• 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)

What narrative exactly? That the Pro GPU is power constrained and cannot hit max clocks in all games? That is directly from Sony themselves. That the Pro supports Tier 2+ VRS which the normal PS5 doesn't? Again, directly from Sony themselves. That the Pro has full mesh shader support instead of primitive shaders (minor difference really, as Alex said simply programmatic). Bingo, directly from Sony again.

There's a lot of information missing here, and we can't rely on how DF are framing the information they are reading unless we see direct documentation and Sony's own words.

DF are saying that PS5 Pro supports Mesh Shaders as well, although a lot of context is missing here and it raises a lot of questions. What do they mean support? on a hardware level or a software level?

We already know PS5 has the hardware for Mesh Shaders, but there's a potential lack of API support. Maybe there's an updated API for the Pro?

We also know that Mesh Shaders aren't natively supported on AMD hardware (RDNA whitepapers don't make mention of Mesh Shaders either), they are compiled down to Primitive Shaders on all Radeon GPU's including the Series X/S. Maybe RDNA 3.5 and 4 introduce a new hardware feature which allows for native mesh shader support and the Pro will benefit from this.
 

Zathalus

Member
There's a lot of information missing here, and we can't rely on how DF are framing the information they are reading unless we see direct documentation and Sony's own words.
Well the clock speeds and VRS seems pretty self explanatory. As for Mesh Shaders, the article states the following:


"The primitive shader features found in RDNA 1 and the vanilla PS5 are augmented with full support for mesh shaders, which should hopefully see more widespread adoption of a very useful feature. Sony points out how the stricter requirements of this feature up against PC APIs are no longer an issue."
 

ChiefDada

Gold Member
Yes, the 33tf figure remains strange. It doesn't matter whether 2.35ghz can be attained 100% of the time or just 1% of the time, all GPUs communicate specs and TF in particular under best case scenario/max throughput. I can only assume that the original 33tf weren't based on final clocks.
 

S0ULZB0URNE

Member
Developer portal, so information they are relaying directly to everyone using the dev kit. I doubt Sony would state any of this in marketing material. The clock speeds, VRS, mesh shaders, and new cache configuration is all from Sony directly.
You said Sony stated this though.
 
Well the clock speeds and VRS seems pretty self explanatory. As for Mesh Shaders, the article states the following:


"The primitive shader features found in RDNA 1 and the vanilla PS5 are augmented with full support for mesh shaders, which should hopefully see more widespread adoption of a very useful feature. Sony points out how the stricter requirements of this feature up against PC APIs are no longer an issue."

I'm struggling to understand what they mean by that last part about the stricter requirements.

This doesn't answer a lot of the questions surrounding the Pro's Mesh Shader support as far as I'm aware. My theory is that the Pro alongside RDNA 3.5/4 will handle Mesh Shaders more in line with how Nvidia and Intel do it.
 

Zathalus

Member
You said Sony stated this though.
Yes, in the development portal. DF is not going to make this stuff up and risk getting backlash for it. Any developer or Sony themselves can correct them if they are making up all the claims they have made so far. Besides, the VRS/Mesh Shaders line up with what we know of the PS5 and the cache configuration is similar to RDNA 3.
 

S0ULZB0URNE

Member
Yes, in the development portal. DF is not going to make this stuff up and risk getting backlash for it. Any developer or Sony themselves can correct them if they are making up all the claims they have made so far. Besides, the VRS/Mesh Shaders line up with what we know of the PS5 and the cache configuration is similar to RDNA 3.
So you don't have anything showing Sony saying anything.
 

Zathalus

Member
I'm struggling to understand what they mean by that last part about the stricter requirements.

This doesn't answer a lot of the questions surrounding the Pro's Mesh Shader support as far as I'm aware. My theory is that the Pro alongside RDNA 3.5/4 will handle Mesh Shaders more in line with how Nvidia and Intel do it.
I'd assume it means Mesh shaders are simply more programmatic and thus easier to use. So the difference lies in flexibility of the feature and not any performance improvements.
 

