• 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.

PS5 Pro Specs Leak are Real, Releasing Holiday 2024(Insider Gaming)

Bojji

Member
In a world of AI acceleration they really didn't "went crazy and invented some bullshit system" as you say. When you are doing massive matrix multiplication, like you do in AI training or AI inference, being able to do twice the calculations per clock does yield that benefit consistently.

Against the backdrop of BVH8 (Pro) structures vs BVH4(PS5) structures that should also yield twice as much granular intersection testing per clock per level, leading to a lot more than twice the efficiency in RT if I'm not mistaken primarily because of memory accesses.

Pro has dedicated Ai/ML hardware, why would they use shaders for that? So far only use for ML in gaming so for reconstruction and frame generation, deidacted ML core on Pro will be plenty for that. In the future more uses cases will probably appear but I doubt it will happen before PS6.

Since 2020 (and 2018 on PC) like 90% of console games don't have any Ray Tracing, for PC this number is probably like 75% (both numbers are from my imagination but I doubt they are far from truth). Standard raster performance is the most important thing and will be the most important thing in gaming hardware for next 3 years (who knows after that). Even Sony first party don't give a shit about RT, only Insomniac is doing something with it.
 

PaintTinJr

Member
Pro has dedicated Ai/ML hardware, why would they use shaders for that? So far only use for ML in gaming so for reconstruction and frame generation, deidacted ML core on Pro will be plenty for that. In the future more uses cases will probably appear but I doubt it will happen before PS6.

Since 2020 (and 2018 on PC) like 90% of console games don't have any Ray Tracing, for PC this number is probably like 75% (both numbers are from my imagination but I doubt they are far from truth). Standard raster performance is the most important thing and will be the most important thing in gaming hardware for next 3 years (who knows after that). Even Sony first party don't give a shit about RT, only Insomniac is doing something with it.
It has dedicated ability to use the Flops, possibly specialist hardware async paths, in the way RT has dedicated ability on PS5, and can do shaders with Texture lookup, and RT intersection tests simultaneously. But from what I've read in the specs shown on here, the same 67TF/s for AI at FP16 is the same 33.5TF/s dual issue number just x2 as a RPM multiplier, and is sharing of the CU capability.
 
Last edited:

Sw0pDiller

Banned
With all the leaks about a true handheld xbox coming our way i must say i understand the descision of MS not to make a pro console this gen. We allreay have a X + a budget S and a mobile series S in the future. this mobile series console feel like a better addition to the series line-up tha a pro console would.

Still, day one PS5 pro for me.
 
PS5 doesn't support Mesh Shaders as MS from DX12U spec, but developers can use Primitive shaders to achieve similar results.

vmsEjCb.jpeg



There is nothing there which shows that PS5 does NOT support Mesh Shaders, the 4Gamer article you shared even talks about how the underlying hardware is the same, and that ALL AMD cards including the Series X/S compile Mesh Shaders down to Primitive Shaders in code.

Back to my original point, the PS5 already has the hardware for Mesh Shaders, so the Pro is not gaining anything in that regard by switching to RDNA 3+ and Mesh Shaders aren't exclusive to DirextX either so there's nothing stopping developers from implementing on the PS5 other than potential lack of software/API support but again the comments from the Remedey developers seem to contradict this.
 

McGILLAZ

Member
Even if this is the most powerful gaming system known to man, I'm becoming increasingly aggravated at Sony and there forced censorship. Stellar Blade seems like a strange case, not only with the costume censorship, but also with the toned down violence. Knowing something like Devil May Cry V is offered uncensored on Microsoft's platform while it is edited on Sony's console for example, I'm debating whether I'd ever upgrade or purchase another console from them.
 

onQ123

Member

Gaiff

SBI’s Resident Gaslighter
Even if this is the most powerful gaming system known to man, I'm becoming increasingly aggravated at Sony and there forced censorship. Stellar Blade seems like a strange case, not only with the costume censorship, but also with the toned down violence. Knowing something like Devil May Cry V is offered uncensored on Microsoft's platform while it is edited on Sony's console for example, I'm debating whether I'd ever upgrade or purchase another console from them.
Y’all can just get a PC if you care that much about tits and blood.
 

