Topher
Gold Member
Just a guess on my part, but I'm thinking Sony forked FSR into their own solution.
-
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
Just a guess on my part, but I'm thinking Sony forked FSR into their own solution.
Loxus
Member
Is this with PSSR?
Cause the 4070 is 60% faster and the 7800xt is 72% faster than AMD Radeon RX 6700, which is the PS5 equivalent.
The leak states the PS5 Pro is only 45% faster than the PS5, which lands it around the 7700XT.
In terms of RT.
6650XT is around PS5 level, which has 12fps in this test.
12 × 3 = 36 fps, which lands it around 3070TI like I've been saying.
I didn't use the 4× RT cause that seems like a best case scenario and not really a average in fps.
Doesn't FSR and DLSS both incorporate motion vectors? The biggest difference between Nvidia's and AMDs form of super resolution is the former is accelerated with machine learning will the latter isn't on top of whatever algorithm each company uses achieve a perceived higher resolution.
farmerboy
Member
Thing is, I feel like we still haven't seen what they can do with the ps5.
I haven't had my "wow" moment yet.
Tqaulity
Member
Preach brother! Why people continue to hold onto TF numbers is beyond me when it's proven so many times to be a faulty way to compare power. Let's add another data point relevant to this discussion:
If you look at the 7700XT and the 7800XT on paper, you'd think they are pretty close in performance based on the TF ratings. The 7700XT = 35TFLOPS while the 7800XT = 37.2TFLOPs. That's a difference of <10% right. Yet what's the real world in game performance delta between those 2 card? Generally between 15-25% in favor of the 7800XT!! TechpowerUP averages this out to +21% in their relative performance graph in favor of the 7800XT.
So the question is why? Because there is more to a GPU than just compute right!? 7800 XT has 6 more Compute Units, a wider memory bus, 2x larger L2 cache, and a +44% memory bandwidth advantage. Yes that stuff matters a lot!
So can PS5 Pro be faster than 4070? Absolutely and the people that scoff at that prospect clearly don't understand the data they are receiving. Think of it this way, the 7800XT is consistently faster than a 4070 and the PS5 Pro GPU config is a much closer match to the 7800XT than the 7700XT.
So everyone's still latching onto those leaked specs and that 45% "rendering performance" figure which has had multiple reasons presented as to why that figure doesn't make any sense. I really wouldn't hold onto that figure as a meaningful metric of the PS5 Pro's relative GPU performance for a number of reasons. Yes the +45% does roughly land at the 7700XT in perf from the current PS5. BUT, based on the (skeptical) leaked info on the PS5 Pro's GPU, it should easily beat up the 7700XT for several obvious reasons:
- CU Count: 7700XT has 54 CUs while the (alleged) PS5 Pro GPU has 60 CUs (difference of >10%)
- Memory Bus: 7700XT only has a bus width of 192-bit while the (alleged) PS5 Pro has a 256-bit bus (see what a difference that makes in the 7700XT to 7800XT comparison above)
- Memory Bandwidth: 7700XT only has 432GB/s while the (alleged) PS5 Pro has 576GB/s (difference of +33% in favor of the Pro)
- Cache size: both the L1 and L2 caches are 2x larger on the (alleged) PS5 Pro
- Architecture Improvements: PS5 Pro (allegedly) is based on a RDNA 3.5 or RDNA 4 architecture which will have some efficiency and performance advantages over the RDNA 3 arch in the 7700XT
Again, pay attention to the (conflicting) data being presented. There simply is no logical outcome where a GPU configured as the rumored PS5 Pro is a "match" for a 7700XT (unless they severely underclock it lower than the base PS5). That simply does NOT compute (+45% rendering performance be damned). Either that 45% figure is wrong or the rumored specs are wrong. We shall see
Last edited:
ChiefDada
Gold Member
Preach brother! Why people continue to hold onto TF numbers is beyond me when it's proven so many times to be a faulty way to compare power. Let's add another data point relevant to this discussion:
If you look at the 7700XT and the 7800XT on paper, you'd think they are pretty close in performance based on the TF ratings. The 7700XT = 35TFLOPS while the 7800XT = 37.2TFLOPs. That's a difference of <10% right. Yet what's the real world in game performance delta between those 2 card? Generally between 15-25% in favor of the 7800XT!! TechpowerUP averages this out to +21% in their relative performance graph in favor of the 7800XT.