Panajev2001a

GAF's Pleasant Genius
According to Jim Keller the effectiveness of modern day branch prediction for CPUs, is well above 95%. So the occurrence of pipeline stalls or even a pipeline flush, are relatively rare.

Pentium 4’s branch predictor was estimated to be well above 95% too yet general purpose branchy code hurt its performance quite a bit (due to its very very deeply pipelined design too), the point was not about branch prediction rates, in the paper they talk about memory load/store instructions reordering and conflicts that may arise (as well as decreasing instruction cache hit rate).

Again, the miss rate we were talking about before was data cache misses and how much can the OOOE front end hide (10-20 cycles sure, 60-80 cycles … it depends… going further out to RAM, which is what Fafalada was saying it is miles off of that, no it is not designed to cover that… even very aggressive designs).

Even more, if we then add the effectiveness of modern day software compilers.
OoO is not about maximizing parallelism. It's about reordering instructions as to have pipeline stalls, due to instructions with later dependencies being executed first.
OoO Execution is about maximising ILP as much as SMT is about TLP. Covering stalls, data dependencies, etc… (including and not in small part allowing to overcome the relatively tiny x86 register file limitations) is part of how the the back end will keep the execution units fed extracting as much parallelism from the instruction stream. How do you think they feed these monsters (incredibly wide backends otherwise?):

… and the modern M3 is based on even fatter cores ;). No SMT at all, so a single thread need to keep them fed.

Purely compiler based approach to single threaded ILP maximisation can be seen in Transmeta’s Crusoe line and famously in Intel and HP’s Itanium processors line.

Again, I think we are discussing semantics. Because what keeps you from exploiting all this parallel execution potential? Misprediction, cache misses, other stalls, etc…

For this the CPU will watch it's to have an instruction queue, to assess what data will it need next. And if it's missing, it will reorder those instructions.
Of course, that in modern superscalar processors, parallelism is almost a given. But having a good pre-fetcher, branch predictor and out of order execution is vital to keep all those pipelines fed and with as fewer stall as possible.
I think we get that :).

So although there are issues with OoO, the advantages far outstrip the disadvantages. And that is why all modern CPU cores are OoO. Even ARMs low power cores are now OoO.
Nobody is disputing that, nobody is saying it is not a good tool to have, not suggesting we try Itanium again… not today haha.
 

Zathalus

Member
So you don't have anything showing Sony saying anything.
We don't have anything about the PS5 Pro directly from Sony. Guess it doesn't exist then. PSSR was just something somebody made up.

Come on now, we are talking about reporting on the information Sony is releasing directly on the developer portal. They wouldn't claim VRS is a new feature if it wasn't.

Haha, sure breaking an NDA to correct them? Not worth it at the moment.
Well obviously not right now, but when it is officially unveiled.
 

winjer

Gold Member

Pentium 4’s branch predictor was estimated to be well above 95% too yet general purpose branchy code hurt its performance quite a bit (due to its very very deeply pipelined design too), the point was not about branch prediction rates, in the paper they talk about memory load/store instructions reordering and conflicts that may arise (as well as decreasing instruction cache hit rate).

Again, the miss rate we were talking about before was data cache misses and how much can the OOOE front end hide (10-20 cycles sure, 60-80 cycles … it depends… going further out to RAM, which is what Fafalada was saying it is miles off of that, no it is not designed to cover that… even very aggressive designs).

The problem with the Pentium's 4 pipeline was not the branch predictor, it was the length of the pipeline. It go to the point of being 30+ stages.
So when there was a stall because of a miss with the OoO, the resulting pipeline flush was catastrophic for all operations already in the pipeline.

Now mind you, I'm not saying that a good frontend can nullify cache and memory latencies. I'm saying it can help to hide it.
When it works well, a good branch predictor, pre-fetcher and OoO, can do wonders to keep a superscaler CPU with several pipelines, all running at peak performance.

