Forums - Sony Discussion - Unreal Engine 5 Announced + PS5 Demo

chakkra said:

What I'm finding most fascinating about all of this is that, before the new consoles specs were revealed, just about everybody was putting the blame on the old Jaguar CPUs for the 8th gen's bottleneck, and for the life of me I don't remember anyone even mentioning storage as the culprit. Now all of a sudden the SSD is the only thing that matters.

And yes, I understand that higher SSD speeds will allow to stream more data faster, but that data needs to be processed in the first place. You can have all the storage speed and memory bandwith in the world, but if the CPU and GPU cannot keep up, you're just going to end up with a waste of bandwith.

Everything is relative though, and I guess ultimately the approach both Sony and MS have gone with is fully dependent on the cost of components and the diminishing returns of linearly progressive hardware spec bumps. 5 years ago I don't think many of us thought that things would have slowed down as much as they have comparative to price. Price being the major factor here that has skyrocketed.

That is to say that usually we see huge amounts of RAM upgrades every 5 years. So much so that on average we generally see a RAM increase by a factor of approximately x10. PS1 -> PS2 -> PS3 -> PS4. Going from PS4 to PS5 is only going to be twice as much RAM. Why? Cost of RAM is ridiculous. This presented Sony/MS with a dilemna, how do they achieve a generational leap while still keeping the cost down? Here is where blazing fast asset streaming comes in. Cerny touched on this heavily in his presentation - whereby your SSD (800+gb) can almost act as RAM (not quite but almost).

Ultimately this leads to maximum utilization of your 16GB of RAM. You will see that a general theme of the next generation consoles is maximum efficiency in using what they have.

With regards to CPU power - there wasnt really much thought needed on this one. 8 zen2 cores was always going to be the standard given what we see on desktop along with price. Again bumping this to 16 just isn't feasible from a cost perspective and anything less would be hopelessly underpowered with no way to make up the difference. Perhaps if they did decide to spend the resources on more CPU power it might have come at the cost of their SSD and IO solution along with a better heat solution and bigger form factor etc. So again the logical decision was made, and this is a decision made by both manufacturers leading to little or no needed discussion on the CPU front.

Personally, with regards to the GPU i see it as a exactly the same way as I do with the CPU. Latests RDNA 2 architecture with decent clock speeds, is really all that is required. Anything less again would have been waaay to far behind, and anything more is probably too expensive, too big and too hot. This seems to be the only area of deviation though between the two manufacturers, MS goes with a bigger GPU die, while Sony pushed some hardware processing to assist with IO through put. 

And thats why we find ourselves here with all this SSD talk.

If things had progressed as they have been 10-15 years ago we would probably be talking about huge amounts of RAM like 64gb and 16 core cpus clocked at 10ghz etc..

This also leads to Cerny's comment about equating PS4 teraflops to PS5 teraflops. Devs and Engineers are finding new ways to overcome limitations and the hardware manufacturers are following suite and building to cater to those needs. If they know to focus on certain areas then processing of a particular type of load becomes more efficient - but it doesnt necessarily change the teraflop count. Its the same concept as running Ray Tracing work loads on a monster gpu from 5 years ago - its pointless.



Intel Core i7 3770K [3.5GHz]|MSI Big Bang Z77 Mpower|Corsair Vengeance DDR3-1866 2 x 4GB|MSI GeForce GTX 560 ti Twin Frozr 2|OCZ Vertex 4 128GB|Corsair HX750|Cooler Master CM 690II Advanced|

Around the Network
chakkra said:
CGI-Quality said:

I am not sure who said the SSD is the only thing that matters (if they are in here, they are very much a minority). This also leaves out the I/O (which will be hjust as important and people have pointed that out).

In any case, yes, those Jaguar CPUs were bottlenecks to better performance. That's a fact. So people had good reason to point blame at them when mentioning what needed a big upgrade (which happens to also be part of the next gen package). Worried about next gen bandwidth/bottlenecks (at least in regards to the PS5)? Listen to Mark Cerny.

Well, I don't think you need to literally see the words "SSD is the only thing that matters" to notice that that is the only thing people talk about these days (and the way they talk about it).