So the question is why? Because there is more to a GPU than just compute right!? 7800 XT has 6 more Compute Units, a wider memory bus, 2x larger L2 cache, and a +44% memory bandwidth advantage. Yes that stuff matters a lot!
So can PS5 Pro be faster than 4070? Absolutely and the people that scoff at that prospect clearly don't understand the data they are receiving. Think of it this way, the 78700XT is consistently faster than a 4070 and the PS5 Pro GPU config is a much closer match to the 7800XT than the 7700XT.
So everyone's still latching onto those leaked specs and that 45% "rendering performance" figure which has had multiple reasons presented as to why that figure doesn't make any sense. I really wouldn't hold onto that figure as a meaningful metric of the PS5 Pro's relative GPU performance for a number of reasons. Yes the +45% does roughly land at the 7700XT in perf from the current PS5. BUT, based on the (skeptical) leaked info on the PS5 Pro's GPU, it should easily beat up the 7700XT for several obvious reasons:
With this in mind, the 7700XT literally has no advantage in HW spec except possibly clock speed since I surmise that the PS5 Pro clock speed is still TBD (tons of conflicting info here). But assuming no increase in clocks, the 7700XT has a roughly ~13% advantage is clocks which still wouldn't be enough to make up the differences I just outlined.
- CU Count: 7700XT has 54 CUs while the (alleged) PS5 Pro GPU has 60 CUs (difference of >10%)
- Memory Bus: 7700XT only has a bus width of 192-bit while the (alleged) PS5 Pro has a 256-bit bus (see what a difference that makes in the 7700XT to 7800XT comparison above)
- Memory Bandwidth: 7700XT only has 432GB/s while the (alleged) PS5 Pro has 576GB/s (difference of +33% in favor of the Pro)
- Cache size: both the L1 and L2 caches are 2x larger on the (alleged) PS5 Pro
- Architecture Improvements: PS5 Pro (allegedly) is based on a RDNA 3.5 or RDNA 4 architecture which will have some efficiency and performance advantages over the RDNA 3 arch in the 7700XT
Again, pay attention to the (conflicting) data being presented. There simply is no logical outcome where a GPU configured as the rumored PS5 Pro is a "match" for a 7700XT (unless they severally underclock it lower than the base PS5). That simply does NOT compute (+45 rendering performance be damned). Either that 45% figure is wrong or the rumored specs are wrong. We shall see
Excellent post. I don't wanna hear anything about the damn 7700XT again. Such an insulting and frankly idiotic comparison.
hinch7
Member
7700XT does have 48MB of L3 Cache (or Infinity Cache) which consoles don't have, that will negate some of the memory bandwidth difference. And the 7700XT is also likely to be clocked a few hundred mhz higher for game/boost clocks assuming the Pro uses the same clocks due to BC. Which should make each GPU comparative given both will have similar specs, with architectural improvements in question. And in theory a DGPU shouldn't be as power constrained as opposed to the APU in a tight package.Preach brother! Why people continue to hold onto TF numbers is beyond me when it's proven so many times to be a faulty way to compare power. Let's add another data point relevant to this discussion:
If you look at the 7700XT and the 7800XT on paper, you'd think they are pretty close in performance based on the TF ratings. The 7700XT = 35TFLOPS while the 7800XT = 37.2TFLOPs. That's a difference of <10% right. Yet what's the real world in game performance delta between those 2 card? Generally between 15-25% in favor of the 7800XT!! TechpowerUP averages this out to +21% in their relative performance graph in favor of the 7800XT.
So the question is why? Because there is more to a GPU than just compute right!? 7800 XT has 6 more Compute Units, a wider memory bus, 2x larger L2 cache, and a +44% memory bandwidth advantage. Yes that stuff matters a lot!
So can PS5 Pro be faster than 4070? Absolutely and the people that scoff at that prospect clearly don't understand the data they are receiving. Think of it this way, the 7800XT is consistently faster than a 4070 and the PS5 Pro GPU config is a much closer match to the 7800XT than the 7700XT.