OoO Execution is about maximising ILP as much as SMT is about TLP. Covering stalls, data dependencies, etc… (including and not in small part allowing to overcome the relatively tiny x86 register file limitations) is part of how the the back end will keep the execution units fed extracting as much parallelism from the instruction stream. How do you think they feed these monsters (incredibly wide backends otherwise?):

… and the modern M3 is based on even fatter cores ;). No SMT at all, so a single thread need to keep them fed.

Purely compiler based approach to single threaded ILP maximisation can be seen in Transmeta’s Crusoe line and famously in Intel and HP’s Itanium processors line.

Again, I think we are discussing semantics. Because what keeps you from exploiting all this parallel execution potential? Misprediction, cache misses, other stalls, etc…


I think we get that :).


Nobody is disputing that, nobody is saying it is not a good tool to have, not suggesting we try Itanium again… not today haha.

You might be mistaking what OoO means. It's just Out of Order execution.
Meaning the CPU look at the instruction queue and reorder it to avoid instruction stalls.
The feature in modern CPUs that allow for greater parallelism is superscaler, meaning there are several pipelines in the CPU. Though, of course that each stage has different amounts of units.
SMT is just an opportunistic feature that tries to fit secondary instructions, in a pipeline that already has different thread on it, but that ha left unused units.
In a way, good frontend and having SMT are counter productive, because a good frontend will leave fewer stages open in the pipeline.
That is why Intel is ditching SMT for it's future CPU architectures. And why Apple already left it behind.

For a while, many supposed that X86 could only have an 4-wide decode stage. While ARM isa, could have significantly more.
Intel proved everyone wrong, by having a 6-wide decode stage since 12th gen.
AMD is still with a 4-wide decode, but with strong throughput. I wonder what they will do with Zen5 regarding this.
 
Developer portal, so information they are relaying directly to everyone using the dev kit. I doubt Sony would state any of this in marketing material. The clock speeds, VRS, mesh shaders, and new cache configuration is all from Sony directly.

Again, it's how you spin the narrative, mesh shaders & VRS are available on PS5 also, it's an update to the API's and will be available across PS5/PS5 Pro
 

winjer

Gold Member
Again, it's how you spin the narrative, mesh shaders & VRS are available on PS5 also, it's an update to the API's and will be available across PS5/PS5 Pro

The PS5 does not have support for Mesh Shaders. Only for primitive shaders. And neither for hardware VRS.
The Pro will have them, very likely, considering it's based on RDNA3 and/or RDNA4.
 
The only performance metric we got was for the AI engine and FP16 numbers,so only we have for a reference point was those things.

Ehm no

Both Henderson and DF clearly that the TF number in the documents is 33.5 Tflops FP32 (dual-issue) but I guess it assumes the lowest possible clock speed that can reach the power limit

"Leaked specifications, derived from Sony's developer portal, suggest that the PS5 Pro has 30 WGP (Work Group Processors) delivering 33.5 teraflops of performance. This is up against the standard model with 18 WGP offering up an equivalent 10.23 teraflops.

"On the surface level, that's an extra 227 percent of performance, except that the same Sony documents suggest only an extra 45 percent of actual game throughput. Part of the explanation comes from the RDNA 3 architecture with its dual-issue FP32 support, which doubles the amount of instructions processed, but which does not typically double game performance."
 
Last edited:

Panajev2001a

GAF's Pleasant Genius
The peak clocks thing is directly from Sony themselves, not DF speculation.
It is not. Sony is saying the same thing (just watched the section in the video) as they did with PS5 boost clock. That it is the maximum it can run, anything else is speculation by DF and they mention it as such.
 

S0ULZB0URNE

Member
We don't have anything about the PS5 Pro directly from Sony. Guess it doesn't exist then. PSSR was just something somebody made up.

Come on now, we are talking about reporting on the information Sony is releasing directly on the developer portal. They wouldn't claim VRS is a new feature if it wasn't.


