive isn't the same as claiming victory based on results one year in or from a single title when we know the consoles have much left in the tank. And let's not pretend Series X isn't already demonstrating it's the more capable hardware by the expected percentages. It kinda is.

Below is what we're talking about

12TF 52 Compute Unit Full RDNA 2 GPU (or 26 Dual Compute Units)
3,328 Stream Processors
10GB @ 560GB/s 6GB @ 336GB/s
1825MHz clock speed (locked)
CPU @ 3.8GHz (without SMT) 3.6GHz (with SMT) both locked
Mesh Shaders, VRS Tier 2, Sampler Feedback Streaming, Ray Tracing, Hardware Machine Learning

The contention some are making is that the 12TFLOPs GPU of the Series X, with all the above features, doesn't command as much of a performance edge over the PS5 as many suggested it would. The problem with any such assumption is that it's largely premature since much of what's listed above from an advanced graphics performance feature standpoint, hasn't sniffed, let alone even appeared, in a game this early into the gen. Ray Tracing we know harms performance, and knew that entering the gen, so I won't label that a performance enhancer. Every other feature supported by the Series X GPU is designed to offer enhanced performance.

We have one instance total of a released game for both systems where one of these performance enhancing features are in use, just one. Doom Eternal. It used not the inferior version of VRS, the one which can be software emulated, but the more advanced hardware Tier 2 VRS. What was the performance outcome in what we know to be one of the most advanced and graphically impressive game engines out there?





That's a 23% performance advantage for Series X in balanced mode



That's a 14% performance advantage for Series X in RT mode.



That's a 29% performance edge for Series X in 120fps mode.




Then there's Avengers, another visually advanced looking game.

In the Quality mode where both consoles used dynamic res with native resolutions, Series X maintained a 13% performance advantage.

In the performance mode it's an even more insane 62% performance advantage for the Series X in terms of performance because the PS5 is literally rendering way less pixels. If we go by Digital Foundry's pixel count, then it's an even higher 74% edge. Now, I'll go with VGTech simply because they provide way more data compared to Digital Foundry.





Now something I've always been curious about is this. We know both consoles turn in excellent performance figures in avengers. Neither console is performing badly, which the stats back up. But we also know the PS5 is rendering, at any given moment, many less pixels compared to the Series X. At minimum it appears 62% less. And even then we get these kinds of stats.

ZvDlTem.jpg



You would think that if the PS5 in this particular game is rendering so many less pixels, how is it possible that its performance can be THIS close to the Series X, and not be pretty much flawless? It suggests that if the PS5 were anywhere close to the native resolutions of the Series X, it would performed quite a bit worse. Or, simply put, the resolution difference would have needed to have been even lower. The PS5's maximum checkerboard resolution of 3840x2160 works out to 1920x2160, that's 12.5% more pixels than 1440p. Series X's maximum native 4K is literally 100% more pixels by comparison. Yes, the Series X drops from that native 4K more often than the PS4 drops from its half checkerboard 4K, but at worst the Series X advantage is in the realm of 62% more resolution performance. But to be more fair, 50% more resolution if you leave PS5 at its version of 4K and instead use a lower native pixel count for Series X under its native 4K.

This makes what Crystal Dynamics told NXGamer even more valid. Had they not gone with Checkerboarding, they would have ended up with potentially worse image quality or something softer in appearance overall. Checkerboarding, even with its imperfections, guaranteed a better visual outcome than if they hadn't gone with it. Now we know CD takes their performance seriously, so clearly the game wouldn't have performed like shit at native, because they wouldn't have allowed it to. They would have simply run at a lower overall resolution to maintain the performance, and it would looked worse next to the Series X version. So they made the smartest decision. And just in case someone tries to claim PS5 has better overall performance because Series X had the lowest minimum framerate, which appears less than a quarter of 1% of the time

4L1QSIk.jpg




Finally, another highly visually advanced title on a game engine that's no joke.

Metro Exodus, yet another demanding game pushing visuals, stop me if you've heard this one before, but Series X is maintaining roughly 21-23% more performance in resolution. And no, PS5 maintaining 60fps less than 1% of the time over Series X does not constitute a performance win for the PS5 when it's already running at a lower resolution and dropping to lower than 1080p in more demanding scenes.