Loxus

Member
I think people should be more focused on the 45% uptick, rather than getting led astray over misleading 33.5TF figures.
I disagree. That 45% could be only referring to unpatched games. Sony isn't going to implement something they’re not going to use.

Microbenchmarking AMD’s RDNA 3 Graphics Architecture
On the other hand, VOPD does leave potential for improvement. AMD can optimize games by replacing known shaders with hand-optimized assembly instead of relying on compiler code generation. Humans will be much better at seeing dual issue opportunities than a compiler can ever hope to. Wave64 mode is another opportunity. On RDNA 2, AMD seems to compile a lot of pixel shaders down to wave64 mode, where dual issue can happen without any scheduling or register allocation smarts from the compiler.

It’ll be interesting to see how RDNA 3 performs once AMD has more time to optimize for the architecture, but they’re definitely justified in not advertising VOPD dual issue capability as extra shaders. Typically, GPU manufacturers use shader count to describe how many FP32 operations their GPUs can complete per cycle. In theory, VOPD would double FP32 throughput per WGP with very little hardware overhead besides the extra execution units. But it does so by pushing heavy scheduling responsibility to the compiler. AMD is probably aware that compiler technology is not up to the task, and will not get there anytime soon.



The main problem is developers build games first on Nvidia, then port them to AMD. To make things worse, dev won't waste time coding for AMD specific hardware, mainly because AMD are not as popular / widespread as Nvidia hardware.

MSI is already doing away with AMD GPUs.


Consoles on the other hand will utilize dual-issue. It's part of the console feature set and API will make it easier for devs to work with dual-issue.

A good example of a game that best utilize AMD hardware thanks to the consoles is Spider-Man.
sw3lnYK.jpeg


This is with AMD supposedly crap RT.
Now imagine when PS5Pro games utilize dual-issue and 3rd gen RT and not to mention AI/ML upscaling on top of that.

Here we can see a glimpse of dual-issue capabilities if AMD hardware was properly utilized.
Microbenchmarking AMD’s RDNA 3 Graphics Architecture
In this test, we’re running a single workgroup to keep the test local to a WGP. Because boost behavior is quite variable on recent GPUs, we’re locking clocks to 1 GHz to drill down on per-clock behavior.

rdna2_wgp_insrate-1.png
My test is definitely overestimating on Ada for FP32 and INT32 adds, or the assumption of 2.7 GHz clock speed was off.
Unfortunately, testing through OpenCL is difficult because we’re relying on the compiler to find dual issue opportunities. We only see convincing dual issue behavior with FP32 adds, where the compiler emitted v_dual_add_f32 instructions. The mixed INT32 and FP32 addition test saw some benefit because the FP32 adds were dual issued, but could not generate VOPD instructions for INT32 due to a lack of VOPD instructions for INT32 operations. Fused multiply add, which is used to calculate a GPU’s headline TFLOPs number, saw very few dual issue instructions emitted. Both architectures can execute 16-bit operations at double rate, though that’s unrelated to RDNA 3’s new dual issue capability. Rather, 16-bit instructions benefit from a single operation issued in packed-math mode. In other major categories, throughput remains largely similar to RDNA 2.
 

King Dazzar

Member
I disagree. That 45% could be only referring to unpatched games. Sony isn't going to implement something they’re not going to use.

Microbenchmarking AMD’s RDNA 3 Graphics Architecture
On the other hand, VOPD does leave potential for improvement. AMD can optimize games by replacing known shaders with hand-optimized assembly instead of relying on compiler code generation. Humans will be much better at seeing dual issue opportunities than a compiler can ever hope to. Wave64 mode is another opportunity. On RDNA 2, AMD seems to compile a lot of pixel shaders down to wave64 mode, where dual issue can happen without any scheduling or register allocation smarts from the compiler.

It’ll be interesting to see how RDNA 3 performs once AMD has more time to optimize for the architecture, but they’re definitely justified in not advertising VOPD dual issue capability as extra shaders. Typically, GPU manufacturers use shader count to describe how many FP32 operations their GPUs can complete per cycle. In theory, VOPD would double FP32 throughput per WGP with very little hardware overhead besides the extra execution units. But it does so by pushing heavy scheduling responsibility to the compiler. AMD is probably aware that compiler technology is not up to the task, and will not get there anytime soon.