Well obviously not right now, but when it is officially unveiled.
We have nothing from Sony this is the only correct thing you said.

So you might want to go back and edit your posts claiming Sony said things.
 

Zathalus

Member
Again, it's how you spin the narrative, mesh shaders & VRS are available on PS5 also, it's an update to the API's and will be available across PS5/PS5 Pro
It's great that you believe that, but Sony claims the opposite. Nobody has reported that any of these features will be backported to the original PS5. Not that it really matter in the end as hardware VRS and the difference between primitive and mesh shaders are minor.

It is not. Sony is saying the same thing (just watched the section in the video) as they did with PS5 boost clock. That it is the maximum it can run, anything else is speculation by DF and they mention it as such.
The fact that Sony needs to clarify the difference between peak clocks, average TFLOP numbers, and they have stated it is the maximum clock speed but is usually power limited. Sony also said it is obtainable "on a limited subset of games". Not that is really matters, 2.18Ghz vs 2.35Ghz is basically unnoticeable by anyone.


We have nothing from Sony this is the only correct thing you said.

So you might want to go back and edit your posts claiming Sony said things.
Sure, if you want to be extra pedantic about things, I'll change my statement below:

What narrative exactly? That the Pro GPU is power constrained and cannot hit max clocks in all games? As reported by DF, that is directly from Sony themselves. That the Pro supports Tier 2+ VRS which the normal PS5 doesn't? Again, reported as directly from Sony themselves. That the Pro has full mesh shader support instead of primitive shaders (minor difference really, as Alex said simply programmatic). Bingo, reported as directly from Sony again.
 

S0ULZB0URNE

Member
It's great that you believe that, but Sony claims the opposite. Nobody has reported that any of these features will be backported to the original PS5. Not that it really matter in the end as hardware VRS and the difference between primitive and mesh shaders are minor.


The fact that Sony needs to clarify the difference between peak clocks, average TFLOP numbers, and they have stated it is the maximum clock speed but is usually power limited. Sony also said it is obtainable "on a limited subset of games". Not that is really matters, 2.18Ghz vs 2.35Ghz is basically unnoticeable by anyone.



Sure, if you want to be extra pedantic about things, I'll change my statement below:
Again you are claiming things to be said by Sony without showing Sony saying a single thing.
 

winjer

Gold Member
As far as the underlying RDNA2 HW they are the same thing as an AMD exec interview clarified. Mesh Shaders get compiled to primitive shaders, but that may have changed in RDNA4.

That is true. But there has top be something more to it.
Remember that RDNA1 on PC has trouble running a game like Alan Wake, with Mesh Shaders. But RDNA2 runs it just fine.
So why is the primitive Shaders in RDNA1 not good enough to be used with Mesh Shaders? Is it just drivers? Some hardware feature missing?

In this article AMD talks about they integrate AMDs Primitive Shaders into the Mesh Shader pipeline.
 
Last edited:

Zathalus

Member
Again you are claiming things to be said by Sony without showing Sony saying a single thing.
The implication for anyone following the conversation is that the Sony claims are being reported by DF. I'm not going to preface every single discussion with the Verge reported, or DF reported, or as per the leak from MLiD. In this context the Sony claims are what are being reported by DF (and others) on the official developer portal. As is everything being discussed in this thread.
 

S0ULZB0URNE

Member
The implication for anyone following the conversation is that the Sony claims are being reported by DF. I'm not going to preface every single discussion with the Verge reported, or DF reported, or as per the leak from MLiD. In this context the Sony claims are what are being reported by DF (and others) on the official developer portal. As is everything being discussed in this thread.
So no statement from Sony(like you clearly stated)
Just stuff from the known to have issue with PlayStation... DF.
 

shamoomoo

Member
Ehm no

Both Henderson and DF clearly that the TF number in the documents is 33.5 Tflops FP32 (dual-issue) but I guess it assumes the lowest possible clock speed that can reach the power limit

"Leaked specifications, derived from Sony's developer portal, suggest that the PS5 Pro has 30 WGP (Work Group Processors) delivering 33.5 teraflops of performance. This is up against the standard model with 18 WGP offering up an equivalent 10.23 teraflops.