And yes, I watched (and read) Sony's presentation, and I watched (and read) MS's presentation.

You want to hear my opinion on the matter?
Well, Series X designers got to the conclusion that a SSD running at 2.4GB have the enough streaming speed needed for the amount of data that a GPU with 52CUs x1.8GHz and a 8x Zen 2 Cores at 3.8GHz are going to be able to process/generate.

PS5 designers, on the other hand, concluded that a SSD at 5.5GB is needed for the amount of data that a GPU with 36 CUs x2.2GHz and a 8x Zen 2 Cores at 3.5GHz are going to be able to process/generate.

One of them is going to end up being wrong. If MS is wrong then their SSD solution might end up being a bottleneck, but if Sony is wrong then the PS5 will end up with just a bunch of bandwith unused/wasted.

Those differences are sooo small in the grand scheme of things though. I highly doubt either console is going to be hugely bottlenecked in any significant way to the point where one console outshines the other to a significant degree.



Intel Core i7 3770K [3.5GHz]|MSI Big Bang Z77 Mpower|Corsair Vengeance DDR3-1866 2 x 4GB|MSI GeForce GTX 560 ti Twin Frozr 2|OCZ Vertex 4 128GB|Corsair HX750|Cooler Master CM 690II Advanced|

chakkra said:

You want to hear my opinion on the matter?
Well, Series X designers got to the conclusion that a SSD running at 2.4GB have the enough streaming speed needed for the amount of data that a GPU with 52CUs x1.8GHz and a 8x Zen 2 Cores at 3.8GHz are going to be able to process/generate.

PS5 designers, on the other hand, concluded that a SSD at 5.5GB is needed for the amount of data that a GPU with 36 CUs x2.2GHz and a 8x Zen 2 Cores at 3.5GHz are going to be able to process/generate.

That is nonsense.

The whole hardware chain of <data from ssd at speed x> -> <data in gpu memory at speed x> is incredibly complex if you look at the Cerny slides. The work on all the extra processors was started at the beginning of the PS5 development (maybe even before). That was long before cpu and gpu geometries were set so these numbers were totally unknown. What they could do is estimate the minimum bandwidth required by analyzing test game setups.

The speed of 5.5GB/s was chosen to be high enough to do the job while at the same time it does not require particularly expensive chips. A 12 lane ssd could run at much higher speeds but at much hotter temperatures (a key element to watch over).



chakkra said:
CGI-Quality said:

I am not sure who said the SSD is the only thing that matters (if they are in here, they are very much a minority). This also leaves out the I/O (which will be hjust as important and people have pointed that out).

In any case, yes, those Jaguar CPUs were bottlenecks to better performance. That's a fact. So people had good reason to point blame at them when mentioning what needed a big upgrade (which happens to also be part of the next gen package). Worried about next gen bandwidth/bottlenecks (at least in regards to the PS5)? Listen to Mark Cerny.

Well, I don't think you need to literally see the words "SSD is the only thing that matters" to notice that that is the only thing people talk about these days (and the way they talk about it).

And yes, I watched (and read) Sony's presentation, and I watched (and read) MS's presentation.

You want to hear my opinion on the matter?
Well, Series X designers got to the conclusion that a SSD running at 2.4GB have the enough streaming speed needed for the amount of data that a GPU with 52CUs x1.8GHz and a 8x Zen 2 Cores at 3.8GHz are going to be able to process/generate.

PS5 designers, on the other hand, concluded that a SSD at 5.5GB is needed for the amount of data that a GPU with 36 CUs x2.2GHz and a 8x Zen 2 Cores at 3.5GHz are going to be able to process/generate.

One of them is going to end up being wrong. If MS is wrong then their SSD solution might end up being a bottleneck, but if Sony is wrong then the PS5 will end up with just a bunch of bandwith unused/wasted.

Neither is going to end up being wrong, because the two devices don’t work exactly the same and will have their internals leveraged for game development (as consoles generally do). So, no there won't be "just a bunch of bandwith unused/wasted". In either case.