The main problem is developers build games first on Nvidia, then port them to AMD. To make things worse, dev won't waste time coding for AMD specific hardware, mainly because AMD are not as popular / widespread as Nvidia hardware.

MSI is already doing away with AMD GPUs.


Consoles on the other hand will utilize dual-issue. It's part of the console feature set and API will make it easier for devs to work with dual-issue.

A good example of a game that best utilize AMD hardware thanks to the consoles is Spider-Man.
sw3lnYK.jpeg


This is with AMD supposedly crap RT.
Now imagine when PS5Pro games utilize dual-issue and 3rd gen RT and not to mention AI/ML upscaling on top of that.

Here we can see a glimpse of dual-issue capabilities if AMD hardware was properly utilized.
Microbenchmarking AMD’s RDNA 3 Graphics Architecture
In this test, we’re running a single workgroup to keep the test local to a WGP. Because boost behavior is quite variable on recent GPUs, we’re locking clocks to 1 GHz to drill down on per-clock behavior.

rdna2_wgp_insrate-1.png
My test is definitely overestimating on Ada for FP32 and INT32 adds, or the assumption of 2.7 GHz clock speed was off.
Unfortunately, testing through OpenCL is difficult because we’re relying on the compiler to find dual issue opportunities. We only see convincing dual issue behavior with FP32 adds, where the compiler emitted v_dual_add_f32 instructions. The mixed INT32 and FP32 addition test saw some benefit because the FP32 adds were dual issued, but could not generate VOPD instructions for INT32 due to a lack of VOPD instructions for INT32 operations. Fused multiply add, which is used to calculate a GPU’s headline TFLOPs number, saw very few dual issue instructions emitted. Both architectures can execute 16-bit operations at double rate, though that’s unrelated to RDNA 3’s new dual issue capability. Rather, 16-bit instructions benefit from a single operation issued in packed-math mode. In other major categories, throughput remains largely similar to RDNA 2.
I don't think focusing on figures which suggest a 300% uptick are realistic. I do however think that the 45% uptick could be more when we take into account the new upscaling. But its fine for us to disagree, whilst I still enjoyed reading some of your take. 😊
 

Tqaulity

Member
Personally I think people should stop thinking about the 33.5TF number

Looking Season 6 GIF by This Is Us
THANK YOU HeisenbergFX4 HeisenbergFX4 ! I've been just shaking my head for months now that people are treating this 33.5 TFLOP number like it's the bible and known fact when there are so many obvious things about it that make it highly suspect to say the least:
  • The leaked documents didn't actually mention 33.5TFLOPs or anything about the GPU config (i.e shader count, clock speed, ROPs etc).
  • The only thing mentioned in the docs about the GPU is that it's larger with faster RAM and "rendering is 45% faster" which in itself is a very ambiguous figure with little context to draw conclusions.
  • The 33.5TFLOPs is a calculated figure from a single sub-bullet point in the "Machine Learning capabilities of the GPU" mentioning a 67TFLOP FP16 figure. A.) Since when does anyone use FP16 TFLOPs to communicate shader performance in video games (that would be a first)? B.) Even if that figure were true for machine learning, how folks think that would directly apply to FP32 shader performance is...interesting :pie_thinking:
  • You can't have a TFLOP figure without a known GPU clock speed right? No where in the docs did they mention clock speed and the consensus among the "insiders" is that the GPU clocks are not known yet. History tells us that the clocks are typically the last thing to get finalized which is probably why there is no credible leaks to this point with an actual GPU clock. Yet somehow, the TFLOP number is a "known quantity" to some :LOL:
  • To that point above, many have already seen that if you try to reverse engineer a GPU clock speed from that 33.5TFLOP figure, you get a clock lower than the base PS5. NEVER in the history of video game hardware have we seen any "revision" of a system releasing AFTER the original system (of the same generation) with a lower clock speed than the base. That makes absolutely no sense, has never happened, and would have negative implications to how even existing games would run from a backwards compatibility perspective. Yet, 33.5 TFLOPs it is :messenger_smirking:
People just took the bait and are eating it up. I've said it before...I can't wait for the meltdown when the actual confirmed specs come out from Sony :pie_roffles:



That would be dumb unless we are stopping to talk about the 67 TFLOPS number.