"On the surface level, that's an extra 227 percent of performance, except that the same Sony documents suggest only an extra 45 percent of actual game throughput. Part of the explanation comes from the RDNA 3 architecture with its dual-issue FP32 support, which doubles the amount of instructions processed, but which does not typically double game performance."
That wasn't the information DF had when MLID posted that screen shot a while back.
 

Imtjnotu

Member
PS4 Pro was announced in September, they won't announce PS5 Pro any sooner than that
Jimmy Fallon Of Course GIF by The Tonight Show Starring Jimmy Fallon
 

S0ULZB0URNE

Member
There won't be any statements from Sony until the Pro officially launches. But other sources will be able to confirm the information listed on the developer portal.
He clearly said Sony said things.

What narrative exactly? That the Pro GPU is power constrained and cannot hit max clocks in all games? That is directly from Sony themselves. That the Pro supports Tier 2+ VRS which the normal PS5 doesn't? Again, directly from Sony themselves. That the Pro has full mesh shader support instead of primitive shaders (minor difference really, as Alex said simply programmatic). Bingo, directly from Sony again.

Yet not a single thing has been said by Sony.
 

Zathalus

Member
Continue to lie about Sony saying things with no proof as well.
If you want proof of anything then I'm not sure what you are doing in a speculation and leak thread. Obviously the implication of anything being discussed here is through the lens of it being leaked by gaming websites and media.
 
That is true. But there has top be something more to it.
Remember that RDNA1 on PC has trouble running a game like Alan Wake, with Mesh Shaders. But RDNA2 runs it just fine.
So why is the primitive Shaders in RDNA1 not good enough to be used with Mesh Shaders? Is it just drivers? Some hardware feature missing?

In this article AMD talks about they integrate AMDs Primitive Shaders into the Mesh Shader pipeline.
Also curious that the PS5 performs similarly to the Series X in this game even though the latter uses Mesh Shaders as opposed to the PS5 which has some sort of "RDNA 1 Primitive Shader feature".
 
All this talk about it while my confidence in SIE's leadership as a platform holder at an all-time low. Needless to say, I just don't have much interest in the Pro anymore.

Some of the technology is neat, like PSSR, but I have little faith the 1P studios will have the freedom or support to push the thing even as an upres/framerate boosting box, let alone 3P studios. And that's on top of the litany of other issues I've come to have with SIE over the past few months.

It is what it is I guess. But outside of the small curiosities here or there, the PS5 Pro itself just feels...very boring...to me.
 

S0ULZB0URNE

Member
No need to jump through any hoops. We just need independent confirmation of the contents of the developer portal.
But you are as he quoted Sony and Sony saying what he said is nowhere to be found.
We do need confirmation for all of this and speculating is great. Making stuff up isn't.
 

FireFly

Member
But you are as he quoted Sony and Sony saying what he said is nowhere to be found.
We do need confirmation for all of this and speculating is great. Making stuff up isn't.
His original quote was missing a "DF claims" in reference to the developer portal information, since it has not been independently verified. So yes, that's his terrible crime, for which he must be punished by the forum gods. But my point is simply that the DF article may turn out to be perfectly accurate representation of what is on the portal.
 

twinspectre

Member
This is BS I won't buy it (as a protest), but I know theres tons of people ready with their wallet so, I can't change the entire industry, but at least I'm not the one to blame when Sony continues to screw over their customers, so I can continue to say "YOUR FAULT PEOPLE", lol.
 

SonGoku

Member
If it wasn't for the fact that this will be more about Ray-Tracing I would be worried about the 2 SE


Going from 18 CUs per SE to 30 CUs per SE is crazy
I worry about this too but from the latest DF thread it seems like the SEs have been beefed up to accommodate the extra CUs
 
Top Bottom