• 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.

Matt weighs in on PS5 I/O, PS5 vs XSX and what it means for PC.

THE:MILKMAN

Member
We have yet to see streaming per frame file IO capability other than the UE5 demo, streaming per frame in ms is allot more than streaming in advance X seconds of expected gameplay which just needs a fast SSD and not the full IO optimisations..

True. We just need to see actual game footage showing the advances and/or differences.
 

killatopak

Member
Funnily this is one area that doesn't get talked about enough.
That’s because we already know what Raytracing does.

Part of the reason I think SSDs is talked about is because we actually haven’t seen the big revolutionary stuff these devs keep talking about. The unknown is just more exciting to contemplate about. Like actually play with it. Yes there is the Unreal demo but that’s exactly what it is, a demo.
 

Thirty7ven

Banned
He's probably talking about Ray-Tracing. And as Dictator from DF has said, the Series X has a massive advantage in compute units, which Ray-Tracing is directly tied to. Funnily this is one area that doesn't get talked about enough.

There's going to be a noticeable difference between both machines 3rd party games. And it's not going to come from the SSD. I'm talking about clear graphical differences from better effects and Ray-Tracing. Interesting times ahead.

It’s also tied to clock speed. It’s just people trying really hard at widening the GPU gap because 2 Tflops were a really big deal until people realized it was just a 15%-18% difference.

It gets talked plenty, but no matter how hard you keep looking for it, it isn’t there.

There’s no massive difference and if there were, you would be hearing it from devs already. We aren’t.
 

MaulerX

Member
That’s because we already know what Raytracing does.

Part of the reason I think SSDs is talked about is because we actually haven’t seen the big revolutionary stuff these devs keep talking about. The unknown is just more exciting to contemplate about. Like actually play with it. Yes there is the Unreal demo but that’s exactly what it is, a demo.


I was referring to the performance delta between the two machines respective capabilities. You only seem to hear the PS5's SSD advantage. Even though those same devs quietly acknowledge that it won't do much with regards to third party multiplatform games. Whereas the one area that will yield actual visible graphical differences isn't talked about enough (at least not by the people that believe that an SSD advantage will miraculously mitigate the disadvantage in TF's, Bandwidth, CPU, GPU, Compute Units/Ray-Tracing). Although I'm sure they will once the DF comparison videos start rolling out. The Series X's massive compute unit advantage (in which Ray-Tracing is directly tied to) is probably the biggest differentiator between the two.
 
He's probably talking about Ray-Tracing. And as Dictator from DF has said, the Series X has a massive advantage in compute units, which Ray-Tracing is directly tied to. Funnily this is one area that doesn't get talked about enough.

There's going to be a noticeable difference between both machines 3rd party games. And it's not going to come from the SSD. I'm talking about clear graphical differences from better effects and Ray-Tracing. Interesting times ahead.
Talking out your ass?
A developer already stated the differences in 3rd party titles will be modest and youd need Digital Foundry type analysis to spot them.
So who to believe. A random forum poster or a developer working on both next gen consoles.
 
Last edited:

MaulerX

Member
It’s also tied to clock speed. It’s just people trying really hard at widening the GPU gap because 2 Tflops were a really big deal until people realized it was just a 15%-18% difference.

It gets talked plenty, but no matter how hard you keep looking for it, it isn’t there.

There’s no massive difference and if there were, you would be hearing it from devs already. We aren’t.



More like people trying really hard to narrow the gap. That 2TF's is bare minimum and is actually greater when you account for the PS5's "variable clocks". You just can't ignore the massive compute unit advantage in which Ray-Tracing is directly tied to. Higher clocks will not mitigate that.


If you repeat that every day, once during the morning and twice before bedtime, it might become true.


One look at the spec sheet is enough. :cool:



Talking out your ass?
A developer already stated the differences in 3rd party titles will be modest and youd need Digital Foundry type analysis to spot them.
So who to believe. A random forum poster or a developer working on both next gen consoles.


If there's one thing I've noticed is that different devs have been saying different things. Contradicting statements even. At the end we'll just have to wait and see how/if each machine's respective strengths materialize into anything tangible.
 
Last edited:

yurinka

Member
True. We just need to see actual game footage showing the advances and/or differences.
Well, the final version of UE5 won't be released until late 2021, so add a couple of years after that until we see games released taking proper advantage of the new stuff UE5 will offer. Late 2023 maybe. Consider also that during the first couple of years of the generation or so, many 3rd party games will be crossgen. So anyways they won't use these features at all to make sure the game works well in the previous gen.