And, yeah, no shit the SSDs are largely talked about. Their potential and the fact that no consumer-grade part (such as the PS5's SSD) exists like them for purchase (yet) makes it worth discussing. These aren't teraflops. I'd understand the bitchin' if they weren't so advanced and/or simply numbers on a page.

drkohler said:
chakkra said:

You want to hear my opinion on the matter?
Well, Series X designers got to the conclusion that a SSD running at 2.4GB have the enough streaming speed needed for the amount of data that a GPU with 52CUs x1.8GHz and a 8x Zen 2 Cores at 3.8GHz are going to be able to process/generate.

PS5 designers, on the other hand, concluded that a SSD at 5.5GB is needed for the amount of data that a GPU with 36 CUs x2.2GHz and a 8x Zen 2 Cores at 3.5GHz are going to be able to process/generate.

That is nonsense.

The whole hardware chain of -> is incredibly complex if you look at the Cerny slides. The work on all the extra processors was started at the beginning of the PS5 development (maybe even before). That was long before cpu and gpu geometries were set so these numbers were totally unknown. What they could do is estimate the minimum bandwidth required by analyzing test game setups.

The speed of 5.5GB/s was chosen to be high enough to do the job while at the same time it does not require particularly expensive chips. A 12 lane ssd could run at much higher speeds but at much hotter temperatures (a key element to watch over).

Correct.

Last edited by CGI-Quality - on 05 June 2020

                                                                                                             

.



 "I think people should define the word crap" - Kirby007

Join the Prediction League http://www.vgchartz.com/predictions

Instead of seeking to convince others, we can be open to changing our own minds, and seek out information that contradicts our own steadfast point of view. Maybe it’ll turn out that those who disagree with you actually have a solid grasp of the facts. There’s a slight possibility that, after all, you’re the one who’s wrong.

Around the Network
CGI-Quality said:

"Neither is going to end up being wrong, because the two devices don’t work exactly the same and will have their internals leveraged for game development (as consoles generally do). So, no there won't be "just a bunch of bandwith unused/wasted". In either case."

I hope you're right about this part. Although I find it weird that you seem to have more issue with the PS5 part of the statement.

"I'd understand the bitchin' if they weren't so advanced and/or simply numbers on a page."

This is literally the first time I mention this topic, so I fail to see how that can be considered "bitching."

"And, yeah, no shit the SSDs are largely talked about. Their potential and the fact that no consumer-grade part (such as the PS5's SSD) exists like them for purchase (yet) makes it worth discussing. These aren't teraflops."

Now I have an issue with this statement; Teraflops is basically the GPU processing power. And I can totally understand that talking about TFs from different arquitectures is pointless, like trying to compare AMD to Nvidia cards, or cards from the RTX line to the GTX line; but when we are talking about GPUs from the same line, the GPU with more TFs will ALWAYS perform better than the one with less. In fact, I'm willing to bet my life that when AMD releases their next RDNA 2 graphics cards, the one with more teraflops will be the best performer one. So I totally fail to see how storage speed is more worthy to talk about than GPU processing power.



TallSilhouette said:

https://youtu.be/4ehDRCE1Z38

Linus commits seppuku for his misinformed comments pertaining to Tim Sweeney and the PS5 SSD. Classy move and a good summary of the drive's advantages.

I just watched it. That was amazing. It REALLY takes it on the chin at the beginning.

He didn't cry. He didn't monetize. He apologized. He asked other people that used his thoughts/words/video to apologize.  He went on to talk about the SSD and tried to explain it in a different way to "get through" to the people that didn't get Sweeny video. Then he called out the people who think that Cerny was some how paid off by Sony.

He nailed it 100%.

Last edited by Bandorr - on 05 June 2020

  • Deadliest mass shooting by an individual in US history (10/01/2017)
  • Deadliest high school shooting in US history (02/14/2018)
  • Deadliest massacre of Jews in US history (10/27/2018)
  • Political assassination attempt of TWO former presidents(and 10+ other people)  (10/23/2018 - and beyond)
chakkra said:
CGI-Quality said:

"Neither is going to end up being wrong, because the two devices don’t work exactly the same and will have their internals leveraged for game development (as consoles generally do). So, no there won't be "just a bunch of bandwith unused/wasted". In either case."

