twilo99
Gold Member
Yes, yes they do.
-
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)
- Thread starter jm89
- Start date
Yes, yes they do.
Moses85
Member
Audiophile
Member
Hopefully we'll get 30th Anniversary Limited Edition Pros and Slims like the launch PS4 had for the 20th Anniversary. As much as I like the PS1 though, it'd be cool if they themed it around the PS2 this time (or one of each).
I'm pretty sure that's how phones were sold for years until companies successfully inflated their prices.
Last edited:
It depends on what you mean by "cutting edge technology" in some ways it is cutting edge technology but it certainly isn't "most powerful" for $500. People are not dumb. They know that.
Loxus
Member
Supposedly, because lots of things not adding up.
Mibu no ookami
Demoted Member® Pro™
What isn't adding up?
lh032
I cry about Xbox and hate PlayStation.
they do, especially now. People change.
Eiknarf
Banned
Oh yeah, people do. There are thousands of people who are ignorant to what’s inside a PC and console
YeulEmeralda
Linux User
The PS3 was pretty much cutting edge but it almost sank Sony.
MikeM
Gold Member
Unless the issue is on the CPU side (which I doubt), you’ll see probably a close to 60fps on a Pro with that about 1.7x GPU increase (ignoring other potential bottlenecks).
PaintTinJr
Member
Do we have a expanded acronym for the "MF" part of the PSML MFSR in that document?
Along with my much older predictions that Spectral Analysis is using dynamic ML based on histogram frequency analysis to concentrate ML processing on identifying high frequency areas in the native renderer used for training, first and then trying to recover as much of that lost high frequency detail in the upscale, where the player will notice it most.
If the "MF" part hasn't been labelled, yet I wonder if it stands for "multiple frequency".
Gaiff
SBI’s Resident Gaslighter
Good thing the world doesn’t revolve around you.
Should be multiple-frame or some such given how it operates.
Daneel Elijah
Member
When I first learned about RT, it was presented to me as the graal of light rendering. Being able to simply simulate light instead of loosing time to fake it, it would allow devs to iterate much faster and get better results. Spiderman 2 shows that you can go RT only in a AAA PS5 game, and in the future some of that promise may become reality for this gen and future ones. I see the PS5 Pro, and the focus on a DLSS equivalent to do the same in the long term. Helping getting better games, faster. I know that it won't be the case at first, but I hope that it will one day.
SF Kosmo
Al Jazeera Special Reporter
Yeah unfortunately I don't think it's quite enough on the RT front. I think some of these PC games that are doing path tracing really are realizing that dream of truly next gen lighting but it requires more than what's on offer here, and nVidia is so very far ahead on both RT and AI-baded technologies needed to make it work that I don't think Song can catch up with AMD silicon.
Daneel Elijah
Member
The PS5 is probably not goot enough. The PS5 Pro, with PSSR and dual compute, who knows? Hopefully the PS6 will have path tracing, and that the PS5 Pro will somehow be able to cope. A man can dream. As for AMD not being able to compete, I don't think that will be a problem for Sony. As long as they can sell 100+ millions consoles each gen, they will get to be part of the cutting edge by default, like they made SSD standard now, even if it was obviously already the case for PC for a certain time already, but they and Microsoft raised the bar. Same with RAM for another example. Unless Nividia can use their advance to control how we will do path tracing, be it by patents or people being so accoustumed to do it their way that AMD will have to follow?
Loxus
Member
MLID and DF having access to this kind of info without prior history and not having a good history of Sony leaking console specs.
2 Shader Engine with 60CUs don't match Cerny's previous design of 18CUs.
300TOPs and 67TFLOPs not matching.
300TOPs not matching the CU count or clocks.
The 7900XTX with 96CUs (192 AI Accelerators) and 2.5 GHz only has 123TOPs.
2.35 GHz with 60CUs not matching the 300TOPs and 67TFLOPs.
Drop clocks to 2.18 GHz not making sense. Strix Halo is rumored to run at upto 3.0 GHz @ 175W.
Sony having a NPU for that 300TOPs would double die size looking at Strix Point NPU size, which only has 50TOPs.
PSSR doesn't make sense from an R&D prospective. It makes more sense to share R&D with AMD to add AI/ML into FSR. Which makes sense for PC releases and we see Sony already updating the some of their PC games to FSR3.
The list goes on.
At the end of the day, Sony is still using AMD hardware. Sharing R&D helps keep the cost of the PS5 Pro down.