At least it's what MS announced they will do with the games they will publish. Sony instead said they will bet on PS5 only games, so I think we may see stuff like that before in PS5 exclusives, even if some early PS5 Sony game may be some delayed PS4 project.
 

killatopak

Member
Well, the final version of UE5 won't be released until late 2021, so add a couple of years after that until we see games released taking proper advantage of the new stuff UE5 will offer. Late 2023 maybe.
Won’t take that long. Epic already said games in development for UE4 can easily be imported to UE5.
 

Panajev2001a

GAF's Pleasant Genius
That 2TF's is bare minimum and is actually greater when you account for the PS5's "variable clocks".
It may go from 18% in 99% of the cases to 18.08% in the remaining ones... maybe you do want to read this reasonable explanation of PS5’ssolution: https://www.resetera.com/threads/pl...ve-ot-secret-agent-cerny.175780/post-36022468

You just can't ignore the massive compute unit advantage in which Ray-Tracing is directly tied to. Higher clocks will not mitigate that.

Nobody is ignoring that: compute is per clock throughout * clockspeed... it seems like you are the one ignoring it and all the common fixed function hardware outside of the CU which will run faster on average on PS5 :). XSX won the TFLOPS war, get over it ;).
 
Last edited:

Shmunter

Member
He's probably talking about Ray-Tracing. And as Dictator from DF has said, the Series X has a massive advantage in compute units, which Ray-Tracing is directly tied to. Funnily this is one area that doesn't get talked about enough.

There's going to be a noticeable difference between both machines 3rd party games. And it's not going to come from the SSD. I'm talking about clear graphical differences from better effects and Ray-Tracing. Interesting times ahead.
Regarding RT, it doesn’t get talked about much because the difference is no bigger than the compute difference.

XsX can send out more ‘rays’ per cycle, but PS5 does more cycles in the same amount of time. If both were at the same clock speed, there would be a conversation to be had, otherwise it doesn’t even inspire the most basic of tech conversations.
 

Md Ray

Member
It may go from 18% in 99% of the cases to 18.08% in the remaining ones... maybe you do want to read this reasonable explanation of PS5’ssolution: https://www.resetera.com/threads/pl...ve-ot-secret-agent-cerny.175780/post-36022468



Nobody is ignoring that: compute is per clock throughout * clockspeed... it seems like you are the one ignoring it and all the common fixed function hardware outside of the CU which will run faster on average on PS5 :). XSX won the TFLOPS war, get over it ;).
That's a great fucking post by Liabe Brave.
 

quest

Not Banned from OT
It may go from 18% in 99% of the cases to 18.08% in the remaining ones... maybe you do want to read this reasonable explanation of PS5’ssolution: https://www.resetera.com/threads/pl...ve-ot-secret-agent-cerny.175780/post-36022468



Nobody is ignoring that: compute is per clock throughout * clockspeed... it seems like you are the one ignoring it and all the common fixed function hardware outside of the CU which will run faster on average on PS5 :). XSX won the TFLOPS war, get over it ;).
Its really poor it ignores the cpu side of things that the gpu is using to get its clocks. Its just sony fantasy land where they invented this perfect technology lol. Just more astroturfing because Sony won't put out a white paper on this. No just cutting the peak power does not do shit except predictable cooling. Stealing power budget from the cpu allows the clocks. There is a reason the dev kit has power profiles for the gpu and cpu because one wins out. Devs are choosing the gpu.
 
I'm not saying the SSD will do the raytacing calculations for the gpu or that it will process the sound for the cpu.

Although technically, I suppose if you can drastically lower your vram usage once you can rely on on being able to load a certain amount of data per frame. You could use the now free vram on the GPU to raise the quality of may effect including shadow quality and ray-tracing. So we will have to see if that "I'm the GPU now" meme actually ends up holding some weight
[
/quote]

This would apply with both systems however, so seeing it usually only used in reference to PS5 is baffling. Depending on what extent the ML customizations have been done on XSX's GPU they could use a combination of that and the SSD I/O to achieve effective parity with what Sony's SSD I/O alone seems to provide. For most use-cases, anyhow.

Point is speculation based on understanding the tech is still just speculation, Devs are speculating with more information than you and tons of this speculation will be wrong, you need real testing to find out what you are now capable of.

Not every developer has full access to information on the systems, either, and I'd be comfortable saying most of them don't 100% know what the systems are capable of since they haven't had enough time. That literally takes years.

You're right it's still speculation, but at least it's informed speculation. None of the "16 GB GDDR6, 16 GB HBM2, 128 GB ReRAM dual-GPU 518-bit bus" Reddit or 4Chan garbage that was gaining traction during earlier next-gen speculation.

