MisterXDTV
Member
+67% refers to the Shading Units (CUs)
2304->3840 = +67%
same as:
36->60 = +67%
Last edited:
-
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.
Digital Foundry - Playstation 5 Pro specs analysis, also new information
- Thread starter Gaiff
- Start date
+67% refers to the Shading Units (CUs)
2304->3840 = +67%
same as:
36->60 = +67%
I'll probably be stating the obvious but I disagree on this.
XBox actually frequently had bandwidth contention issues - that was the weakest part of its chain, even though hardware itself was stupidly overpowered compared to competition (basically it was the 1X console of that generation - but there was no other 'pro/X' competitor on the market, the gulf was that big).
But it was just that - the box was all about brute-forcing things - so design philosophy was actually a lot like the PS2 - they just took a different architecture approach to the same end-goal.
DC was the other extreme of the spectrum - all about doing more with less, and it did that brilliantly. But the console was also built around design constraints aiming for far less fill/bandwidth hungry processing than the rest of the market allowed. Ie. this was the generation when practical postprocessing pipelines became a reality (HDR, camera-lens effects, accumulation effects), increasing operations per pixel (first iterations of per-pixel compute), and of course high-density particle simulations, and DC design was not particularly friendly to any of it.
It had the best ratio of power/compute+features for graphics-pipeline practices from the late 90ies, but it was just not built to cope with what early 00s became all about.
PS2s biggest weakness was the trade-offs made to ship that GPU in March 2000. GS initially was supposed to have certain quality of life improvements (proper MipMap filters, bi-directional bus addressing etc.) and 8MB of eDram. Some of those are more minor than others (texture filters would have just put IQ on par with others, so it would remove 95% of jaggy complaints, but not add anything to capabilities as such).
Larger eDram would have made those practical post-process computations a 'lot' more practical (while what we got could contend with the XBox 'sometimes', 8MB would have put it comfortably up there), and bi-directional bus would have made deferred shading a practical solution on a 1999 hardware - some 7 years before it became practical everywhere else.
The rest of the the system was - IMO - basically as good as it could have been for the time it released, they had the most sophisticated geometry pipeline of that generation(it literally introduced concepts like meshlets 20 years into the future), packed a lot of extra useful I/O hardware (like the stupidly overpowered MPEG2 Asic, geometry compression hw...) etc.
XBox had some similar things (Vertex shaders with more limitations, but more raw power, Dolby Asic etc.) but a year later and with a BOM at like 2x.
I mean if we talk about design goals as part of 'balance' - I think all 3 achieved what they set out to do (as did the GC, really), but usually it's all about the 'relative to market', and that's what I debate above.
Lysandros
Member
I don't think it works quite like that. A X% faster GPU requires this much more bandwidth in order to be truly X% faster overall. Like a bigger engine needing more fuel. If bandwidth happens to be the bottleneck in a specific scene the performance increase from PS5 can well be limited to just around 25%. Based on the leak PS5 PRO is even more lopsided in the matter bandwidth to compute (and other GPU resources) ratio compared to PS5. With the increase in resolution and/or performance targets PS5 PRO could well be bandwidth bound more often.PS5 and XSX often caused frame drops due to bandwidth limitations when using a lot of alpha effects.
The synergistic effect of PS5Pro's efficient access and 28% increase in bandwidth + 45% faster GPU speed may result in (up to) twice the frame rate.
PS5 shares 448GB/s of memory bandwidth between the CPU, GPU, and Tempest 3D Audio Engine(can use over 20GB/s).
That being said we the need official specifications and clarifications, those are just leaks regardless of how genuine they are.
Last edited:
Loxus
Member
This is another credible GPU leaker that leaked Strix Point.
Came in with another leak, this time it's Strix Halo clock speed.
40CU × 4 SIMD32 × 32 × 2 × 3.000GHz = 30.72 TF
This thing is nearly as powerful as the PS5 Pro.
What makes this more interesting, is Strix Halo is rumored to have a max TDP of 120W.
If RDNA3+/4, PS5 Pro hitting 3GHz + could be possible is Sony wanted to.
Came in with another leak, this time it's Strix Halo clock speed.
40CU × 4 SIMD32 × 32 × 2 × 3.000GHz = 30.72 TF
This thing is nearly as powerful as the PS5 Pro.
What makes this more interesting, is Strix Halo is rumored to have a max TDP of 120W.
If RDNA3+/4, PS5 Pro hitting 3GHz + could be possible is Sony wanted to.