I'm not saying the PS5 Pro isn't real or not, I'm just saying don't expect leaks to be 100% accurate. We can't keep falling for rumors like we did with Tommy Fisher.
Zathalus
Member
Playstation Machine Learning Multi Frame Super Resolution.
As per Sony themselves “PSSR is a machine learning-enhanced version of Temporal Anti-Aliasing Upscal (TAAU). It will replace a game's existing temporal anti-aliasing or temporal upsampling. The inputs are "quite similar to DLSS or FSR" with full HDR support.”
It’s all on the second page of that document.
Last edited:
Zathalus
Member
GPU architectures change, you can’t expect the GPU to keep a butterfly approach forever. PS4 Pro was the first attempt at a Pro console, I’d imagine the hardware and software design for regular console emulation has become far more advanced.
We have no idea how those numbers were calculated as the clock speeds of the GPU have a larger variance than prior consoles. Also the 300TOPs could be rounded up or down because it is so specific.
The comparison with the 7900XTX is also useless as we are talking about a new GPU architecture that is bringing up some large updates to ML acceleration.
Different GPU archs, different nodes (likely 6nm for Pro vs 4nm for Strix).Drop clocks to 2.18 GHz not making sense. Strix Halo is rumored to run at upto 3.0 GHz @ 175W.
Sony having a NPU for that 300TOPs would double die size looking at Strix Point NPU size, which only has 50TOPs.
PSSR may very well be based on a ML version of FSR that AMD plans to announce with RDNA4. Or Sony might have developed it because AMD is not bothering to do so. We just don’t have enough information yet.
The leaks have come from like 5 different sources at this point.The list goes on.
At the end of the day, Sony is still using AMD hardware. Sharing R&D helps keep the cost of the PS5 Pro down.
I'm not saying the PS5 Pro isn't real or not, I'm just saying don't expect leaks to be 100% accurate. We can't keep falling for rumors like we did with Tommy Fisher.
Last edited:
ChiefDada
Gold Member
The PSSR architecture is fully custom, therefore I honestly don't think it's appropriate to call this particular area of the silicon "AMD hardware".
Loxus
Member
You do know AMD can provide Sony with two choices for AI/ML right?
XDNA and RDNA with the AI accelerators.
This thinking is like thinking Sony would drop AMD RT to create their own. It just doesn't make sense from a cost prospective.
Just share R&D and build on top of what AMD already has. That way, we can avoid having a $700 console.
TheRedRiders
Member
"quite enough" is subjective in my opinion and will differ depending on who you are asking.
I think when Mark Cerny was designing the PS5 Pro, he didn't say "let's bump up the RT by 2.5-3x" and call it a day. I Imagine judging by his past history that he approached it by saying "how much RT capability do we need so that developers can implement their RT solution in a significantly more meaningful way than they did on the PS5", or "how much RT do we need so that we can make RTGI, RT Shadows and Reflections all run simultaneously in realtime at 60 or 30 FPS". I also imagine that there was heavy developer feedback in this process as well.
Path tracing is still far from becoming a standard on PC hardware when we account for both GPU capability and especially on the number of titles implementing it. We're still years off from even full ray tracing becoming a standard let alone path tracing. I suspect however that the PS5 Pro is a step in the right direction, and hopefully it will encourage more developers to pursue and develop a fully tay-traced pipeline in their game engines.
In a related subject, I'm curious if whether Sony has gone out fishing for marketing deals with 3rd party developers to show off the Pro. If they haven't approached 4A Games yet I think they're fools... Imagine what the next Metro game will be able to pull off in the ray tracing department.
Loxus
Member
Clearly you don't understand AMD hardware and anything from Road to PS5.
We aren't lacking info from RDNA3.5 or RDNA4 either.
We have lots of info here.
AMD RDNA 3.5’s LLVM Changes
Examining AMD’s RDNA 4 Changes in LLVM
I can't be the only one here that's trying to make sense of the leaks. Where are all the tech guys from 2019-2020 are?
ChiefDada
Gold Member
Uh ok but Sony went with their own solution for whatever reason, more likely than not because it makes sense for purposes of levereging strictly for the Playstation console.
This thinking is like thinking Sony would drop AMD RT to create their own. It just doesn't make sense from a cost prospective.
Clearly Sony has done the cost benefit analysis and decided to go "fully custom" with the ML upscaling architecture, and alternatively for RT, go with the "more powerful ray-tracing architecture" that is most likely lifted from RDNA 4.