You say the SSD is subservient to the GPU. But the GPU is often just as subservient to the SSD. The longer the GPU has to wait to use data that isn't in memory the less options we have. The SSD may not be actively doing things as often but it's often just as big of a factor in whether you can achieve the art you want or not. A slower SSD could be exactly what prevents you from doing an extremely dynamic set piece with the quality of assets that you desire.

Well, it's arguable that smart game design does more with less data. Realistically you'd want to cut down on as many accesses to storage as possible, though if the storage solution is good enough it does open up possibilities of smart game design around it. The thing though is that, again, both systems benefit from this, but it doesn't change the fact the in terms of overall grunt work, the GPU is doing much more than the SSD.

Beyond that, it's more arguable that the GPU is dependent on the RAM more than storage, since ultimately it's the data in RAM that it's relying on the vast chunk of the time to operate with. Any system having a very fast storage solution can of course feed data to the RAM that much quicker assuming other hardware is present to accomplish this, but it isn't the only method to reducing data footprint for "just-in-time" usage in RAM. Other methods may be quirkier and have to exploit specific features of the target hardware, but they are technically possible.

Dude can this site stop fuccing with my post formatting please?
 
Last edited:

Panajev2001a

GAF's Pleasant Genius
Its really poor it ignores the cpu side of things that the gpu is using to get its clocks. Its just sony fantasy land where they invented this perfect technology lol. Just more astroturfing because Sony won't put out a white paper on this. No just cutting the peak power does not do shit except predictable cooling. Stealing power budget from the cpu allows the clocks. There is a reason the dev kit has power profiles for the gpu and cpu because one wins out. Devs are choosing the gpu.

:LOL:, yeah really poor haha....

It does not do 100% of the homework for you, it may require 0.2% of thinking 🤔 ... the CPU vs GPU power allocation is a logical extension of it: both CPU and GPU have the same setup to control dynamic frequency, but there is a mechanism where, when the CPU is not needing all of its power budget, the CPU can transfer a bit of power budget to the GPU. Are there more levels that can be pulled, likely, is this cause of “concern”? Nope...

Anyways, the relationship between power and frequency is still the same non linear one that Cerny mentioned (likely it under colts the unit while it is lowering the clockspeed).
 
Last edited:

Shmunter

Member
Its really poor it ignores the cpu side of things that the gpu is using to get its clocks. Its just sony fantasy land where they invented this perfect technology lol. Just more astroturfing because Sony won't put out a white paper on this. No just cutting the peak power does not do shit except predictable cooling. Stealing power budget from the cpu allows the clocks. There is a reason the dev kit has power profiles for the gpu and cpu because one wins out. Devs are choosing the gpu.
I expect the GPU to continue being tilted towards with the smartshift tech. Especially with all I/o and sound processing being offloaded onto custom processors, there is ample general processing available. Especially in contrast to teh JaGuarSzz.
 
Last edited:
Its really poor it ignores the cpu side of things that the gpu is using to get its clocks. Its just sony fantasy land where they invented this perfect technology lol. Just more astroturfing because Sony won't put out a white paper on this. No just cutting the peak power does not do shit except predictable cooling. Stealing power budget from the cpu allows the clocks. There is a reason the dev kit has power profiles for the gpu and cpu because one wins out. Devs are choosing the gpu.

So they're basically saying it's analogous to a limiter in audio engineering with the peak power "spikes" equivalent to high decibel spikes that cause distortion. You use limiters to bring the level of the loudest parts down outside of that peaking range (when mastering audio production you usually don't want the loudest parts peaking "into the red" because that might signal sound distortion), and that poster is basically trying to say Sony've done something similar with their power solution (which, given their history with audio, I guess makes some sense).

In any case, I definitely can see their setup presenting programming challenges later in the generation since devs will have to be very mindful of what type of power consumption their code, algorithms etc. are going to use, and how it will use it. And it seems like it'll provide more of a challenge down the line than the "split" 10 GB/6 GB pools of the XSX as conceptually (and in practice) it's a much smaller and more historically familiar compromise.
 

longdi

Banned
It may go from 18% in 99% of the cases to 18.08% in the remaining ones... maybe you do want to read this reasonable explanation of PS5’ssolution: https://www.resetera.com/threads/pl...ve-ot-secret-agent-cerny.175780/post-36022468

Thats the thing, modern gpu, cpu works this way. I bet.
Is there any evidence that PS5 gpu is much special in terms of activity based loading? Did any developers chime in, all i see are just continued stream of random posters astroturfing it as a fact or exclusive to PS5.
Seems like we have another movement besides the 'micro-center' one.