So everyone's still latching onto those leaked specs and that 45% "rendering performance" figure which has had multiple reasons presented as to why that figure doesn't make any sense. I really wouldn't hold onto that figure as a meaningful metric of the PS5 Pro's relative GPU performance for a number of reasons. Yes the +45% does roughly land at the 7700XT in perf from the current PS5. BUT, based on the (skeptical) leaked info on the PS5 Pro's GPU, it should easily beat up the 7700XT for several obvious reasons:
With this in mind, the 7700XT literally has no advantage in HW spec except possibly clock speed since I surmise that the PS5 Pro clock speed is still TBD (tons of conflicting info here). But assuming no increase in clocks, the 7700XT has a roughly ~13% advantage is clocks which still wouldn't be enough to make up the differences I just outlined.
- CU Count: 7700XT has 54 CUs while the (alleged) PS5 Pro GPU has 60 CUs (difference of >10%)
- Memory Bus: 7700XT only has a bus width of 192-bit while the (alleged) PS5 Pro has a 256-bit bus (see what a difference that makes in the 7700XT to 7800XT comparison above)
- Memory Bandwidth: 7700XT only has 432GB/s while the (alleged) PS5 Pro has 576GB/s (difference of +33% in favor of the Pro)
- Cache size: both the L1 and L2 caches are 2x larger on the (alleged) PS5 Pro
- Architecture Improvements: PS5 Pro (allegedly) is based on a RDNA 3.5 or RDNA 4 architecture which will have some efficiency and performance advantages over the RDNA 3 arch in the 7700XT
Again, pay attention to the (conflicting) data being presented. There simply is no logical outcome where a GPU configured as the rumored PS5 Pro is a "match" for a 7700XT (unless they severally underclock it lower than the base PS5). That simply does NOT compute (+45 rendering performance be damned). Either that 45% figure is wrong or the rumored specs are wrong. We shall see
The PS5 Pro's GPU will likely be quite a bit faster in RT though and perhaps more in line with current gen Nvidias mid range.
Last edited:
dano1
A Sheep
PS5 pro is for the hardcore fans like me and a bunch of others on this site. Keeping pace with PC.
saintjules
Gold Member
Preach brother! Why people continue to hold onto TF numbers is beyond me when it's proven so many times to be a faulty way to compare power. Let's add another data point relevant to this discussion:
If you look at the 7700XT and the 7800XT on paper, you'd think they are pretty close in performance based on the TF ratings. The 7700XT = 35TFLOPS while the 7800XT = 37.2TFLOPs. That's a difference of <10% right. Yet what's the real world in game performance delta between those 2 card? Generally between 15-25% in favor of the 7800XT!! TechpowerUP averages this out to +21% in their relative performance graph in favor of the 7800XT.
So the question is why? Because there is more to a GPU than just compute right!? 7800 XT has 6 more Compute Units, a wider memory bus, 2x larger L2 cache, and a +44% memory bandwidth advantage. Yes that stuff matters a lot!
So can PS5 Pro be faster than 4070? Absolutely and the people that scoff at that prospect clearly don't understand the data they are receiving. Think of it this way, the 7800XT is consistently faster than a 4070 and the PS5 Pro GPU config is a much closer match to the 7800XT than the 7700XT.
So everyone's still latching onto those leaked specs and that 45% "rendering performance" figure which has had multiple reasons presented as to why that figure doesn't make any sense. I really wouldn't hold onto that figure as a meaningful metric of the PS5 Pro's relative GPU performance for a number of reasons. Yes the +45% does roughly land at the 7700XT in perf from the current PS5. BUT, based on the (skeptical) leaked info on the PS5 Pro's GPU, it should easily beat up the 7700XT for several obvious reasons:
With this in mind, the 7700XT literally has no advantage in HW spec except possibly clock speed since I surmise that the PS5 Pro clock speed is still TBD (tons of conflicting info here). But assuming no increase in clocks, the 7700XT has a roughly ~13% advantage is clocks which still wouldn't be enough to make up the differences I just outlined.