I hope you're right about this part. Although I find it weird that you seem to have more issue with the PS5 part of the statement.

"I'd understand the bitchin' if they weren't so advanced and/or simply numbers on a page."

This is literally the first time I mention this topic, so I fail to see how that can be considered "bitching."

"And, yeah, no shit the SSDs are largely talked about. Their potential and the fact that no consumer-grade part (such as the PS5's SSD) exists like them for purchase (yet) makes it worth discussing. These aren't teraflops."

Now I have an issue with this statement; Teraflops is basically the GPU processing power. And I can totally understand that talking about TFs from different arquitectures is pointless, like trying to compare AMD to Nvidia cards, or cards from the RTX line to the GTX line; but when we are talking about GPUs from the same line, the GPU with more TFs will ALWAYS perform better than the one with less. In fact, I'm willing to bet my life that when AMD releases their next RDNA 2 graphics cards, the one with more teraflops will be the best performer one. So I totally fail to see how storage speed is more worthy to talk about than GPU processing power.

Only "issue" I had was you trying to minimize the impact of the upcoming SSDs (yes, both) and "bitchin" was the general term I used. It's not the end of the world.

In any event, floating point operations/sec. (MEGA/GIGA/TERA/PETA, etc.) are a theoretical measure of power, not "basically processing power". No matter what architecture types are discussed, flops will remain theoretical and never tell the real story. Otherwise, a 2080Ti should not be able to keep up with a TITAN RTX in any capacity (given it is 3000 GIGAflops behind it — a bigger difference than those in these upcoming consoles). So, yes, they're meaningless for serious discussion, storage speed is not.

Want to know what a GPU can do? Look at its...

  1. RAM Amount/Speed/Type
  2. Shader Cores (CUDA, for example)
  3. Clock Speed
Last edited by CGI-Quality - on 05 June 2020

                                                                                                             

CGI-Quality said:
chakkra said:

"Neither is going to end up being wrong, because the two devices don’t work exactly the same and will have their internals leveraged for game development (as consoles generally do). So, no there won't be "just a bunch of bandwith unused/wasted". In either case."

I hope you're right about this part. Although I find it weird that you seem to have more issue with the PS5 part of the statement.

"I'd understand the bitchin' if they weren't so advanced and/or simply numbers on a page."

This is literally the first time I mention this topic, so I fail to see how that can be considered "bitching."

"And, yeah, no shit the SSDs are largely talked about. Their potential and the fact that no consumer-grade part (such as the PS5's SSD) exists like them for purchase (yet) makes it worth discussing. These aren't teraflops."

Now I have an issue with this statement; Teraflops is basically the GPU processing power. And I can totally understand that talking about TFs from different arquitectures is pointless, like trying to compare AMD to Nvidia cards, or cards from the RTX line to the GTX line; but when we are talking about GPUs from the same line, the GPU with more TFs will ALWAYS perform better than the one with less. In fact, I'm willing to bet my life that when AMD releases their next RDNA 2 graphics cards, the one with more teraflops will be the best performer one. So I totally fail to see how storage speed is more worthy to talk about than GPU processing power.

Only "issue" I had was you trying to minimize the impact of the upcoming SSDs (yes, both) and "bitchin" was the general term I used. It's not the end of the world.

In any event, floating point operations/sec. (MEGA/GIGA/TERA/PETA, etc.) are a theoretical measure of power, not "basically processing power". No matter what architecture types are discussed, flops will remain theoretical and never tell the real story. Otherwise, a 2080Ti should not be able to keep up with a TITAN RTX in any capacity (given it is 3000 GIGAflops behind it — a bigger difference than those in these upcoming consoles). So, yes, they're meaningless for serious discussion, storage speed is not.

Want to know what a GPU can do? Look at its...

  1. RAM Amount/Speed/Type
  2. Shader Cores (CUDA, for example)
  3. Clock Speed

The gap between 2080 TI and RTX Titan is 14.2 to 16.3 TFs, which is a %15 difference, and Titan still manages to consistently over perform it. So that actually helps to prove my statement that more TFs within the same arquitecture equals more performance.