Isnt it funny the lead engineer of PS5, whose explanation seems half a dozen and confusing when it was discussed the previous month.
So we needed to rely on some random internet posting to convince us of PS5 high 2.23ghz possibility? 🤷‍♀️
 
Last edited:
For "reasons" I feel like it's not even Sony's technology:

Sony's definitely leveraging Smartshift, but it's just one component of their variable frequency setup. The other major component is their cooling system. We know now what the devkit was using for the cooling system (six fans sounds kind of comical in isolation though, ngl), but the final retail system might be using something different for all we know.

It better be damn good, though.
 

Panajev2001a

GAF's Pleasant Genius
Thats the thing, modern gpu, cpu works this way.
Is there any evidence that PS5 gpu is much special in terms of activity based loading? Did any developers chime in, all i see are just continued stream of random posters celebrating it as a fact.
Isnt it funny the lead engineer of PS5, whose explanation seems half a dozen and confusing when it was discussed the previous month. So we needed to rely on some random internet posting to convince us of PS5 high 2.23ghz possibility? 🤷‍♀️
Yes there are devs there too, you do your homework to find ways to ignore what they say or try to discredit the. Other systems do not have the same constraints and design goals, so they choose different paths to achieve it. So?

It appears you are now shifting the goalpost to a defender someone can sum up as “I could not understand it at all the first time it had been explained, so the technology can not possibly be any good” which if you were a console lead system architect level engineer would almost make sense (it implies nobody can think of novel ways to do things you did not think before yourself).
 

longdi

Banned
Not some press, EVERYONE.

People judge what they see, the result.

The main fact is last gen was ~ 2 TF and 8 GB RAM, a ratio of 4

This gen, all being equal., 10 TF should have 40 GB of RAM. 12 TF 48 GB. I guess OS should not increase much, but if assets do then even 50 GB does not cut it.

So the new consoles, to do higher detail, are massively short of RAM, that is the deficiency. You either add 100 GB RAM or stream data in every frame as fast a spossible for balance.

Or you keep the same detail as current gen and add resolution / frame rate.

Since you are just assuming, so can i.

What makes you think MS engineers didnt think through the synergy between 2.4gbs SSD + the 12tf power? I think they designed Series X for $499, and wanted to make the best next gen experience.
While Sony wanted to hit $399. Can their SSD somehow able to close the gap between a $100 difference in raw materials? If im a betting man, and looking at MS efforts, i put my money on the 2.4gbs + 12tf performance to give a 25% superiority. :messenger_bicep:
 

longdi

Banned
Yes there are devs there too, you do your homework to find ways to ignore what they say or try to discredit the. Other systems do not have the same constraints and design goals, so they choose different paths to achieve it. So?

It appears you are now shifting the goalpost to a defender someone can sum up as “I could not understand it at all the first time it had been explained, so the technology can not possibly be any good” which if you were a console lead system architect level engineer would almost make sense (it implies nobody can think of novel ways to do things you did not think before yourself).

Nope, when i read it yesterday, no real developers chimed in to agree this is an exclusive hw customisation to PS5.
I understand it just fine, i dont see what my 1080Ti isnt doing it now, or my 3950x isnt either.

Lets hope DF next chat with Mark or Phil, are allowed to ask sensitive questions, go for jugular on the smartshift, highclocks and ssd compression. 🤷‍♀️
 
Last edited:

geordiemp

Member
Since you are just assuming, so can i.

What makes you think MS engineers didnt think through the synergy between 2.4gbs SSD + the 12tf power? I think they designed Series X for $499, and wanted to make the best next gen experience.
While Sony wanted to hit $399. Can their SSD somehow able to close the gap between a $100 difference in raw materials? If im a betting man, and looking at MS efforts, i put my money on the 2.4gbs + 12tf performance to give a 25% superiority. :messenger_bicep:

I think Ps5 can stream data per frame like the UE5 demo, and games that do made for ps5 only from sony will look a gen above anything on XSX.

There will be 2 classes of game visuals next gen.

Do you think 20 TF would of been able to run that UE5 demo at that detail - no.
 
Last edited:

Exodia

Banned
Uh huh.... So would you mind posting your verfiable facts for this claim then? I'm genuinely curious...

I already did:

 

Goliathy

Banned
I think Ps5 can stream data per frame like the UE5 demo, and games that do made for ps5 only from sony will look a gen above anything on XSX.

There will be 2 classes of game visuals next gen.

