• 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 devkits arrive at third-party studios, Sony expects Pro specs to leak

A $500 console, would likely have a BOM of $480-$520. Would be equivalent to a ~$800-$1000 PC. Or a PC gamer spending around $400 -$600 for a GPU upgrade. In such a situation, you can entice them to pick up a PS5pro instead.

A $700 Pro will likely have a BOM of $680-$750. Will be equivalent to ~$1300 -$1500 PC hardware. In this scenario, you are still selling to the people in the scenario above, as the people in this scenario who can afford to spend $1500 on a PC, wouldn't be interested in a PS5 for its specs. They would rather just spend on the PC.

Regardless of all that though, there really is no basis for Sony to make a $700 console. No matter what, there will be more people to sell to at $500 or even $600 than there would be to sell to at anything above that. We have seen how poorly a $600 console sells. I don't know where you are getting this impression that there are so many people willing to buy such a console.
A $700 pro would be equal to a $1,600-$1,700 pc your underselling and with how the gpu market is right now 700$ for a pro would be enticing
 

rnlval

Member
Are you being ridiculous? The shaders only have critical path blocking behaviour for their own compute shader, and only if written that way if not using acceleration. You have it back-to-front.

The blocking in vanilla RDNA2 like the PC cards and Series was shown in Xbox's technical breakdown of the series, and the blocking is caused by a CU being limited to either use a BVH accelerator or a texture unit at any one time, not both, and as shaders largely rely on texture samplers to get work done -and that capability is facilitated by the TMU - that is why it is either/or, BVH accelerator or TMU.
Prove PS5 RT beats RX 6700 XT.
 

MikeM

Member
Sure, but a 300mm at 4nm will be a lot beefier and more power at only $20 more. Again, the base PS5 will cover the marketshare play, the pro doesn't have to conform to that. They should not be afraid to price it at $200-$300 more than the base model.



The PS5 base model has to be cheap. It has to be affordable for most in order to gain marketshare. Sony receives around 30% or less for every game sold, let alone PS+ subscription. That's how the console market operates. In other words, using the most reasonably affordable node for the PS5 that would cut the budget for perhaps $450 BOM was chosen. Again, the PS5 needs to be affordable to gain market saturation.



You are correct and I agree. That's why it is unrealistic to expect the PS6 to use the latest and best chip process. Unless you want it to be priced at $600 - $800 and Sony start to lose marketshare. That would be stupid. But we are not talking about new gen yet. This is mid-gen.


That's now how it works. There is this thing called value for money. If it's simply about the price, then PS5 should have been as weak as Series S selling at $299. Remember, it's you who gave the $20 dollar increase in value from 5nm to 4nm.
$20 extra cost at 14million units (using PS4 Pro sales as an example) is $280million in additional costs. Obviously costs will likely go down over time, but volume is important to factor into these BOM discussions.
 

Mr.Phoenix

Member
A $700 pro would be equal to a $1,600-$1,700 pc your underselling and with how the gpu market is right now 700$ for a pro would be enticing
No, I am not. A 8800XT will likely cost $500/$600, based on the 7800XT pricing of $500 which you can check on pcpartspicker. The replacement card will not cost anymore than $600 though. A Ryzen 7 5700x (which is better than whatever CPU the PS5pro will have mind you) is $200. The remaining $400-$700 (depending on if your budget is $1300/$1500) is more than enough to cover a case, motherboard, storage, RAM, PSU, keyboard, mice, and even a controller. And this is assuming you are not even interested in the used market.

And a system like that will match and in most cases exceed the PS5pro.
 

rnlval

Member
Nothing on the hardware BVH transverse. Intersection engine processes geometry data.

7CcijiI.jpg


-----

Both Xbox Series X and PS5 consoles don't have PC RDNA 2's large infinity cache for storing the BVH working set.
w2ZLGnM.jpg



-----
AMD had obscured the following hardware problem when RDNA 2 GPUs were released.

OWcSHAZ.jpg


With RDNA 3's release, AMD bashed its own RDNA 2 products.
 
Last edited:

SonGoku

Member
There is a 34.5% RT improvement between similar 80-to-84 CU-scaled AMD GPUs.
If you normalize CUs and Clock frequencies.(6900XT is 23TF and the 7900XT is 25.7TF (discarding dual issue shaders) thats a 8% flop advantage for the 7900XT)
7900XT is 26.5% faster and 6900XT is 21% slower
The 7900XT also has 56% memory bandwidth advantage, i think at 4K the 6900XT infinity cache pool will be insufficient forcing it to access its main memory pool more often

So when factoring in that 56% bandwidth advantage, that 26.5% RT improvement is less impressive because it means some of it can be attributed to bandwidth bottleneck on the 6900XT running at 4K