- CU Count: 7700XT has 54 CUs while the (alleged) PS5 Pro GPU has 60 CUs (difference of >10%)
- Memory Bus: 7700XT only has a bus width of 192-bit while the (alleged) PS5 Pro has a 256-bit bus (see what a difference that makes in the 7700XT to 7800XT comparison above)
- Memory Bandwidth: 7700XT only has 432GB/s while the (alleged) PS5 Pro has 576GB/s (difference of +33% in favor of the Pro)
- Cache size: both the L1 and L2 caches are 2x larger on the (alleged) PS5 Pro
- Architecture Improvements: PS5 Pro (allegedly) is based on a RDNA 3.5 or RDNA 4 architecture which will have some efficiency and performance advantages over the RDNA 3 arch in the 7700XT
Again, pay attention to the (conflicting) data being presented. There simply is no logical outcome where a GPU configured as the rumored PS5 Pro is a "match" for a 7700XT (unless they severely underclock it lower than the base PS5). That simply does NOT compute (+45% rendering performance be damned). Either that 45% figure is wrong or the rumored specs are wrong. We shall see
I guess the question I ask you and everyone else - If you put in a game on a PS5 Pro are you going to think about any of these specs while actually playing?
Whoa look at that RT. That 45% is really kicking things into gear.
Just playing around
Last edited:
bitbydeath
Member
Doesn’t RDNA 3 have 96MB L3 Cache?
Zathalus
Member
PSSR should be as easy to implement as like DLSS it is a temporal scaler. So if the game has TAA it’s so easy a indie dev can implement it in minutes (DLSS that is). I’d imagine most AAA games would have it included then outside of the technically incompetent ones like FROM.
marquimvfs
Member
Will happily eat crow if a pro ever get an official announcement. But I seriously doubt a PS5 pro will ever exist.
vivftp
Member
Good for you Skippy
The rest of us in the real world will enjoy seeing the PS5 Pro announcement this September
marquimvfs
Member
Well, it's not like that I won't enjoy the existence of it, or like I'm wanting for it to not exist. I just don't believe any of the rumours and I think that the mere reference of it is just a hunt for clicks.
Last edited:
vivftp
Member
It's happening. The time for doubt was about a year ago. We are well into practically confirmed territory now
bitbydeath
Member
Navi 31 is 96MB
AMD RDNA 3 "Navi 31" GPU Block Diagram Detailed: 1st Chiplet Gaming GPU, 54% Perf/Watt, Larger L0/L1/L2 Cache, 80% Better Ray Tracing
The block diagram of AMD's RDNA 3 "Navi 31" GPU has been leaked, showcasing the world's first and most advanced chiplet design for gamers.
Question is which will be in the Pro.
It’ll be at least RDNA 3 if not 4.
Last edited:
PaintTinJr
Member
With the talk of infinity cache sizes, I would wonder if the Pro might have 32GBs of GDDR6, given that the IO complex having priority levels could probably - with double the esram size - could probably transparently present more RAM as an infinity cache/LLC, and the benefit of using more RAM would ease in next-gen RAM uplift IMO
The ninja farmer
Member
I don’t think thats on Sony agenda this time around. If the leaked doc is to be believed the low level caches double in size and the current high level cache stays the same
I do agree with you though. I would think that Sony could gain some serious performance benefits from using an infinity cache solution, or something similar but the current cost ceiling, or power/silicon budget just doesn’t allow it (it surely can’t be for back compat reasons)
But I guess that will have to keep for the ps6 or next Xbox generation.
PaintTinJr
Member
My main thoughts for using more RAM as a cheap LLC for the Pro is that RT is a memory hog, as is 4K, and a quality neural network for PSSR could benefit from more RAM in addition to needing scratch space for preparing data/matrices for alignment to fully exploit the Dual Issue compute and to store intermediate reusable computations, and because increasing memory size in the last mid-gen refresh worked better for X1X than the PS4 Pro keeping the RAM size the same as the base PS4
Loxus
Member
The leak does not specify if the L1 cache is Shader Array or Shader Engine.Preach brother! Why people continue to hold onto TF numbers is beyond me when it's proven so many times to be a faulty way to compare power. Let's add another data point relevant to this discussion:
If you look at the 7700XT and the 7800XT on paper, you'd think they are pretty close in performance based on the TF ratings. The 7700XT = 35TFLOPS while the 7800XT = 37.2TFLOPs. That's a difference of <10% right. Yet what's the real world in game performance delta between those 2 card? Generally between 15-25% in favor of the 7800XT!! TechpowerUP averages this out to +21% in their relative performance graph in favor of the 7800XT.