There's not much "sharing" that can be done when you're building special purpose architecture that is to be used specifically for one platform, similar to PS5 I/O. Besides, all evidence points towards AMD FSR tech leaning general purpose for all GPUs to be able to run without the need for specialized hardware.
PanzerCute
Member
Even the most casual of my friends know that PC is far better, what the fuck are you talking about.
Loxus
Member
This is the guy who first leaked the PS5 Pro specs before that leak document.
AMD doesn't have SSD/IO hardware available, so it's obvious Sony would have to create their own.
Last edited:
ChiefDada
Gold Member
Right, Sony is using their own proprietary hardware for ML upscaling. That's what I've been saying.
The document isn't fake. Sony successfully struck it down via copyright claim. They were legit.
Loxus
Member
Hurricane Beryl is affecting my island and I had stuff doing.
I meant to say this
So either the document is fake and Sony is using their own hardware or the document is real but that 300TOPs is just the total TOPs.
Like this.
But I edited out to make it a separate post, cause the 300TOPs still ain't adding up.
Last edited:
MisterXDTV
Member
Yeah, the combination of AMD's RDNA4 and Sony's PSSR is gonna be what makes or breaks this machine in appeal
I'm really confident about it
Last edited:
Physiognomonics
Member
Sure and no surprises here. They also were and still are fully custom with their ID buffer that is still used in some PS5 games for different things, not only reconstruction. People are still surprised that some games using CBR are showing better results on PS5 like in Dragon dogma 2? Of course it's easy to say it's a bug on Xbox. But the most logical explanation would be that PS5 has hardware advantages with CBR reconstruction in Dragon Dogma 2 as it's the actual truth.
Last edited:
ChiefDada
Gold Member
Stay safe.
Which document are we talking about? The MLiD leak, which is what I'm referring to, states the Pro is using "fully custom architecture" for PSSR. My position is the MLiD docs which cites custom ML architecture is real because Sony successfully executed a copyright strike against the leak.
or the document is real but that 300TOPs is just the total TOPs.
Like this.
But I edited out to make it a separate post, cause the 300TOPs still ain't adding up.
Nah, it wouldn't make sense for Sony to list TOPS performance related to other areas of GPU/CPU. the 300 TOPS is specifically related to the dedicated PSSR hw imo.
Zathalus
Member
Road to PS5 is irrelevant when discussing the Pro. GPU architecture and priorities change over time. The link to RDNA4 mentions improved tensor capability and sparsity. The 300 TOPs figure is likely calculated with sparsity, as NVIDIA has done with newer generations.
Loxus
Member
The PS5 Pro is still a PS5, just more powerful. If not, it would be the PS6, so Road to PS5 still applies.
We can't ignore the things Mark Cerny said in Road to PS5, for example:
"First we have a custom AMD GPU based on there "RDNA2" technology. What does that mean? AMD is continuously improving and revising their tech for RDNA2. Their goals were roughly speaking, to reduce power consumption by re-architecting the GPU to put data close to where it's needed, to optimize the GPU for performance and to add new more advanced feature set.
But that feature set is malleable, which is to say that we have our own needs for PlayStation and that can factor into what the AMD roadmap becomes.
So collaboration is born.
If we bring concepts to AMD that are felt to be widely useful, then they can be adopted into RDNA - and used broadly, including in PC GPUs.
If the ideas are sufficiently specific to what we're trying to accomplish, like the GPU cache scrubbers I was talking about, then they end up being just for us.
If you see a similar discrete GPU available as a PC card at roughly the same time as we release our console, that means our collaboration with AMD succeeded in producing technology useful in both worlds.
It doesn't mean that we, as Sony, simply incorporated the pc part into our console."
The bold in particular is very important.
RT and the AI Accelerators are useful in both the PS5 Pro and AMD GPUs.
RDNA4 and the PS5 Pro are releasing roughly the same time, so Sony and AMD collaborated again. This would apply to FSR as well. If PSSR is true, it'll just be FSR with Sony branding. Nothing majorly different.
We have some RDNA4 specs already.
The closest thing to the PS5 Pro is either Navi32 or Navi48 but then that contradicts the leaks, since Navi32 has 3 Shader Engines and Navi48 has 4 Shader Engines and the PS5 Pro is rumored to have 2 Shader Engines.
PS5 Pro with 1 WGP(2CU) disabled for yields.