And the only reason I'm "minimizing" the SSD solutions is because plenty of people are acting like that is the only thing that matters, and it's not the only thing nor the most important. I mean, the PS5 SSD wouldn't have been able to stream all those billions of polygons on the demo if said polygons hadn't been processed by the GPU in the first place; I mean if the PS5's GPU/CPU had been much slower, the SSD would have had to stand there waiting for the data to be processed. Dont you agree?

To put it simply, in one hand we have the CPU/GPU combo processing/generating the data, and in the other we have the RAM/SSD storing and streaming said data. The big question is at what point do the RAM/SSD become a bottleneck for the GPU/CPU or at what point is the RAM/SSD so fast that the GPU/CPU cannot keep up with them and fully utilize their speed.

Which brings me back to my point: MS concluded that their SSD solution provided enough speed for the amount of data that their GPU/CPU can process, while Sony obviously got to a different conclusion.

So yeah, I think one of them might end up being wrong in their approach. What I don't understand is why you seem so upset about the possibility of Sony being wrong. I mean, the worst that could happen is that the PS5 could end up with bandwith to spare. I see no harm there.



chakkra said:
CGI-Quality said:

Only "issue" I had was you trying to minimize the impact of the upcoming SSDs (yes, both) and "bitchin" was the general term I used. It's not the end of the world.

In any event, floating point operations/sec. (MEGA/GIGA/TERA/PETA, etc.) are a theoretical measure of power, not "basically processing power". No matter what architecture types are discussed, flops will remain theoretical and never tell the real story. Otherwise, a 2080Ti should not be able to keep up with a TITAN RTX in any capacity (given it is 3000 GIGAflops behind it — a bigger difference than those in these upcoming consoles). So, yes, they're meaningless for serious discussion, storage speed is not.

Want to know what a GPU can do? Look at its...

  1. RAM Amount/Speed/Type
  2. Shader Cores (CUDA, for example)
  3. Clock Speed

The gap between 2080 TI and RTX Titan is 14.2 to 16.3 TFs, which is a %15 difference, and Titan still manages to consistently over perform it. So that actually helps to prove my statement that more TFs within the same arquitecture equals more performance.

So yeah, I think one of them might end up being wrong in their approach. What I don't understand is why you seem so upset about the possibility of Sony being wrong. I mean, the worst that could happen is that the PS5 could end up with bandwith to spare. I see no harm there.

The flops difference between the Founders Edition 2080Ti and TITAN RTX is exactly as I stated (13.5/16.3 — you got your info from Gamesradar, who is incorrect), but even if I were to humor the higher number, it still doesn't really kill the point. And, until I see otherwise, the myth remains that anyone said the SSDs are either the only things that matter or the most important. People are arguing that they play a seriously important role in the upcoming generation (and they are absolutely correct).

Anyway, in regards to flops, as a tech guy, used to working with hardware on a more surgical basis, I'm vocal about them for a reason. On top of that, I listen closely to other informed tech guys, such as techradar, who state...

"So teraflops are a convenient and all-encompassing measurement of graphical power on a games console or GPU. But as is often the case with computing, the reality isn’t quite as simple and you probably shouldn’t use tflops as the ultimate barometer when researching your next GPU or console."

Sure, perhaps saying "they don't matter at all" is a stretch, but in the grand scheme of architecture, game development, and throughput, they are simply a baseline of theory. I know 'more flops' sounds nice, but the way these consoles will work (or any console, for that matter) isn't/won't be in theory. Rather, how the intertwined anatomy of each device flows and functions to pump out the goods. Flops sit beside "bits" in the gaming world — fun to discuss, theoretical at best. It isn't that one (or both) of the manufacturers will be "wrong in their approach". One will simply be more efficient (there's always going to be a winner, no matter how good each device is crafted, that's just science). Both are masterfully built, regardless of theoretical performance.

And, I'm not upset about anything which is why I continue trying to inform you better on this topic. :P

Last edited by CGI-Quality - on 06 June 2020