Funnily enough if you compare 6800XT 20.7TF and the 7800XT (18.6TF) 7800XT 10% less theoretical flop performance
On Alan Wake 2 Path Tracing, 7800XT is 16% faster before factoring in its 10% flop deficit that would make the 7800XT 26% faster with normalized flop performance
Q59lL4n.png

Phantom Liberty RT 7800XT is ~10% faster before factoring in its 10% flop deficit that would make the 7800XT 20% faster with normalized flops
ocvcX8x.png
8g7UW19.png


Its not a perfect way to compare but it shows the alleged 1.5x RT improvement was an exaggeration
IF you take the comparative teraflops of the bottom two cards, the 7700XT is wasting roughly 42% of its higher TFLOP(FP32)
I believe the 7700XT is bottlenecked by bandwidth, 90% of the 7800XT CUs but only 70% (30% less) the memory bandwidth and 75% (25% less) the amount of infinity cache
If it only costs $20 more to manufacture in 4nm, then 5nm is a mistake.
Theres no 4nm, TSMC next node is 3nm
The only options are based on 5nm, so its either N5 or N4P which is a "a performance-focused enhancement of the 5-nanometer"
Nothing stops them from switching to 4nm a year or two later when its price has come down. Same way the 6nm process they switched to for the PS5 in 2022 was available back in 2020 when the console launched.
True... but if im remembering correctly N6 entered production in 2020 same year as PS5 launch
N7 entered production in 2018 two years before PS5 launch
N4P entered production in 2022 which would also make it 2 years before PS5 Pro launch giving AMD/Sony enough time to produce wafers on N4 for 2024, so while nothing is set in stone I think the probability is high that they will go with N4

Also TSMC claims 4NP lowers process complexity and improves wafer cycle time so that should make it cheaper
 
Last edited:

Mr.Phoenix

Member
$20 extra cost at 14million units (using PS4 Pro sales as an example) is $280million in additional costs. Obviously costs will likely go down over time, but volume is important to factor into these BOM discussions.
Exactly... but its not even 14M, thats not how Sony does deals for chips.

When sony is doing the deal for their chips on the next node they go to, its not going to be for just the PS5pro. Sony would look at a two-year window, and if they plan on shipping 40M consoles (PS5 + PS5pro) in those two years they do a deal with AMD for those 40M chips on say 5nm. Whatever they pay for that deal, can be as much as 20-25% more if they were to instead use 4nm.

That can be as much as $1B extra for what? Just to be on a 4nm node? For performance targets, they can still comfortably hit using 5nm and save themselves a billion dollars?
True... but if im remembering correctly N6 entered production in 2020 same year as PS5 launch
N7 entered production in 2018 two years before PS5 launch
N4P entered production in 2022 which would also make it 2 years before PS5 Pro launch giving AMD/Sony enough time to produce wafers on N4 for 2024, so while nothing is set in stone the probability is high that they will go with it imo.

Also TSMC claims 4NP lowers process complexity and improves wafer cycle time so that should make it cheaper
Yes, however, N5 was in production by the time the PS5 launched. Weird I know, but N5 came to market before N6.

From the beginning, I have said that whatever node Sony uses would come down to cost. If it ends up being cheaper for them to use N4? then that's what they will do. If you see N5 in the PS5pro and an internally revised PS5slim with an N5 chip, then it means it was cheaper.

To be fair, if they went with N4, it could amount to some cost savings elsewhere, like using a slightly smaller cooler, smaller chassis and smaller PSU. So its not all black and white.
 
Last edited:
I wish I could remember rn but I can tell you it had an official ps5-ready endorsement from Sony. It even kinda matched the console. Heatsink, yes. Iirc, Sony has some kind of stake in the company.
You got the sn850p it’s the one I got? I’m also looking at the Seagate game drive as an extra
 
No, I am not. A 8800XT will likely cost $500/$600, based on the 7800XT pricing of $500 which you can check on pcpartspicker. The replacement card will not cost anymore than $600 though. A Ryzen 7 5700x (which is better than whatever CPU the PS5pro will have mind you) is $200. The remaining $400-$700 (depending on if your budget is $1300/$1500) is more than enough to cover a case, motherboard, storage, RAM, PSU, keyboard, mice, and even a controller. And this is assuming you are not even interested in the used market.

And a system like that will match and in most cases exceed the PS5pro.
Your using the specs of a 500-600 pro not a 700$ one I don’t think you understand how much console hardware really opens up once they breach the $600 barrier by even 1% it gets an exponential increase
 

Mr.Phoenix

Member
Your using the specs of a 500-600 pro not a 700$ one I don’t think you understand how much console hardware really opens up once they breach the $600 barrier by even 1% it gets an exponential increase
If you look at the specs I listed... doesn't that tell you that there isn't even a $700 PS5 Sony can make? What? they should ditch APU and use a discrete CPU and put in a 7900 equivalent GPU? With 24GB of RAM, infinity cache and a chassis that looks like a PC?