Last edited:
Lysandros
Member
This is another credible GPU leaker that leaked Strix Point.
Came in with another leak, this time it's Strix Halo clock speed.
40CU × 4 SIMD32 × 32 × 2 × 3.000GHz = 30.72 TF
This thing is nearly as powerful as the PS5 Pro.
What makes this interesting, is it's Strix Halo is rumored to have a max TDP of 120W.
If RDNA3+/4, PS5 Pro hitting 3GHz + could be possible is Sony wanted to.
Now that's the kind of design i want to see. That would be a little beast.
Last edited:
MisterXDTV
Member
I don't get this, a GPU clock of 3 Ghz???
Unreal as in: I don't believe it
Unreal as in: I don't believe it
Last edited:
playsaves3
Member
Hmm I see what your saying how exactly are they doing the conversion to get the 45% it it based ins tflops or rops
MisterXDTV
Member
"Playstation says these combined make the Pro 45% faster than the standard PlayStation 5 and can provide twice the rendering speed of the standard console."
so 100% faster rendering speed but 45% faster of what kind of speed??
Too confusing
so 100% faster rendering speed but 45% faster of what kind of speed??
Too confusing
Last edited:
I guess the obvious implication would be that 2x rendering speed does not mean 2x framerate. But they might just be referencing the fact some games will/can double resolution on the Pro while maintaining performance. But indeed too little details to really know what this is all about.
Lysandros
Member
It wouldn't make sense to base it on a single GPU metric, just like it didn't make sense in XSX' case (vs. PS5). For me 45% faster rendering simply means 45% more FPS at the same resolution/IQ using the whole GPU (outside of increased ML/RT capabilities). Just 'drop' the PS5 code of a game running at ~35 FPS on PS5 PRO and reach ~50 FPS without further optimization for example.
Last edited:
MisterXDTV
Member
So it's like the RAW performance upgrade you would get doing nothing to the original code?
Seems like a pretty big number then
Last edited:
Bojji
Member
It's just GPU upgrade, PS4 Pro was over 100% more powerful than PS4 and on much newer architecture.
Here you have 63% in pure Teraflops and architecture for raster performance is more or less the same, only gets much bigger boost with RT.
People are disappointed because this is weak upgrade even compared to PS4 Pro that was considered very mediocre upgrade (and full of bottlenecks) vs Xbox One X.
They are just talking nonsense. Numbers we have are:
63% in TF
45% in actual in game performance (Sony words)
2-4x RT speed upgrade
Last edited:
MisterXDTV
Member
Anyway looking at the AMD leaks it seems like PS5 Pro could actually have a true RDNA 4 GPU - a Navi 48 chip with 4 CUs disabled for yields and an underclock
The latest leak suggests that Navi 48 has 32 WGP = 64 CUs, 256 Bit GDDR6 Memory
Just right on point
The latest leak suggests that Navi 48 has 32 WGP = 64 CUs, 256 Bit GDDR6 Memory
Just right on point
Last edited:
Panajev2001a
GAF's Pleasant Genius
Nope, CUs and ROPS are likely changed too where they fall between RDNA3 and RDNA4 is beyond me but them specifying VRS in the docs means they upgraded that unit too, so it is not the same as PS5.
Which was perhaps the biggest problem with the base PS5, but then again there is more and we will see it (AI driven PSSR is not a small fest, but the tone of your posts makes it clear you have other tendencies).
3-4 years after PS5 at acceptable costs has limits, will not post for the 1,000th time how semiconductor tech has been slowing down more and more (for a good while) and designing chips on newer nodes and manufacturing them is also getting more and more expensive… more consoles more often is not the answer. Still, they have taken as an ambitious crack as PS4 Pro if not better.
The 45% is tied to the fixed function units in the Shader Engines
So if it's going from 2 SE to 3 SE
So the simple way of getting the answer is 2 x 2.23 (GHz ) = 4.46 units of fix function rendering
While 3 x 2.18 GHz) = 6.54 units of fix function rendering
4.46 + 46% = 6.51
(This is without using the exact clock speed it's rough math )
So if it's going from 2 SE to 3 SE
So the simple way of getting the answer is 2 x 2.23 (GHz ) = 4.46 units of fix function rendering
While 3 x 2.18 GHz) = 6.54 units of fix function rendering
4.46 + 46% = 6.51
(This is without using the exact clock speed it's rough math )
Bojji
Member
On paper it's much weaker and less impressive jump vs the PS4 Pro.