Do you think 20 TF would of been able to run that UE5 demo at that detail - no.

What are you smoking? Lol. You only need a notebook to run the UE5 demo with the same amount of detail at 1440p@40fps. XSX will easily run it at a higher res. Stop being to delusional.
And STOP with this stupid narrative that XSX means last-gen jaguar based games and PS5 is the only platform on this planet to have nextgen visuals.
This is ridiculous.

Your thinking of streaming like current gen, with spiderman the buildings block the view and streaming in blocks worth of data as you move holding all the block of the city in the available RAM. Think of the blocks as levels.

If the game does the same then yes all SSD will be fine, but you still dont get it, that is loading the level by blocks.

Ps5 can load the data it needs each frame , if the game is made for that.

Then there is RAM, Ps4 Pro and Xb1X current half gen have 12 GB, next gen is only 16 GB, good luck fitting high quality assets into that every 10 seconds or whaetver the game "block size and timing is".

But the disk storage you ask, currently 45 GB typical for say HZD1, add some kraken compression and 100 GB disks (or 2), NO dublicate data and you have got 3-4 x asset quality potential.

All of this doesn't matter at all. Developers will not create based on a single SSD. They will use the PC SSD as the lowest common denominator and that's it. So, we will see better loading times, but that's it.
Again, UE5 demo could run on a notebook - according to epic engineers in china, at 40fps.

I don't know about that, the Series X is well put together and custom. But PS5 is apparently more efficient, and I'm not just talking about the SSD. I wouldn't be surprised if some early multiplatform games perform better on PS5 because of its commitment to efficiency and removal of all bottlenecks.

We had a dev on here (BGs) that went as far to say that the PS5 has achieved 'perfect efficiency'. TFLOPs just being one theoretical measurement of a system which doesnt translate to real performance is a truthbomb that will hit many hard when the games and certainly multiplats are eventually compared.

what? no, XSX has way more TFLOps, and it's sustained 100% of the time. PS5 clocks are variable, I think Devs will just use the base/lowest possible clock to make sure the performance is consistent.
But we will see as soon as the games arrive.
 
Last edited:

Shmunter

Member
What are you smoking? Lol. You only need a notebook to run the UE5 demo with the same amount of detail at 1440p@40fps. XSX will easily run it at a higher res. Stop being to delusional.
And STOP with this stupid narrative that XSX means last-gen jaguar based games and PS5 is the only platform on this planet to have nextgen visuals.
This is ridiculous.



All of this doesn't matter at all. Developers will not create based on a single SSD. They will use the PC SSD as the lowest common denominator and that's it. So, we will see better loading times, but that's it.
Again, UE5 demo could run on a notebook - according to epic engineers in china, at 40fps.



what? no, XSX has way more TFLOps, and it's sustained 100% of the time. PS5 clocks are variable, I think Devs will just use the base/lowest possible clock to make sure the performance is consistent.
But we will see as soon as the games arrive.
Hehe, next gen flagship demo running even better on a current gen notebook, China says so. I love this one, one of my favourites.
 

FZW

Member
Why do ppl focus so much on The TF advantage that the XSX has when it has more than just TF advantage. Faster CPU, Stronger GPU (with 50% more compute Units) and wider memory bus. Literally every category except SSD. Defining the power difference as just a TF difference is really selling it short.
 

THE:MILKMAN

Member
Why do ppl focus so much on The TF advantage that the XSX has when it has more than just TF advantage. Faster CPU, Stronger GPU (with 50% more compute Units) and wider memory bus. Literally every category except SSD. Defining the power difference as just a TF difference is really selling it short.

Mark Cerny explained this in Road to PS5. Raising clocks affects multiple parts of the GPU where as CUs alone just increase ALU. Even Andrew Goossen agrees with this BTW.
 

chilichote

Member
More like people trying really hard to narrow the gap. That 2TF's is bare minimum and is actually greater when you account for the PS5's "variable clocks". You just can't ignore the massive compute unit advantage in which Ray-Tracing is directly tied to. Higher clocks will not mitigate that.


The mistake that many (gladly!) make is that, although the XSeX theoretically has 12 TF thanks to the fixed clock, the load is practically important. But it is the case that game code rarely exploits the full potential of hardware, even on consoles. This is one of the reasons why Cerny took this step with the PS5.