And finally, though I won't waste time posting it, but you can go look at the stats if you don't believe me, Resident Evil Village actually performs better on Series X Better than 2% of the time in RT mode. Both consoles are flawless in non RT mode, not a single drop according to the stats. And due to how fantastic Capcom's checkerboarded solution is it's near damn impossible to get a proper resolution count on the game, so inconclusive so we'll assume they must have been 100% identical. There are some titles where despite having a resolution advantage Series X's performance is just unacceptable, that's a PS5 win. The 4 games I just ticked through are not one of them.

Another game, the next gen version of Star Wars Jedi: Fallen Order is also again better on Series X, but that one is much closer. A roughly 8% resolution edge with less than 1%, not even a tenth of 1% framerate edge going to PS5 in both modes. But admittedly that game is far less, to practically nothing, to bark about because they are just that damn close.

Series X has more convincing cases, in more visually demanding games on very advanced engines. And guess what? That advantage is often indeed precisely the GPU performance advantage that it's been suggested to have. There are times it seems to almost exceed it. Feel free to look on the PC side of things for nearly matching GPU configurations and you will notice generally similar or smaller percentage advantages in framerate performance at identical resolutions. Translation: in console terms the stronger GPU maintaining higher fps at a particular resolution would be the one running at the higher resolution compared to the weaker GPU running at a lesser resolution for performance reasons, and in so doing they would get roughly equal framerates, exactly what you're seeing between PS5 and Series X. A familiar trend you will often see if you look at these PC game benchmarks is that often the weaker GPU running at the lower resolution will see generally better framerates than the stronger GPU was getting at its higher resolution, forcing the stronger chip to come down from its perch just a little bit to get closer to performing like the weaker card, just at a higher resolution because that's what it should be doing.

Just for fun people can use the 2080 Super to represent Series X and the 2070 Super or 5700XT to represent PS5 (differences in hardware and core makeup are roughly close to PS5 and Series X). What you'll often see is that the lower resolution framerate performance of weaker cards tend to be better than their more capable counterparts running at the higher resolutions. This would mean that the two cards through dynamic resolution would need to draw closer in resolution to meet their targets


So if people want to label Touryst a benchmark, then we can show benchmarks for much more graphically demanding titles developed under similar circumstances and timeframes on more advanced game engines that demonstrate which platform is more capable. I'm unsure how a game that released nearly an entire year after the Series X launch with an entirely rewritten game engine and the added benefit of a lot more time under their belt (things Xbox couldn't benefit from) can be considered a fair benchmark of the capabilities of the two systems, but that won't stop people from trying to hold it up as evidence of platform capability. Then again, we certainly know people on the Xbox side would have done the same if the roles were reversed, so I guess all is fair. But there's a shiny asterisk all over this one.

You Know It GIF by MOODMAN

Doom Eternal is often used as a comparison but it doesn't look like the same IQ to me. Please compare the two images of the real machine. I don't understand why the resolution is so high but it looks blurry.(Both are RT modes.)Is it due to some setting other than resolution?
W8gG9wW.jpg
PepcFIv.jpg
 

onQ123

Member
RX 6700 XT would probably give us the answer if it's the ROPS holding the Series X back in this game because it has the same RBE setup with 64 Color ROPS & 128 Depth ROPS , It's clocked higher & has infinity cache but running this game in 8K on it should give us a little info.
 
it isn't, but if you work on the same game again and again, you will most likely find ways to improve it over time.
the question is if they could improve the Xbox version if they wanted to, which is hard to say.

this could very well be a game where the PS5 simply has an advantage due to the fact that the way the renderer works in this game is very unusual compared to most modern games.

what could also be the case is that they early on had 6K as a goal on the Xbox version
it's a nice and round resolution to aim for after all. and there is most likely headroom above that on Xbox even as it is now (would be cool if they had an FPS unlock setting so we could see just how much headroom there is)
so they might have hit that target and didn't even try to push for 7K or 8K, it's only used for Antialiasing after all.

it's hard to say without a dev speaking in detail about the differences in play here. and the devs themselves might also not really know how far they could push the Xbox version if they really targeted 6K from the start and were satisfied with that once they had it running locked at 60fps

Developers only mentioned higher clocks and the memory setup. Seems like it's something they found they could do with their engine based on the hardware that they had.