If you're talking in the context of actual graphics performance, then you really should

Don't think about it because actual results will end up looking far below typical expectations for a 33tf gpu? Or because end results are so awesome with PSSR that we should no longer be thinking in terms of TF, a la Mark Cerny messaging.


Your statement really could go either way. You sneak devil you😆
It's not really about the dual issue inflation of the TFLOPs figure, which is known at this point. Everything I said above (and more that I won't) should be enough to at least give cause to pause on running with that 33TF figure.

This whole thing is setup for some really fun days ahead :messenger_grinning_smiling:



giphy.gif
 

Akuji

Member
Dont think Remake will get a another big boost. It runs pretty great on PS5 already if iam not mistaken?
So you can play remake, rebirth on the other hand, iam holding out as well on this one :D
 
Dont think Remake will get a another big boost. It runs pretty great on PS5 already if iam not mistaken?
So you can play remake, rebirth on the other hand, iam holding out as well on this one :D
What are you talking about?? People complain about Rebirth not the Remake(or do you talk about the second game can't tell)and second the image quality is shit compared to the first game.

The framerate is okay but the image quality is shit.
 
Last edited:

S0ULZB0URNE

Member
THANK YOU HeisenbergFX4 HeisenbergFX4 ! I've been just shaking my head for months now that people are treating this 33.5 TFLOP number like it's the bible and known fact when there are so many obvious things about it that make it highly suspect to say the least:
  • The leaked documents didn't actually mention 33.5TFLOPs or anything about the GPU config (i.e shader count, clock speed, ROPs etc).
  • The only thing mentioned in the docs about the GPU is that it's larger with faster RAM and "rendering is 45% faster" which in itself is a very ambiguous figure with little context to draw conclusions.
  • The 33.5TFLOPs is a calculated figure from a single sub-bullet point in the "Machine Learning capabilities of the GPU" mentioning a 67TFLOP FP16 figure. A.) Since when does anyone use FP16 TFLOPs to communicate shader performance in video games (that would be a first)? B.) Even if that figure were true for machine learning, how folks think that would directly apply to FP32 shader performance is...interesting :pie_thinking:
  • You can't have a TFLOP figure without a known GPU clock speed right? No where in the docs did they mention clock speed and the consensus among the "insiders" is that the GPU clocks are not known yet. History tells us that the clocks are typically the last thing to get finalized which is probably why there is no credible leaks to this point with an actual GPU clock. Yet somehow, the TFLOP number is a "known quantity" to some :LOL:
  • To that point above, many have already seen that if you try to reverse engineer a GPU clock speed from that 33.5TFLOP figure, you get a clock lower than the base PS5. NEVER in the history of video game hardware have we seen any "revision" of a system releasing AFTER the original system (of the same generation) with a lower clock speed than the base. That makes absolutely no sense, has never happened, and would have negative implications to how even existing games would run from a backwards compatibility perspective. Yet, 33.5 TFLOPs it is :messenger_smirking:
People just took the bait and are eating it up. I've said it before...I can't wait for the meltdown when the actual confirmed specs come out from Sony :pie_roffles:


If you're talking in the context of actual graphics performance, then you really should


It's not really about the dual issue inflation of the TFLOPs figure, which is known at this point. Everything I said above (and more that I won't) should be enough to at least give cause to pause on running with that 33TF figure.

This whole thing is setup for some really fun days ahead :messenger_grinning_smiling:



giphy.gif
Everything is rumor thus far but we only want to listen to the 45% part of the rumor 🤣




Oh it's going to be glorious watching the spinning denial when all the specs are officially announced.

Bookmarking this.
 

CloudShiner

Member
What are you talking about?? People complain about Rebirth not the Remake(or do you talk about the second game can't tell)and second the image quality is shit compared to the first game.

The framerate is okay but the image quality is shit.
Yeah mainly Rebirth of course, but I'm sure PSSR can improve Remake to some extent also.
 

King Dazzar

Member
So you are expecting a machine with 3x the power of the current PS5?

I know many are I just want to be crystal here what we are talking about

Btw the 45% rumor actually came from Sony provided slides
Exactly. And I trust more in Sony's own real world metric, than some hypothetical misleading 300%+ TF increase.

That said, I'd love for Sony to have totally under miscalculated their own hardware....
 

S0ULZB0URNE

Member
So you are expecting a machine with 3x the power of the current PS5?