Or in short: it doesn't matter at first whether you have fixed clock rates or variable ones. It's all about using the potential.
 

longdi

Banned

THE:MILKMAN

Member

Ascend

Member
If it ultimately leads to notably higher utilisation of the CPU/GPU or actually the whole system then I'm fine with it. Wait and see.
It doesn't. It's purely about power management. What is explained on that Resetera link is not wrong. It is the way they were able to reach the 2.3 GHz for the GPU with the limited power budget. They are able to keep the load higher to the limit, eliminating the peaks that are wasteful. If they did not do things this way, they would be stuck at or below 2GHz, which Cerny himself implied when he said they had trouble maintaining 2 GHz the traditional way. So everything is already factored in with the higher clock speed for the GPU. There will be no performance gains on top of that.
 

geordiemp

Member
LOL. The demo is running at 30 fps. That means it's 33.3 ms, not 16 ms...

The goal is 60 FPS (epic) this is an early build that ran at 40 fps, likely a game would not need movie assets and would be next step down textures and 60FPS is more practical than that demo, still would blow away current gen visuals and assets.

My statement is that the SSD IO generally needs to load high quality assets in the fraction of the 16 ms allocated to it is correct to be able to asset stream per frame (generally). Most players want 60 FPS,.

I also assume next gen we will continue to get performance and image quality mode options, so for me the IO latency and IO system is more important than outright speed, as I prefer 60 FPS.

It doesn't. It's purely about power management. What is explained on that Resetera link is not wrong. It is the way they were able to reach the 2.3 GHz for the GPU with the limited power budget. They are able to keep the load higher to the limit, eliminating the peaks that are wasteful. If they did not do things this way, they would be stuck at or below 2GHz, which Cerny himself implied when he said they had trouble maintaining 2 GHz the traditional way. So everything is already factored in with the higher clock speed for the GPU. There will be no performance gains on top of that.

He is refering to clock speed of caches and other things that would help other than 2.23 GHz contribution towards the TF number I believe was the sentiment. It would be interesting to see what circumances the GHz drops and more interestingly how quickly volatages and frequency change and for how long (in milliseconds).
 
Last edited:

FZW

Member
Mark Cerny explained this in Road to PS5. Raising clocks affects multiple parts of the GPU where as CUs alone just increase ALU. Even Andrew Goossen agrees with this BTW.
That's cool and all but have you ever seen a faster and narrower GPU beat out a wider and slower GPU within the same architecture? I know I haven't.

the 2080 super is faster than a 2080ti with less compute units yet the TI has much better performance. if you overclocked a 5700 to a higher clock speed than a 5700xt it will only come close, and that only has a 4 compute unit advantage. Imagine if it had a 16 CU advantage.
 

geordiemp

Member
That's cool and all but have you ever seen a faster and narrower GPU beat out a wider and slower GPU within the same architecture? I know I haven't.

the 2080 super is faster than a 2080ti with less compute units yet the TI has much better performance. if you overclocked a 5700 to a higher clock speed than a 5700xt it will only come close, and that only has a 4 compute unit advantage. Imagine if it had a 16 CU advantage.

We have never had RDNA2 GPU before, lets see what the PC parts clock are before quoting older parts which are not relevant. Cerny did say, if you listen carefully, that they capped at 2.23 GHz as logic and other things stopped the performance gains.

He has already given a clue to RDNA2 scaling.#

Both machines have good and bad points, if you could merge the 2 you would have 52 CU, 2.2 Ghz, 15 TF, smart shift and predictive power control, 5.5 raw GBs SSD with crazy IO latency, and chuck in 24 GB for good measure - perfect.
 
Last edited:

THE:MILKMAN

Member
It doesn't. It's purely about power management. What is explained on that Resetera link is not wrong. It is the way they were able to reach the 2.3 GHz for the GPU with the limited power budget. They are able to keep the load higher to the limit, eliminating the peaks that are wasteful. If they did not do things this way, they would be stuck at or below 2GHz, which Cerny himself implied when he said they had trouble maintaining 2 GHz the traditional way. So everything is already factored in with the higher clock speed for the GPU. There will be no performance gains on top of that.

Watch and read the included gif. Whether that is correct we'll have to wait and see. The benefit of a few percent and variable down clock seems worth it though (if correct)

Liabe Brave said:
But now we have lots of new headroom between our highest spikes and the power supply buffer zone. How can we easily use that? Simply by raising the clockspeed until the highest peaks are back at the limit. Since total power draw is a function of number of transistors flipped, times how fast they're flipping, the power drawn rises across the board. But now, the non-peak parts of your code have more oomph. There's literally more computing power to throw at the useful work. You can increase visible quality for the user in all the non-blitz scenes, which is the vast majority of the game.