That's all we know from the developers themselves.
 
RX 6700 XT would probably give us the answer if it's the ROPS holding the Series X back in this game because it has the same RBE setup with 64 Color ROPS & 128 Depth ROPS , It's clocked higher & has infinity cache but running this game in 8K on it should give us a little info.
Sadly nobody seems to actually own this game so we'll never know.
 

Loxus

Member
Am I tripping or does that show PS5 with 72 Color ROPS 288 Depth I ROPS , Xbox Series X 64 Color ROPS 128 Depth ROPS ?


Locuza said 72 cROPs in total, 64 are active (16RBs from 18).
I think he came to that conclusion because ROPx4 are grouped in 4s. 16/4 = 4 clusters, with 2 ROPx4 disabled for yields.

I've searched high and low for GPUs that disable ROPs for yields and couldn't find anything. Maybe other can provide more information on this. Also, it seems like a waste of space to include extra ROPs for yields.

Navi 10 didn't have any extra RPOs for yields.
wltzZIM.jpg


But Navi 14 did had one extra ROPx4.
ZDMmZ3N.jpg


So maybe it's for yields or Mark Cerny found a use for the extra ROPs and made it work.
wxPp6qY.jpg


But I'm going with Locuza with 64 active ROPs.
 
Last edited:

Lysandros

Member
While both machines have the same number of color ROPs, PS5 has twice more Depth / stencil ROPs than XSX. PS5 ROPs take almost twice more space on the APU than XSX ROPs.

Locuza wondered if that would bring some benefits in some games. Well I think we have found one of those games with The Touryst. A game heavily limited by pixel throughput performing almost twice better on a machine having almost twice more ROPs?




I think this can be even more important than the frequency difference to fill rate performance, it would be pretty illogical to expect the same throughput with 50% less RBE hardware. This looks like to be a clear compromise to save die space to add few more CUs reach to the most wanted 12 TF. It's a trade-off for more compute at the expense of fill rate. It seems PS5's advantage here is more substantial than we thought in begining of the generation.
 

onQ123

Member


Locuza said 72 cROPs in total, 64 are active (16RBs from 18).
I think he came to that conclusion because ROPx4 are grouped in 4s. 16/4 = 4 clusters, with 2 ROPx4 disabled for yields.

I've searched high and low for GPUs that disable ROPs for yields and couldn't find anything. Maybe other can provide more information on this. Also, it seems like a waste of space to include extra ROPs for yields.

Navi 10 didn't have any extra RPOs for yields.
wltzZIM.jpg


But Navi 14 did had one extra ROPx4.
ZDMmZ3N.jpg


So maybe it's for yields or Mark Cerny found a use for the extra ROPs and made it work.
wxPp6qY.jpg


But I'm going with Locuza with 64 active ROPs.

They had extra ROPS on PS4 & PS4 Pro for VR I guess so maybe they added extra ROPS for PS5 VR
 

onQ123

Member
I was just coming to add this.
The 1080TI has 88 ROPs, so I'm gonna go out on a limb and say the PS5 does have 72 active ROPs.

Have not found anything about disabling ROPs for yields.
Yeah I couldn't find anything about disabled ROPS from the past also why would they need extra ROPS for yields when PS5 is using the RBEs from the last generation of cards & shouldn't have any problems with yields.
 
I think this can be even more important than the frequency difference to fill rate performance, it would be pretty illogical to expect the same throughput with 50% less RBE hardware. This looks like to be a clear compromise to save die space to add few more CUs reach to the most wanted 12 TF. It's a trade-off for more compute at the expense of fill rate. It seems PS5's advantage here is more substantial than we thought in begining of the generation.

Don't see here what is the problem in fact. The new RB+ unit is designed to double the fillrate compared to the previous RB used in NAVI1 GPU and in the PS5. You can use two time less RB to reach the same pixel fillrate. The "gap" between the XsX and the PS5 will be only due to the frequency difference.
We have a good example : The 6700XT uses this new RB+ have the same number of Compute Units, total ROPs, and TMUs than the 5700XT which include the previous RB. The 5700XT has two time more RB units than the 6700XT but with the double number of pipeline included in the new version, they have the same theoretical pixel fillrate when running at the same frequency (and 30/35% higher when running at their "normal" frequency"). That's confirmed with benchmarks done with such setting, they offer really similar performances when put at the same frequency with all game tested (techspot for example has done a clock-for-clock performance test with these two GPUs running at 1.8GHz)
Same situation than the PS5 vs XsX in fact.
 