But I get it, it's much harder to make better console where there almost no nodes reductions anymore. Only thing you can do is to make GPU bigger and more power hungry and that's exactly what they are doing here.
But there are power and cost limits to this ^ I think PS6 will be the weakest gen to gen jump ever, last time we had half gen jump after three years and it was over 2x in GPU power, next gen (PS5) was 4x/5x of PS4 power. Now half gen upgrade after four years is just 1.5x in GPU power. PS6 looks to be just 3x more powerful than PS5? 4090 level in 2027/8...
Gaiff
SBI’s Resident Gaslighter
That doesn’t make much sense either though. That only puts it on the level of a 6800 and according to
HeisenbergFX4
, it’s more like a 4070/3070 which are ~80% faster in rasterization and 100% in some RT workloads.You also have to take into account that these stats are meant for developer and they further state: Playstation says these combined make the Pro 45% faster than the standard PlayStation 5 and can provide twice the rendering speed of the standard console.
The numbers don’t add up at all. 45% is suspiciously close to the ROPs increase + clock speed which is 46% faster than the base console.
Perhaps they do mean worst-case scenario but that’d just be weird. They would provide a number which can be generally trusted.
Last edited:
Panajev2001a
GAF's Pleasant Genius
It is what it is with only 3-4 years of R&D and constraints a mid generation upgrade has (you cannot always double performance, it ain’t that easy
). PS6 Pro to your eyes will be even worse and you will not be wrong, but that is not because mean Sony is cheapening out though… again see above .Physiognomonics
Member
Yep. This should be PS5 Pro GPU. Navi 48 (prototype).
This is another credible GPU leaker that leaked Strix Point.
Came in with another leak, this time it's Strix Halo clock speed.
40CU × 4 SIMD32 × 32 × 2 × 3.000GHz = 30.72 TF
This thing is nearly as powerful as the PS5 Pro.
What makes this more interesting, is Strix Halo is rumored to have a max TDP of 120W.
If RDNA3+/4, PS5 Pro hitting 3GHz + could be possible is Sony wanted to.
Not on 6nm. But anyways Strix lacks AI upscaling and new RT units. Such a machine would be pointless for a PS5 Pro machine and would just be a PS4 Pro repeat. But wait, what's the supposed serial number of PS5 Pro vs this thing? I mean the biggest disappointment is PS5 Pro using 6nm IMO and this is RDNA3.X and is using N4?
EDIT: PS5 Pro serial is lower indeed. 1115 vs 1150 for Strix Point. Everything points to PS5 Pro being 6nm only.
Last edited:
Panajev2001a
GAF's Pleasant Genius
I think PS6 will have the advantage of being a full new generation and they can completely revamp the architecture. It is also about ~4 years off, but sure it gets more and more difficult to make these kind of large jumps.
Loxus
Member
If the Pro is Navi 48 prototype, it should be on N4P as well.
Strix Halo should have the same AI Accelerators + XDNA 2. But your right about the RT.
It would represent a rough doubling in performance per watt when compared with the 7700 XT. I am not sure AMD has ever made such a leap.
Last edited:
Loxus
Member
Physiognomonics
Member
That leak also states 56 CUs, 4.4 ghz CPU and 2ghz GPU...That GFX:1115 is from this leak, which states N4P.
Last edited:
Loxus
Member
Exactly, the PS5 Pro GFX is currently unknown.
TheRedRiders
Member
Would be curious as to see what process Viola is fabbed on. Some are saying 6nm, and some are saying 4nm.That GFX:1115 is from this leak, which states N4P.
Do we know what kind of TDP we can expect from the Pro?
Loxus
Member
I'd expect the same as the PS5 base on the fact CPU and GPU clocks are so low.
350W power supply: 200-220W while gaming.
TheRedRiders
Member
Assuming N6 or N4?
Loxus
Member
Physiognomonics
Member
Based on this 6nm would be incredibly short-sighted indeed. Litteraly a deadend with no cheap evolution possible. This should be on N5 indeed at the minimum. It would be stupid to release an APU you plan to make 20 million units on using N6 in 2024. Cheap for short term, costly for long term.N5.
It's not too expensive and gives a good power reduction.
I am trying to find ways this thing is not on 6nm guys...
Last edited:
Bojji
Member
With something completely beyond RDNA I can see this. AMD likes to ride their architectures for long, this was true with GCN but with GCN there were IPC gains with some new gens. So far between RDNA1-3 IPC looks to be the same pretty much.
With something completely new it will be a different story obviously.