So the question is why? Because there is more to a GPU than just compute right!? 7800 XT has 6 more Compute Units, a wider memory bus, 2x larger L2 cache, and a +44% memory bandwidth advantage. Yes that stuff matters a lot!
So can PS5 Pro be faster than 4070? Absolutely and the people that scoff at that prospect clearly don't understand the data they are receiving. Think of it this way, the 7800XT is consistently faster than a 4070 and the PS5 Pro GPU config is a much closer match to the 7800XT than the 7700XT.
So everyone's still latching onto those leaked specs and that 45% "rendering performance" figure which has had multiple reasons presented as to why that figure doesn't make any sense. I really wouldn't hold onto that figure as a meaningful metric of the PS5 Pro's relative GPU performance for a number of reasons. Yes the +45% does roughly land at the 7700XT in perf from the current PS5. BUT, based on the (skeptical) leaked info on the PS5 Pro's GPU, it should easily beat up the 7700XT for several obvious reasons:
With this in mind, the 7700XT literally has no advantage in HW spec except possibly clock speed since I surmise that the PS5 Pro clock speed is still TBD (tons of conflicting info here). But assuming no increase in clocks, the 7700XT has a roughly ~13% advantage is clocks which still wouldn't be enough to make up the differences I just outlined.
- CU Count: 7700XT has 54 CUs while the (alleged) PS5 Pro GPU has 60 CUs (difference of >10%)
- Memory Bus: 7700XT only has a bus width of 192-bit while the (alleged) PS5 Pro has a 256-bit bus (see what a difference that makes in the 7700XT to 7800XT comparison above)
- Memory Bandwidth: 7700XT only has 432GB/s while the (alleged) PS5 Pro has 576GB/s (difference of +33% in favor of the Pro)
- Cache size: both the L1 and L2 caches are 2x larger on the (alleged) PS5 Pro
- Architecture Improvements: PS5 Pro (allegedly) is based on a RDNA 3.5 or RDNA 4 architecture which will have some efficiency and performance advantages over the RDNA 3 arch in the 7700XT
Again, pay attention to the (conflicting) data being presented. There simply is no logical outcome where a GPU configured as the rumored PS5 Pro is a "match" for a 7700XT (unless they severely underclock it lower than the base PS5). That simply does NOT compute (+45% rendering performance be damned). Either that 45% figure is wrong or the rumored specs are wrong. We shall see
That L1 cache is also from the same leak document.
Kepler_L2 actually knows the PS5 Pro specs.
And even Heisenberg acknowledgehttps this.
All of you all are a bunch of hypocrites.
Y'all pick and choose which part of the leak fits for the day. And run with it ignoring other parts.
It's so ridiculous.
I literally explained Kepler_L2 reasoning for comparing the 6700(36CUs) and 7700XT to the PS5 and PS5 Pro using the same good damn TechPowerUP you did.
Not only that, when I said the leak document is full of inconsistencies, and explain everything with sources, I got backlash and lols, so I decided to go along with the leak.
You do the same but without using sources and get tons of positive reactions???
I be putting work following all AMD leaks and putting pieces of the puzzle together.
I even scaled Zen and RDNA in AutoCAD to see if the leaks can work on a chip awhile ago.
This is some BS guys.
Zathalus
Member
The Pro does have more RAM available for use, base PS5 allows for 12.5GB to used by games while the Pro allows for 13.7GB. PSSR utilises around 200MB so that leaves an extra 1GB for the Pro to use for RT or whatever else the developer wants to.
PaintTinJr
Member
But that hardly feels like a flex for competing with mid-high PC as is the Pro's intention. A high end PC GPU will have at least 16GB of VRAM, if not 24GB, along with at least 32GB RAM, so sticking with 16GB seems like a worse comparison than the original base PS5 vs PC of 3years ago.
Zathalus
Member
I fully agree, but I think the realities of cost reared its head. Personally I would happily shell out a hundred or two hundred dollars more for higher specifications and VRAM, but alas I think I’m probably in the minority compared to the wider gaming audience.
The reason to have a LLC is to avoid the latency penalty of going out to main memory in the first place. Simply increasing the amount of RAM doesn't do anything to solve this.
Edit: And also of course to have access to more bandwidth than the memory bus can offer.