Last edited:

onQ123

Member
Don't see here what is the problem in fact. The new RB+ unit is designed to double the fillrate compared to the previous RB used in NAVI1 GPU and in the PS5. You can use two time less RB to reach the same pixel fillrate. The "gap" between the XsX and the PS5 will be only due to the frequency difference.
We have a good example : The 6700XT uses this new RB+ have the same number of Compute Units, total ROPs, and TMUs than the 5700XT which include the previous RB. The 5700XT has two time more RB units than the 6700XT but with the double number of pipeline included in the new version, they have the same theoretical pixel fillrate when running at the same frequency (and 30/35% higher when running at their "normal" frequency"). That's confirmed with benchmarks done with such setting, they offer really similar performances when put at the same frequency with all game tested (techspot for example has done a clock-for-clock performance test with these two GPUs running at 1.8GHz)
Same situation than the PS5 vs XsX in fact.
PS5 also has double the depth ROPS & it's 72 Color ROPS
 

Boglin

Member
Don't see here what is the problem in fact. The new RB+ unit is designed to double the fillrate compared to the previous RB used in NAVI1 GPU and in the PS5. You can use two time less RB to reach the same pixel fillrate. The "gap" between the XsX and the PS5 will be only due to the frequency difference.
The PS5 has 16 RB units with 4 Color ROPs and 16 Z/Stencil ROPs each giving it 64 Color ROPs and 256 Z/Stencil ROPs total.
The XSX has 8 RB+ units with 8 Color ROPs but still only has 16 Z/Stencil ROPs per RB unit giving it 64 Color ROPs and 128 Z/Stencil ROPs.

If the benchmarks you're referencing are testing the fillrate then of course the only difference will be due to the frequency they are running at and the PS5 would be running only 22% faster. However the PS5 is capable of doing 150% more Z/stencil operations.

If those 2 additional RB units seen on the PS5 aren't actually disabled then then numbers change to it having 18 RB units with a total of 72 Color ROPs and 288 Z/Stencil ROPs. In comparison to the XSX in this scenario, the PS5 has 38% faster fillrate and 177% more Z/stencil operations.

I don't actually know what kind of difference this would make in modern engines and it's especially curious because the Unreal 5 engine actually focuses on software rasterization which pushes work to the CUs.
 
Last edited:
Don't see here what is the problem in fact. The new RB+ unit is designed to double the fillrate compared to the previous RB used in NAVI1 GPU and in the PS5. You can use two time less RB to reach the same pixel fillrate. The "gap" between the XsX and the PS5 will be only due to the frequency difference.
We have a good example : The 6700XT uses this new RB+ have the same number of Compute Units, total ROPs, and TMUs than the 5700XT which include the previous RB. The 5700XT has two time more RB units than the 6700XT but with the double number of pipeline included in the new version, they have the same theoretical pixel fillrate when running at the same frequency (and 30/35% higher when running at their "normal" frequency"). That's confirmed with benchmarks done with such setting, they offer really similar performances when put at the same frequency with all game tested (techspot for example has done a clock-for-clock performance test with these two GPUs running at 1.8GHz)
Same situation than the PS5 vs XsX in fact.
It's more complicated than that. They put twice more color ROPs by RB units (your "pipeline"), but half depth/stencil inside them. So they have the same pixel fillrate (done using color ROPs which is the main job to do if you will), but ROPs are more than that with Depth ROPs used for several jobs related to final rasterization (that can be very important in some scenes).

XSX is also the only RDNA AMD GPU with so many CUs in each shader arrays (that could lead to CU inefficiency in some game related tasks, notice that all 3 GPUs have the same L1 cache size that have to feed all CUs).

6JAYS8f.png
 
PS5 also has double the depth ROPS & it's 72 Color ROPS

The PS5 has 16 RB units with 4 Color ROPs and 16 Z/Stencil ROPs each giving it 64 Color ROPs and 256 Z/Stencil ROPs total.
The XSX has 8 RB+ units with 8 Color ROPs but still only has 16 Z/Stencil ROPs per RB unit giving it 64 Color ROPs and 128 Z/Stencil ROPs.