7700XT alone takes around 230W, I think with rest of components we will see around 300W maximum with this console.
I don't think nod changes do... anything at this point, it's mostly marketing. I don't think jump to N5 will do much unless they really improved power efficiency between RDNA 3 and 4.
Loxus
Member
Sony could do the same thing in this video and run the GPU in power savings and run it at lower clocks.
Cap GPU clocks at 2450MHz and limit power to 170W.
Last edited:
Bojji
Member
Sony could do the same thing in this video and run the GPU in power savings and run it at lower clocks.
Cap GPU clocks at 2450MHz and limit power to 170W.
But you can't undervolt console GPU when there are millions of units (requires fine tuning for each one). This console will run at lower clocks than 7700xt but it's also rumored to have more cores so I doubt it will take less power overall
This test is in 1080p, 4k is much harder on power draw.
Imtjnotu
Member
it can be cheaper all it wants but if Chip Fabs dont have space to supply an order at scale it doesnt matter.N5.
It's not too expensive and gives a good power reduction.
sony will go with whom ever has room either at TSMC or Samsung because they will be having to make 2 separate dies now just like with the Pro before
Mr.Phoenix
Member
I think people always fall into this trap of looking at the PS5 as if it were just a run-of-the-mill PC. Maybe its because we are more familiar with systems like that. Or maybe its because since we arent developers, we tend to only draw conclusions on hardware based on hardware we are familiar with as opposed to look at the hardware and understand what and how it can be used.
Anyways, you have to take something into consideration when thinking about your bandwidth stuff. The PS5pro is not trying to increase rez and push higher framerates per se. In an ideal scenario, the PS5 is going to be natively rendering fewer pixels than the base PS5fidelity mode, and using PSSR to get that to 60fps. And have you seen what GPUs with as little as 44GB/s and or 512GB/s of bandwidth are doing? The PS5pro is not ever going to need to push past 120fps at its lowest sup[ported rez mind you, to reach a point where its bandwidth limited.
If I told you back in 2014 when the best of the best GPU was clocked at 1500Mhz, that 10 years on we would have GPUs clocked at 2.7Ghz.
I doubt you would have believed it either.
Last edited:
Gaiff
SBI’s Resident Gaslighter
The Titan in 2014 and 980 Ti in 2015 were clocked at like 1000-1100MHz. We were still far from 1500MHz in 2014 lol.
Maxwell and Kepler had insane OC potential though.
MisterXDTV
Member
That's a good point!
But an APU has clear power limits, so the efficiency for a 3 Ghz+ clock has to be incredible
Last edited:
What's incredible is not hitting 3 GHz, but rather hitting 3 GHz with 40 CUs at ~100W (taking into account the CPU).
It would imply the possibility of desktop GPUs hitting 3.5 GHz or more.
Radical_3d
Member
John to the rescue again. Everyone else:
MasterCornholio
Member
winjer
Gold Member
There is nothing special there, expect one sentence from John, where he states that a few devs told him that they are ok with the CPU performance.
But that is not very different from what several users said here, that most often than not, the PS5 is bottlenecked by the GPU, not by the CPU.
Skifi28
Member
Wasn't it the same for the PS5?
Radical_3d
Member
Yeah but they are drawing all the conclusions. So they’ll be right no matter what. The problem is that the original episode didn’t have John to moderate the message. Now we have the Leonard Nimoy cycle, PS5 Pro CPU version:
• The CPU is not enough.
• The CPU is enough.
• The CPU is also saving PC gaming.
Lysandros
Member
I personaly think PS5 APU is most often limited by its shared memory bandwidth which (in turn) limits GPU's raw potential (be it fill rate and/or compute etc.) to be fully taken advantage of. Same for XSX. This is why i would like to see a higher focus on bandwidth to increase the real world per component throughput rather than a raw metric increase with plenty of potential left on the table for the next systems.
winjer
Gold Member
Up to a certain point, yes. The lack of memory bandwidth on consoles is recurring and it's the main reason why consoles use low levels of Anisotropic filtering or none at all.
But the PS5 has much better optimizations for dealing with memory bandwidth, than the PS4, for example.
Not only it uses a tile based renderer, which significantly reduces memory accesses. But it also has delta color compression.
Bojji
Member
This is quite good test, underclocked 7800XT pretty much is PS5 Pro (aside RT performance).
rickybooby87
Member
John to the rescue again. Everyone else:
John is gonna get fired if he ain't careful.
solidus12
Member
Similar threads
- Locked