Last edited:
hinch7
Member
Consoles don't use off the shelf parts and will again be a custom RDNA GPU. There won't be any L3 Cache in the Pro. For varying reasons.Navi 31 is 96MB
AMD RDNA 3 "Navi 31" GPU Block Diagram Detailed: 1st Chiplet Gaming GPU, 54% Perf/Watt, Larger L0/L1/L2 Cache, 80% Better Ray Tracing
The block diagram of AMD's RDNA 3 "Navi 31" GPU has been leaked, showcasing the world's first and most advanced chiplet design for gamers.wccftech.com
Question is which will be in the Pro.
It’ll be at least RDNA 3 if not 4.
bitbydeath
Member
Usually there’s a base level and the customisations are added on top right?
Your theory would work if the base wasn’t RDNA3. What makes you think the L3 cache wouldn’t be up to spec?
YeulEmeralda
Linux User
PSSR catching up with DLSS? I never get the vibe that Nvidia is as stupidly complacent as Intel was.
TrebleShot
Member
Why are you all comparing off the shelf GPUs with custom architecture, you clowns never learn.
OverHeat
« generous god »
Custom piece of shit APU
PaintTinJr
Member
What you are saying in general is certainly true - although L3 cache is more southbridge latency IMO - but specifically within the context of Infinity fabric, Infinity cache and an IO complex with more ESRAM or a 2nd one connected exclusively to RAM, the latency would still be hidden by the priority mechanism via the low latency ESRAM, providing what should still be a very high performance LLC. It should be equal or close enough to Infinity cache or worst case match regular infinity fabric interconnection.
Last edited:
HeisenbergFX4
Gold Member
Some who have talked about the Pro don’t even have any sites or Social Media that they are trying to drive people to
And a few of them know more than what’s been leaked so far but can’t fly too close to the sun as a couple of those people have been told to stop talking about some stuff
Gaiff
SBI’s Resident Gaslighter
I don’t think there’s any intention on the Pro’s part to compete with high-end PCs. If it’s like the PS4 Pro, it’s there to keep PlayStation customers in the ecosystem. Most people buy low to mid-range parts and in 2025, you’ll be able to build a low budget option that easily beats the PS5 which might sound enticing for some.
High-end is completely out of reach. Sony most definitely isn’t aiming to compete against the 5080s that the PS5 Pro will be up against soon. A 5060 though? Why not? Furthermore, the high-end market is tiny so no need to focus on it. If one is hellbent on building a high-end rig, no console will satisfy their needs.
Anchovie123
Member
I'm betting not even the PS6 will have 32GB of VRAM
it'll have 24 with 20 or 18 available for devs. Witch will be enough imo. Barely any game utilizes 16gb or more, even Cyberpunk uses around 16GB on RT overdrive @ native 4k.Skill issue anyways, devs need to better utilize SSD streaming.
PaintTinJr
Member
Mid-High is all the bells and whistles, highest quality texturing and outputting at pristine 4K HDR - the maximum available on TV - and based on the 4070+ predictions by then a 5060 using DLSS will be Mid-High unless Nvidia locked it at 12GBs VRAM.
Comparing to the RX 5700XT that PS4 Pro was compared to and arguable out did when PS5 cross-gen hit - with the relative VRAM memory comparisons of ~6GB vs 8GB- if the PS5 Pro doesn't get a memory uplift then the gulf between PC VRAM (24GB) at that tier and the PS5 Pro VRAM (still base ~12GB) would appear to be getting much bigger IMO.
Gaiff
SBI’s Resident Gaslighter
Next-gen top-tier PC cards are probably staying at 24 GB anyway. Sony doesn't care about high-tier PCs with 24 GB of VRAM. They want to slot the PS5 Pro into the mid-tier where GPUs will have 12-16 GB. There really isn't much they can do about the VRAM situation without altering the GPU design. 32 GB is way too much and while 16 GB might appear to be stretching thin down the line, I think it's still something devs can work with and deliver good-looking games with high-quality texture work at relatively high resolutions too. In a perfect world, they'd go with 20 GB but meh, I ain't getting bent out of shape for this.