The best sony can reasonably make, would cost them around $480-$550 to make. Once they go past that, they cant even make it for $700, they would need to spend more than that. Just take a look at what AMD has on the market, and what can comfortably do 1440p@ 65-75fps. And that should tell you what the PS5pro is going to be.
 
Last edited:

rnlval

Member
If you normalize CUs and Clock frequencies.(6900XT is 23TF and the 7900XT is 25.7TF (discarding dual issue shaders) thats a 8% flop advantage for the 7900XT)
7900XT is 26.5% faster and 6900XT is 21% slower
The 7900XT also has 56% memory bandwidth advantage, i think at 4K the 6900XT infinity cache pool will be insufficient forcing it to access its main memory pool more often

So when factoring in that 56% bandwidth advantage, that 26.5% RT improvement is less impressive because it means some of it can be attributed to bandwidth bottleneck on the 6900XT running at 4K

Funnily enough if you compare 6800XT 20.7TF and the 7800XT (18.6TF) 7800XT 10% less theoretical flop performance
On Alan Wake 2 Path Tracing, 7800XT is 16% faster before factoring in its 10% flop deficit that would make the 7800XT 26% faster with normalized flop performance
Q59lL4n.png

Phantom Liberty RT 7800XT is ~10% faster before factoring in its 10% flop deficit that would make the 7800XT 20% faster with normalized flops
ocvcX8x.png
8g7UW19.png


Its not a perfect way to compare but it shows the alleged 1.5x RT improvement was an exaggeration

I believe the 7700XT is bottlenecked by bandwidth, 90% of the 7800XT CUs but only 70% (30% less) the memory bandwidth and 75% (25% less) the amount of infinity cache

Theres no 4nm, TSMC next node is 3nm
The only options are based on 5nm, so its either N5 or N4P which is a "a performance-focused enhancement of the 5-nanometer"

True... but if im remembering correctly N6 entered production in 2020 same year as PS5 launch
N7 entered production in 2018 two years before PS5 launch
N4P entered production in 2022 which would also make it 2 years before PS5 Pro launch giving AMD/Sony enough time to produce wafers on N4 for 2024, so while nothing is set in stone I think the probability is high that they will go with N4

Also TSMC claims 4NP lowers process complexity and improves wafer cycle time so that should make it cheaper
AMD's up to 50% raytracing improvement claim is based on "per CU", not per clock cycle.

G1nTQmy.jpg

Higher sustained clock speed is part of the improvement.

This is why I used RX 7900 XT with 84 CU against RX 6900 XT with 80 CU.

I also ignored RDNA 3 CU's dual-issue mode.

The transversal workload has the new early cull subtree hardware.

----

AMD is using TSMC's 4 nm node for Phoenix APU. https://www.amd.com/en/product/13041 TSMC "4 nm FinFET". Hawk Point APU is fabricated on TSMC's 4 nm node. TSMC's 4 nm node is an improvement from 5 nm.
 
Last edited:

SonGoku

Member
Yes, however, N5 was in production by the time the PS5 launched. Weird I know, but N5 came to market before N6.

From the beginning, I have said that whatever node Sony uses would come down to cost. If it ends up being cheaper for them to use N4? then that's what they will do. If you see N5 in the PS5pro and an internally revised PS5slim with an N5 chip, then it means it was cheaper.

To be fair, if they went with N4, it could amount to some cost savings elsewhere, like using a slightly smaller cooler, smaller chassis and smaller PSU. So its not all black and white.
I agree it will come down to cost which is why im leaning towards N4 alas nothing is set in stone maybe a lot of customers will buy N4P and raise its cost 🤷‍♂️
The biggest improvement however has to come from AMD not the node, N4 is not that big of a improvement

If AMD somehow fixes RDNA3 bugs, improves power consumption, improves RT to makes RDNA4 (or RDNA3.5) as good as Ada Lovelace then it will be possible for the PS5 Pro to match a RTX 4070. On N4 and 2 years later they should be able to at least match Ada Lovelace right?
What? they should ditch APU and use a discrete CPU
For PS6 I hope this is the way, similar to meteor lake use different chiplets for CPU, AI, GPU, IO etc
That should make it possible to include a bigger GPU die and maybe next time they wont gimp the CPU so much
Just take a look at what AMD has on the market, and what can comfortably do 1440p@ 65-75fps.
Tbf GPUs are severely overpriced still even with the super lineup. Im sure they take a hefty profit
 

Gaiff

SBI’s Resident Gaslighter
If you look at the specs I listed... doesn't that tell you that there isn't even a $700 PS5 Sony can make? What? they should ditch APU and use a discrete CPU and put in a 7900 equivalent GPU? With 24GB of RAM, infinity cache and a chassis that looks like a PC?