Navi32 = 60CU(3SE) / 54CU active for yields.
Navi48 = 64CU(4SE) / 56CU active for yields.
On to the sparsity talk.
Examining AMD’s RDNA 4 Changes in LLVM
RDNA 4 carries these instructions forward with improvements to efficiency, and adds instructions to support 8-bit floating point formats. AMD has also added an instruction where B is a 16×32 matrix with INT4 elements instead of 16×16 as in other instructions.
Since SWMMAC takes a sparse matrix where only half the elements are stored, perhaps RDNA 4 can get a 2x performance increase from sparsity.
Using MLiD clocks.
- FP32 = 60 CU × 4 SIMD32 × 32 × 2 × 2.18 GHz = 33.5 TFLOPs
- FP16 = 256 × 2 AI Accelerators × 60 CU × 2.18 GHz = 67 TFLOPs
- Int8 = 512 × 2 AI Accelerators × 60 CU × 2.18 GHz = 134 TOPs
- 2x performance increase from sparsity = 268 INT8 TOPs
Using Digital Foundry clocks.
- FP32 = 60 CU × 4 SIMD32 × 32 × 2 × 2.35GHz = 36.1 TFLOPs
- FP16 = 256 × 2 AI Accelerator × 60 CU × 2.35 GHz = 72.2 TFLOPs
- Int8 = 512 × 2 AI Accelerators × 60 CU × 2.35 GHz = 144 TOPs
- 2x performance increase from sparsity = 288 Int8 TOPs
If we go by Kepler 2.45 GHz.
- FP32 = 60 CU × 4 SIMD32 × 32 × 2 × 2.45GHz = 37.6 TFLOPs
- FP16 = 256 × 2 AI Accelerator × 60 CU × 2.45 GHz = 75.2 TFLOPs
- Int8 = 512 × 2 AI Accelerators × 60 CU × 2.45 GHz = 150.5 TOPs
- 2x performance increase from sparsity = 301 Int8 TOPs
This is why I call the leak document, supposedly. I be keeping notes on the leaks and seeing a lot of inconsistencies.
me as well i deleted most of my ps5 stuff will redownload them when i get the pro.
TheRedRiders
Member
How do you guys think Sony will make a profit on the Pro? through hardware sales or will they charge extra for Pro versions of the game? or even both?
Nah they won't charge extra for games, that would be suicide. It will just keep focus on them for the next few years and also I doubt they will lost money on hardware.
I know NVidia does this but 'sparsity = fixed multiplier' is incredibly misleading. Not unlike using a fixed multiplier for lossless compression like a certain series...
Anyway I'm also not sure the math necessarily applies - every NVidia GPU since Turing had an exact 2x multiplier on TensorOps vs. RegularOps (ie. clearly there's some sort of execution resource sharing there, regardless of what dedicated 'units' we talk about) - so I'd be more inclined to believe that's what's happening here.
IIRC the leak did reference sparsity but as a separate optimisation - and I've never known Sony to do accounting of things in developer docs - it's not a marketing spec sheet.
The PS5 Pro is still a PS5, just more powerful. If not, it would be the PS6, so Road to PS5 still applies.
We can't ignore the things Mark Cerny said in Road to PS5, for example:
"First we have a custom AMD GPU based on there "RDNA2" technology. What does that mean? AMD is continuously improving and revising their tech for RDNA2. Their goals were roughly speaking, to reduce power consumption by re-architecting the GPU to put data close to where it's needed, to optimize the GPU for performance and to add new more advanced feature set.
But that feature set is malleable, which is to say that we have our own needs for PlayStation and that can factor into what the AMD roadmap becomes.
So collaboration is born.
If we bring concepts to AMD that are felt to be widely useful, then they can be adopted into RDNA - and used broadly, including in PC GPUs.
If the ideas are sufficiently specific to what we're trying to accomplish, like the GPU cache scrubbers I was talking about, then they end up being just for us.
If you see a similar discrete GPU available as a PC card at roughly the same time as we release our console, that means our collaboration with AMD succeeded in producing technology useful in both worlds.
It doesn't mean that we, as Sony, simply incorporated the pc part into our console."
The bold in particular is very important.
RT and the AI Accelerators are useful in both the PS5 Pro and AMD GPUs.
RDNA4 and the PS5 Pro are releasing roughly the same time, so Sony and AMD collaborated again. This would apply to FSR as well. If PSSR is true, it'll just be FSR with Sony branding. Nothing majorly different.