PaintTinJr
Member
Against the criteria I set for high, a 4070+ PSSR PS5 Pro will need to tick all those boxes, but as you said 16GB is a bit thin, and despite base PS5 having texturing that is capable of PC high-end+ (which should massively play into the Pro's PSSR vs DLSS comparison and need PC cards to have 24GBs) it isn't going to leave much VRAM for using the Dual issue and uprated RT of the Pro IMO, and so I'm going to be surprised if they don't go for more RAM in the Pro, as it will give off vibes of the pre-release PS4 4GB specs IMO, and even if they do need to re-engineer the memory controller for a Pro, I don't think they are as constrained by forward looking design in the Pro as they are with decisions made with base consoles.
Last edited:
Loxus
Member
18GB can work as well.
Imagine that 18 was actually memory size and not speed.
288-bit bus
16Gbps speed
576 GB/s bandwidth
Or better yet
18GB
288-bit bus
18Gbps
648 GB/s
Mr.Phoenix
Member
The smart thing for someone like you to do is at least wait till September to say what you just said here.
just look at what happened with PS4pro. Rumors started first, then a leaked document popped up around May-June, then the announcement occurred in September. The PS5pro is two for three on that so far.
If PC games has shown us anything, its that the PS5pro barely even needs 16GB of RAM, especially when you consider how the PS5 uses its own RAM and what framerates and rez it generally runs games at. The majority of games on PC... and I am only looking at the games built on modern game engines and using all of the new modern features (4K, RT and even FG times), all use between 10GB - 15GB (in extreme cases) of VRAM). I am sure on consoles they would use significantly less, especially when you consider the difference in how consoles and PC uses RAM.18GB can work as well.
Imagine that 18 was actually memory size and not speed.
288-bit bus
16Gbps speed
576 GB/s bandwidth
Or better yet
18GB
288-bit bus
18Gbps
648 GB/s
Loxus
Member
That's definitely the case.
I was wondering how the PS5 Pro allocated more RAM for games.
It seems like Cerny already maximize RAM usage on the PS5.
The PS5 Pro is said to have more RAM available for games than the PS5.
I don't think we can get 1GB+ from reducing OS footprint.
Giving the PS5 Pro 18GB seems like a good choice, but I guess we'll know soon enough.
Last edited:
ChiefDada
Gold Member
PS5 can already have 50% advantage over 6700 in games with 4k resolution and/or RT. See DF video below. Two main reasons for Pro existence is better RT and higher resolution output. You buy the Pro to enjoy the bandwidth sensitive graphical features at higher framerate. Therefore, memory setup is far more important than TF here. For the type of workloads PS5 Pro is designed to run, the 7700xt might as well be a 10 tf card beside the Pro. Great example of how benchmark avg can often be useless and misleading.
winjer
Gold Member
Maybe some sort of data compression.
That is a terrible test. That board is limiting the GPU to 4 lanes of PCIe.
So of course it's going to underperform, compared to the PS5.
Remember that communication between CPU and GPU, on the PC, is done via the PCIe bus. And some games transfer a lot of data there.
That is a terrible test. That board is limiting the GPU to 4 lanes of PCIe.
So of course it's going to underperform, compared to the PS5.
Remember that communication between CPU and GPU, on the PC, is done via the PCIe bus. And some games transfer a lot of data there.
Mahavastu
Member
IIRC the PS4 pro had a little bit of slow DDR Ram, which the OS could use for buffering stuff (eg. media apps), so the games had a little bit more of the fast memory.
I guess they will do something similar for the PS5 pro, add some cheaper slow RAM for the OS and such freeing up some memory for the games.
Last edited:
Loxus
Member
This is what that 1GB was used for.
The nitty-gritty details of the PS4 Pro, according to system architect Mark Cerny / More power, at what cost?
An additional 1GB of convention DRAM was put in the Pro to help it switch applications, so software like Netflix will be stored on the DRAM when open but not in use.
Looking at the "stored on the DRAM when open but not in use" part. You really think this applies to the PS5 Pro with the way the SSD works?
Last edited:
SlimySnake
Flashless at the Golden Globes
If Sony themselves are saying 45% better performance, im inclined to believe them.
But they are also saying 2-4x better RT performance so again, im inclined to believe that too.
I think in non-RT games, we will see 45% better performance, and in RT games we will see 80-100% better performance which will put it on par with the 3080 and 4070.
But they are also saying 2-4x better RT performance so again, im inclined to believe that too.