That's cool and all but have you ever seen a faster and narrower GPU beat out a wider and slower GPU within the same architecture? I know I haven't.

the 2080 super is faster than a 2080ti with less compute units yet the TI has much better performance. if you overclocked a 5700 to a higher clock speed than a 5700xt it will only come close, and that only has a 4 compute unit advantage. Imagine if it had a 16 CU advantage.

Nobody is claiming PS5 will beat XSX. The GPU is 18% stronger overall and will show that in some ways no doubt. It is claims of 50% or 44% more CUs that are misleading in the comparisons.
 

Degree

Banned
It’s also tied to clock speed. It’s just people trying really hard at widening the GPU gap because 2 Tflops were a really big deal until people realized it was just a 15%-18% difference.

It gets talked plenty, but no matter how hard you keep looking for it, it isn’t there.

There’s no massive difference and if there were, you would be hearing it from devs already. We aren’t.

If you repeat that every day, once during the morning and twice before bedtime, it might become true. Well, probably not. Never.

So, let's make it clear:

PS5 has 10.28 TFLOPs MAX
XSX has 12.1 TFLOPs ALWAYS, 100%, sustained.

So, there is a minimum 18% difference, but in reality it's a bigger difference, that's due to the architecture of the PS5 and the way PS5 is able to reach that 10.28 TFLOPs.

See, looking at the TFLOPS value of the PS5 you have to know it's variable and, more importantly, if the GPU is running at the max value, the CPU has to be throttled:

Several developers speaking to Digital Foundry have stated that their current PS5 work sees them throttling back the CPU in order to ensure a sustained 2.23GHz clock on the graphics core.

Now, the XSX CPU already is faster, now the difference WILL be even higher, if Devs have to throttle the CPU, so not only is the GPU much more powerful, in this case, even the CPU will be much more powerful.

Throttling the CPU is at the beginning of the Gen not a huge problem, because games are still designed with jaguar in mind, but soon enough, when nextgen only games arrive, devs can't just throttle the CPU to have a sustained GPU core.
What happens if the new games need FULL CPU power? What about the GPU clock? Well, then, devs need to throttle back the GPU to ensure a sustained CPU clock.
In this case the TFLOPS for GPUs would be even higher.

The problem is still, we do not now how much down the CPU and GPU can get, we do not know the average CPU/GPU clock.
Cerny was very unclear about that, only mentioned it a little, bit didn't give any details.

What you also have to keep in mind, the minimum TFLops difference is a whole PS4 of power, and in addition we are talking about RDNA 2.0 TFLOps here.
So, yes currentgen had a higher percentage wise difference of TFLOPs, but a) this couldn't be really used due to weak CPU and b) those are GCN Tflops, that's why it wasn't such a huge deal this gen.
But this completely changes nextgen, because:

1. We do not have a CPU Bottleneck anymore, so we can make much more use of the TFLOPs.
2. RDNA 2.0 is much more efficient than GCN (current gen), so basically, 1 Tflop of RDNA 2.0 is as powerful as 2 TFLops of GCN. So, you can still do a lot more work with 2TF's RDNA2 than you can with 500GF's of GCN.

But anyway, we will see as soon as the games arrive, how different they will be.
 

Gamingbolt just put up an article about all these posts from Matt.

Speaking about the SSDs in both upcoming consoles, Matt remarked that the Xbox Series X’s SSD is an impressive one, and definitely “a huge upgrade” over anything we’ve seen before. That said, according to him, the PS5’s IO is “on a whole other level.”

“The [Xbox Series X] SSD and whole IO setup is great, and a huge upgrade from anything we have ever had before,” he wrote. “Microsoft did a really nice job here, absolutely nothing to complain about.

“But in comparison, the PS5’s IO feels like it was taken from an IO-focused mid-gen upgrade 4 years from now. It is, very simply, on a whole other level. And yes, that allows for things you can’t do on any other platform.”

“There are certainly things developers can do with the PS5 that can’t be done the same, as fast, or as well on any other consumer hardware at the moment. That’s also a fact. The PS5’s IO is a big deal that will lead to actual results in games,” he further stated.

However, he acknowledged that the Xbox Series X is still a stronger system in many ways.

--------------------------

Nobody is claiming PS5 will beat XSX. The GPU is 18% stronger overall and will show that in some ways no doubt. It is claims of 50% or 44% more CUs that are misleading in the comparisons.

I always thought the basis for the percentage more CUs point is that the more CUs will allow for better RT implementation across the board? Surely the extra CUs have to count for something, right?
 
Last edited:

Bryank75

Banned
If you repeat that every day, once during the morning and twice before bedtime, it might become true. Well, probably not. Never.