If the benchmarks you're referencing are testing the fillrate then of course the only difference will be due to the frequency they are running at and the PS5 would be running only 22% faster. However the PS5 is capable of doing 150% more Z/stencil operations.

If those 2 additional RB units seen on the PS5 aren't actually disabled then then numbers change to it having 18 RB units with a total of 72 Color ROPs and 288 Z/Stencil ROPs. In comparison to the XSX in this scenario, the PS5 has 38% faster fillrate and 177% more Z/stencil operations.

I don't actually know what kind of difference this would make in modern engines and it's especially curious because the Unreal 5 engine actually focuses on software rasterization which pushes work to the CUs.

I know for the Z/Stencil ROPs number, but if honestly the division by two of these ROPs is going to be a limiting factor, do you really think AMD would have applied this change on their latest big GPUs ? I mean, the 6900XT has the same total Z/Stencil ROPs compared to the 5700XT (256 for each)....
From what I remember, having too much Z/Stencil ROPs compared to the number of triangle that has been rasterized is useless. The Z/Stencil operation/test is done to reject pixels and avoid useless calculation at shader level etc... The ROPs unit are very limited by your memory bandwidth, and Z/Stencil operations are the worst (Z compression has helped but more use to help the AA tool such as MSAA etc...).
I think AMD has done the choice to reduce the number of Z/stencil ROPS per RB unit in the RX6xxx gpu and in the XsX simply because it's not the bottleneck that will appear in first in your pipeline to generate your frame.
 
It's more complicated than that. They put twice more color ROPs by RB units (your "pipeline"), but half depth/stencil inside them. So they have the same pixel fillrate (done using color ROPs which is the main job to do if you will), but ROPs are more than that with Depth ROPs used for several jobs related to final rasterization (that can be very important in some scenes).

XSX is also the only RDNA AMD GPU with so many CUs in each shader arrays (that could lead to CU inefficiency in some game related tasks, notice that all 3 GPUs have the same L1 cache size that have to feed all CUs).

6JAYS8f.png

I know ;) But the RX6xxx GPU has the same RB unit, and clearly, no game are showing a limitation compared to the RX5xxx (as I said in my other answer, RX6900 has the same Z/stencil ROPS number than the RX5700). I think that's not a real problem, and the AMD engineers has done this change on purposes ;)
 
Last edited:

Boglin

Member
I know for the Z/Stencil ROPs number, but if honestly the division by two of these ROPs is going to be a limiting factor, do you really think AMD would have applied this change on their latest big GPUs ? I mean, the 6900XT has the same total Z/Stencil ROPs compared to the 5700XT (256 for each)....
From what I remember, having too much Z/Stencil ROPs compared to the number of triangle that has been rasterized is useless. The Z/Stencil operation/test is done to reject pixels and avoid useless calculation at shader level etc... The ROPs unit are very limited by your memory bandwidth, and Z/Stencil operations are the worst (Z compression has helped but more use to help the AA tool such as MSAA etc...).
I think AMD has done the choice to reduce the number of Z/stencil ROPS per RB unit in the RX6xxx gpu and in the XsX simply because it's not the bottleneck that will appear in first in your pipeline to generate your frame.

It wasn't my intention to proclaim that the z/stencil will make some radical difference. From your description it sounded like the only difference between the two RBs was that the XSX was twice as space efficient while running at a lower frequency. I just wanted to clarify that there was more to it than that.

The difference in hardware is there. Whether it will be utilized or not is a different matter.
 

Lysandros

Member
I was just coming to add this.
The 1080TI has 88 ROPs, so I'm gonna go out on a limb and say the PS5 does have 72 active ROPs.

Have not found anything about disabling ROPs for yields.
So RX 5700XT (which is quite similar to PS5) has 64 ROPs without any disabled units then, correct?
 

Loxus

Member
I know ;) But the RX6xxx GPU has the same RB unit, and clearly, no game are showing a limitation compared to the RX5xxx (as I said in my other answer, RX6900 has the same Z/stencil ROPS number than the RX5700). I think that's not a real problem, and the AMD engineers has done this change on purposes ;)
The PS5's GPU is custom, they can +/- anything they want. Cache Scrubbers for example.
 
Top Bottom