I think in non-RT games, we will see 45% better performance, and in RT games we will see 80-100% better performance which will put it on par with the 3080 and 4070.
You can't make those extrapolations when at the same time you claim that both systems use memory in very different ways.
Of course Ps5 doesn't need 32GB of RAM like a high en PC, but that doesn't mean 16GB fall short for some games.
They only have 12.5GB dedicated for games and the I/O dedicated hardware to maximize the utilisation. And without a doubt I can say that for data management Ps5 memory architecture is much more advanced than the PC, but as we are seeing since the beginning of this generation, very few developers are taking advantage of the system's potential to improve memory usage.
At the end that hardware advantage on Ps5 don't make the difference because now most devs only target the console as a low/mid tier PC. They only care to port PC code to the high level API Sony gives them, and then make some sacrifices on visual quality of effects to fit in the frame budget, or not get out of RAM and saturate the memory bus.
We still see some studios making dedicated optimization on Ps5 like Massive Entertainment did on Avatar FoP, but even with that you can see how the game engine is suffering to load some ground textures and how some assets don't uses the proper LOD level for the distance because they are running out of memory. The same issue happens with Star wars JS on quality mode on the big open areas of Koboh and some others. And there are a buch of games that are using lower draw distances and effects quality on consoles not because the CPU or the GPU can't hadle it, it is just because they need to fit into the 12,5GB of RAM.
Loxus
Member
Not only that.
Taking into account the entire video, the 6700 is basically the PS5 GPU equivalent.
The average performance is about the same.
TheRedRiders
Member
I'm curious if that 45% figure includes or excludes PSSR.
Mr.Phoenix
Member
The only guess(es) I could arrive at is that it must be something similar to what was done in the PS4pro and adding slower DRAM. I am sure there would be a good chunk of that OS RAM that just stores a lot of data... eg, the fact that with the PS5, the PS store is a part of the OS and not just a separate app as it was in the PS4. That is a whole LOT of cached data right there. Or the RAM that must be reserved for the constant background record feature with the PS5.That's definitely the case.
I was wondering how the PS5 Pro allocated more RAM for games.
It seems like Cerny already maximize RAM usage on the PS5.
The PS5 Pro is said to have more RAM available for games than the PS5.
I don't think we can get 1GB+ from reducing OS footprint.
Giving the PS5 Pro 18GB seems like a good choice, but I guess we'll know soon enough.
I can easily see how adding in 1-2GB of LPDDR RAM could be necessary and would in turn free up some of the precious GDDR6 RAM. And if its anything like with the PS4pro, that's not something you would see on any spec/data sheet. Unless Cerny comes out and says it himself. They have done it before, no reason they cant do it again.
Another way they could be doing this, is that they simply have more parts of the OS moved over to the SSD instead of sitting in RAM. They have enough use data over the last three years now to better optimize their OS with regards to paging using the SSD as a buffer for system RAM now than they would have had at the time the console launched. Think of it as a quick resume but for large chunks of the OS.
Bojji
Member
7700XT is the closest PC part to PS5 Pro. There is no better comparison when it comes to raster performance.
The leak does not specify if the L1 cache is Shader Array or Shader Engine.
That L1 cache is also from the same leak document.
Kepler_L2 actually knows the PS5 Pro specs.
And even Heisenberg acknowledgehttps this.
All of you all are a bunch of hypocrites.
Y'all pick and choose which part of the leak fits for the day. And run with it ignoring other parts.
It's so ridiculous.
I literally explained Kepler_L2 reasoning for comparing the 6700(36CUs) and 7700XT to the PS5 and PS5 Pro using the same good damn TechPowerUP you did.
Not only that, when I said the leak document is full of inconsistencies, and explain everything with sources, I got backlash and lols, so I decided to go along with the leak.
You do the same but without using sources and get tons of positive reactions???
I be putting work following all AMD leaks and putting pieces of the puzzle together.
I even scaled Zen and RDNA in AutoCAD to see if the leaks can work on a chip awhile ago.
This is some BS guys.
This is what I was talking about from the beginning. 6800/7700XT performance when in comes to raster.
Maybe close to 3080/4070 with some heavy RT but we will see...
Vast majority of games are raster only so first figure will be close to typical game performance.