I know many are I just want to be crystal here what we are talking about

Btw the 45% rumor actually came from Sony provided slides
So did the 33.5TF's

If the rumored 33.5TF's rumor is true...I absolutely am expecting 3x the GPU power.
The fruits of this will be seen with higher resolution(not as cpu dependent) PS5 games.
I am aware of the CPU,ram needing, but not getting attention and do think these will hold back seeing games being 3x better.

However when we consider the AI upscaling tech and the GPU muscle given that we will see great results by developers who take advantage of the architecture.

Edit: and I think the 45% is a moddest number Sony gave.
Yes I feel it will be higher of a %.
 
Last edited:
So did the 33.5TF's

If the rumored 33.5TF's rumor is true...I absolutely am expecting 3x the GPU power.
The fruits of this will be seen with higher resolution(not as cpu dependent) PS5 games.
I am aware of the CPU,ram needing, but not getting attention and do think these will hold back seeing games being 3x better.

However when we consider the AI upscaling tech and the GPU muscle given that we will see great results by developers who take advantage of the architecture.
Thanks for clarifying I just wanted to know exactly what you were expecting since I haven't followed this thread super closely as of late
 

King Dazzar

Member
So did the 33.5TF's

If the rumored 33.5TF's rumor is true...I absolutely am expecting 3x the GPU power.
The fruits of this will be seen with higher resolution(not as cpu dependent) PS5 games.
I am aware of the CPU,ram needing, but not getting attention and do think these will hold back seeing games being 3x better.

However when we consider the AI upscaling tech and the GPU muscle given that we will see great results by developers who take advantage of the architecture.

Edit: and I think the 45% is a moddest number Sony gave.
Yes I feel it will be higher of a %.
Well let's hope you're right. Either way I'm day one for sure....
 

Gaiff

SBI’s Resident Gaslighter
33.5 TFLOPs is the theoretical max compute throughput with dual-issue, not GPU horsepower. How many times do we have to repeat this shit for people to pound it into their heads?
 
33.5 TFLOPs is the theoretical max compute throughput with dual-issue, not GPU horsepower. How many times do we have to repeat this shit for people to pound it into their heads?

And the funny thing is that the dual-compute scale was chosen by AMD, not Sony!

Why are people not complaining that the RX 7800 XT is advertised as a 37.3 TFLOPS GPU by AMD???
 

S0ULZB0URNE

Member
And the funny thing is that the dual-compute scale was chosen by AMD, not Sony!

Why are people not complaining that the RX 7800 XT is advertised as a 37.3 TFLOPS GPU by AMD???
Funny is when they try and compare off the shelf pc parts to custom made console components.
I'll never acknowledge that and laugh every time.
 

Thedtrain

Member
One of the most unrecognized enhancements that came along with the PS4Pro was how much snappier the overall UX was compared to the base PS4

With the ps5, I find it to be really very nice to use. Wondering what other nice QoL upgrades will be included with the pro besides the standard increase in perf.
 
Basically I believe GPU vs GPU is a 3x increase but other factors will keep this 3x from being the case in real world performance.
I went back and looked at the docs and the 45% increase is for rendering performance under increased GPU performance

I hope you are right though, regardless I am day one and if people still love me I will get it a few days early
 

Bojji

Member
Funny is when they try and compare off the shelf pc parts to custom made console components.
I'll never acknowledge that and laugh every time.

Pro is actually more customized than standard PS5, it has (not seen before) RT and ML parts.

But PS5 was:

- 36 CU RDNA1 + RT from RDNA2
- 36 CU RDNA2 - VRS, SFS, MS and L3 cache

You can like whatever answer you want but fact is we know exactly what PS5 GPU represents - add 45% to that and you have raw raster power of PS5 Pro.
 

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
33.5 TFLOPs is the theoretical max compute throughput with dual-issue, not GPU horsepower. How many times do we have to repeat this shit for people to pound it into their heads?

Yeah..........I'm confused why people are lying to themselves on this. Guys........the PS5 Pro will perform like a console that's 15 TFs with added\better Ray-tracing and Mach-Learning with PSSR. Overall, it'll probably feel like a console that's 70% better than the Xbox Series X games with raytracing turned on (like Cyberpunk) or 30-40% better without raytracing. But don't expect a next-gen level jump here.
 
Last edited:
Top Bottom