We have some RDNA4 specs already.
The closest thing to the PS5 Pro is either Navi32 or Navi48 but then that contradicts the leaks, since Navi32 has 3 Shader Engines and Navi48 has 4 Shader Engines and the PS5 Pro is rumored to have 2 Shader Engines.
PS5 Pro with 1 WGP(2CU) disabled for yields.
Navi32 = 60CU(3SE) / 54CU active for yields.
Navi48 = 64CU(4SE) / 56CU active for yields.
On to the sparsity talk.
Examining AMD’s RDNA 4 Changes in LLVM
RDNA 4 carries these instructions forward with improvements to efficiency, and adds instructions to support 8-bit floating point formats. AMD has also added an instruction where B is a 16×32 matrix with INT4 elements instead of 16×16 as in other instructions.
Since SWMMAC takes a sparse matrix where only half the elements are stored, perhaps RDNA 4 can get a 2x performance increase from sparsity.
Using MLiD clocks.
Still a ways of from 300TOPs.
- FP32 = 60 CU × 4 SIMD32 × 32 × 2 × 2.18 GHz = 33.5 TFLOPs
- FP16 = 256 × 2 AI Accelerators × 60 CU × 2.18 GHz = 67 TFLOPs
- Int8 = 512 × 2 AI Accelerators × 60 CU × 2.18 GHz = 134 TOPs
- 2x performance increase from sparsity = 268 INT8 TOPs
Using Digital Foundry clocks.
Close but Sony would of said 288 or 290 TOPs.
- FP32 = 60 CU × 4 SIMD32 × 32 × 2 × 2.35GHz = 36.1 TFLOPs
- FP16 = 256 × 2 AI Accelerator × 60 CU × 2.35 GHz = 72.2 TFLOPs
- Int8 = 512 × 2 AI Accelerators × 60 CU × 2.35 GHz = 144 TOPs
- 2x performance increase from sparsity = 288 Int8 TOPs
If we go by Kepler 2.45 GHz.
The 300TOPs are very close but then the TFLOPs are way off from the leak document.
- FP32 = 60 CU × 4 SIMD32 × 32 × 2 × 2.45GHz = 37.6 TFLOPs
- FP16 = 256 × 2 AI Accelerator × 60 CU × 2.45 GHz = 75.2 TFLOPs
- Int8 = 512 × 2 AI Accelerators × 60 CU × 2.45 GHz = 150.5 TOPs
- 2x performance increase from sparsity = 301 Int8 TOPs
This is why I call the leak document, supposedly. I be keeping notes on the leaks and seeing a lot of inconsistencies.
If I'm not mistaken, Naughty Dog and Insomniac Games use TAA, also,Sony Santana Monica used ML for GOW: Ragnarok,I don't see why Sony would need a custom version of FSR of some of their studios were already doing temporal anti aliasing.
Heck, Sony problem could bring back MLAA or get a boost in checkerboard rendering.
Ashamam
Member
One or both could be true. Personally I subscribe to the theory the RT is 'inherited' but PSSR is not. Why? Because Sony has its own IP they use for TV upscaling which in theory could contribute into their own tech running on much faster hardware eliminating the usual TV processing time cost. Im not saying it is the same thing, because obviously PSSR has game centric inputs ala DLSS or FSR, but I think PSSR will utilise a blend of techniques. When comparisons start to be made I predict there will be unexpected results, probably surprising advantages, but at the very least differences that make it obvious its not a clone of existing techniques but has its own distinct characteristics.
The actual hardware itself is derived from AMD and not particularly bespoke to Sony (at least in the core sense, arrangement may be), but the software will be. I also wonder whether the iSize acquisition contributed to the underlying IP in use.
Last edited:
LiquidMetal14
hide your water-based mammals
I wish this was out in time for Astro Bot
ChiefDada
Gold Member
Rocco Schiavone
Member
Wouldn't change much. Astro is 4k/60 on a normal PS5.
Last edited:
DenchDeckard
Moderated wildly
Do we think this will be more successful than ps4 pro or less?
I think if they have an awesome demo and show of the PSSR well it could be more impressive than the PS4 pro and lead to more sales.
I think if they have an awesome demo and show of the PSSR well it could be more impressive than the PS4 pro and lead to more sales.
DJ12
Member
Well I'm definitely buying one and skipped the ps4 pro so they will get at least one more sale
VanCityGamer
Member
Definitely less successful.