So, let's make it clear:

PS5 has 10.28 TFLOPs MAX
XSX has 12.1 TFLOPs ALWAYS, 100%, sustained.

So, there is a minimum 18% difference, but in reality it's a bigger difference, that's due to the architecture of the PS5 and the way PS5 is able to reach that 10.28 TFLOPs.

See, looking at the TFLOPS value of the PS5 you have to know it's variable and, more importantly, if the GPU is running at the max value, the CPU has to be throttled:



Now, the XSX CPU already is faster, now the difference WILL be even higher, if Devs have to throttle the CPU, so not only is the GPU much more powerful, in this case, even the CPU will be much more powerful.

Throttling the CPU is at the beginning of the Gen not a huge problem, because games are still designed with jaguar in mind, but soon enough, when nextgen only games arrive, devs can't just throttle the CPU to have a sustained GPU core.
What happens if the new games need FULL CPU power? What about the GPU clock? Well, then, devs need to throttle back the GPU to ensure a sustained CPU clock.
In this case the TFLOPS for GPUs would be even higher.

The problem is still, we do not now how much down the CPU and GPU can get, we do not know the average CPU/GPU clock.
Cerny was very unclear about that, only mentioned it a little, bit didn't give any details.

What you also have to keep in mind, the minimum TFLops difference is a whole PS4 of power, and in addition we are talking about RDNA 2.0 TFLOps here.
So, yes currentgen had a higher percentage wise difference of TFLOPs, but a) this couldn't be really used due to weak CPU and b) those are GCN Tflops, that's why it wasn't such a huge deal this gen.
But this completely changes nextgen, because:

1. We do not have a CPU Bottleneck anymore, so we can make much more use of the TFLOPs.
2. RDNA 2.0 is much more efficient than GCN (current gen), so basically, 1 Tflop of RDNA 2.0 is as powerful as 2 TFLops of GCN. So, you can still do a lot more work with 2TF's RDNA2 than you can with 500GF's of GCN.

But anyway, we will see as soon as the games arrive, how different they will be.
No GPU runs a peak theoretical performance 100% of the time, so that is absolute nonsense.
 

Ascend

Member
Watch and read the included gif. Whether that is correct we'll have to wait and see. The benefit of a few percent and variable down clock seems worth it though (if correct)
It's definitely worth it. Things would have been very different if they didn't go this route. There's a reason SmartShift is laptop tech. It's about maximizing performance while being on a limited power budget. Consoles can stretch that power budget a lot further than laptops, so, the benefit can be tremendous here.
And I'm 99.9% sure that I'm correct on this one. You're not somehow getting more efficient calculations on top of the higher clocks. This method is what gave the higher clocks and are the increased benefits. That the additional clock speeds might bring other benefits to other components is still true.

That being said, there's always a risk of lower yields when going for high clocks, because less chips on the wafer might be able to reach your required specification... And then we have this...;

I hope it's not true, but many people say this guy is reliable. I don't know him though.
 

CurtBizzy

Member
Does the PS5 custom CU size applies to the Series X ? I know the Series X and the Xbox One X die size are identical
bC04UjT.png
 
Last edited:

Bryank75

Banned

Gamingbolt just put up an article about all these posts from Matt.





--------------------------



I always thought the basis for the percentage more CUs point is that the more CUs will allow for better RT implementation across the board? Surely the extra CUs have to count for something, right?
Mostly resolution, I don't think true ray tracing will take effect for a few more years from what I heard from some of the people here that are more technical.

So while SX may reach 4K more consistently, it will not be able to s9tream the detailed assets that PS5 will be capable of.... IF the developer uses the full-fat assets.
 

FZW

Member
Nobody is claiming PS5 will beat XSX. The GPU is 18% stronger overall and will show that in some ways no doubt. It is claims of 50% or 44% more CUs that are misleading in the comparisons.
what's misleading about it? The console has 16 more compute units than the PS5 and that is over 40% more than 36. Simple math.
 

THE:MILKMAN

Member
I always thought the basis for the percentage more CUs point is that the more CUs will allow for better RT implementation across the board? Surely the extra CUs have to count for something, right?

On paper the RT will also be 18% higher right? But all we can go on are the examples we've seen this gen: PS4 vs Xbox One and One X vs PS4 Pro and the 40-50% differences in both cases in my opinion amounted to far less than the percentages suggested.

what's misleading about it? The console has 16 more compute units than the PS5 and that is over 40% more than 36. Simple math.

The context of clock speed differences?
 
Last edited:
Top Bottom