The best sony can reasonably make, would cost them around $480-$550 to make. Once they go past that, they cant even make it for $700, they would need to spend more than that. Just take a look at what AMD has on the market, and what can comfortably do 1440p@ 65-75fps. And that should tell you what the PS5pro is going to be.
Good luck getting through his head that the limitations are the laws of physics and thermodynamics, not just budget. This kid thinks you could stuff a 4090 inside a PS5 Pro if Sony wanted to sell it for $1000, not considering that they would need a power budget far in excess of 300W to achieve this and unless Sony doesn't mind selling an enormous console with a huge external power brick and 3-4 fans, then that $1000 console cannot exist.

The PS5 Pro will still operate within roughly the same physical and power constraints as the regular PS5.
 

SonGoku

Member
AMD's up to 50% raytracing improvement claim is based on "per CU", not per clock cycle.
I see, you are right, misleading marketing though so most of the "1.5x" improvement comes from increased frequency not improvements to the RT hardware as shown on the stats and benchmarks i posted earlier
AMD is using TSMC's 4 nm node for Phoenix APU. https://www.amd.com/en/product/13041 TSMC "4 nm FinFET". Hawk Point APU is fabricated on TSMC's 4 nm node. TSMC's 4 nm node is an improvement from 5 nm.
its just a marketing name for the same 5nm node with enhancements https://investor.tsmc.com/static/annualReports/2022/english/ebook/files/basic-html/page96.html
TSMC 4nm FinFET=N4=5nm
TSMC 4nm FinFET Plus=N4P=5nm
 
Last edited:
If you look at the specs I listed... doesn't that tell you that there isn't even a $700 PS5 Sony can make? What? they should ditch APU and use a discrete CPU and put in a 7900 equivalent GPU? With 24GB of RAM, infinity cache and a chassis that looks like a PC?

The best sony can reasonably make, would cost them around $480-$550 to make. Once they go past that, they cant even make it for $700, they would need to spend more than that. Just take a look at what AMD has on the market, and what can comfortably do 1440p@ 65-75fps. And that should tell you what the PS5pro is going to be.
The 7900m is a thing of course I using zen 5 cores stacked with 3d cache maybe an increase to 20-24gb of memory and pcie gen 5 support. That’s the kind of pro I envision for $700
 

rnlval

Member
A $700 pro would be equal to a $1,600-$1,700 pc your underselling and with how the gpu market is right now 700$ for a pro would be enticing
NVIDIA can request a higher asking price when ADA has superior raytracing on an SM/CU and RT performance per chip area basis.
I see, you are right, misleading marketing though so most of the "1.5x" improvement comes from increased frequency not improvements to the RT hardware

its just a marketing name for the same 5nm node with enhancements https://investor.tsmc.com/static/annualReports/2022/english/ebook/files/basic-html/page96.html
TSMC 4nm FinFET=N4=5nm
TSMC 4nm FinFET Plus=N4P=5nm

TSMC's 4 nm improvement still has benefits over the older 5 nm. This is like 1st gen 7 nm vs 2nd gen 7 nm (7nm+) vs 6 nm (based on 7nm with further improvement i.e. 7nm++). AMD wouldn't reject process node improvements.


AMD has posted its latest financial results for Q3 2023, with the $5.8 billion 4% year-over-year growth attributed to Ryzen and "record server processor sales," according to AMD Chair and CEO Dr. Lisa Su.
The biggest growth sector for AMD was the Client segment (where most enthusiasts reside), with revenue up 42% year-over-year, driven by Ryzen mobile processor sales. AMD specifically highlights the success of its AMD Ryzen 7000 Series CPU in the mobile space as driving growth.


AMD is prioritizing the mobile PC CPU market. AMD's discrete GPUs, desktop Ryzens, and consoles are on older chip fabrication technology.
-----

AMD notes that the dip in its Gaming sector comes from the "semi-custom" sector, which includes consoles like the PlayStation 5 and Xbox Series X. So the good news (if we can call it that) is that the new Radeon RX 7700 XT and Radeon RX 7800 XT launches helped minimize the loss in revenue.
 
Last edited:

SonGoku

Member
The 60 CU 7800 XT only runs at 2.4 GHz while consuming 250 Watts and is already on the 5nm node.
I don't think its right to compare with RDNA3 as we know it, for PS5 Pro Sony will either use RDNA4 or a heavily improved RDNA 3.5 that fixes RDNA3 bugs, power consumption and increases RT performance to hopefully match a RTX 4070. I mean 2 years later and on a enhanced node AMD has to match 2022 nvidia, they cant be that far behind

TSMC's 4 nm improvement still has benefits over older 5 nm. This is like 1st gen 7 nm vs 2nd gen 7 nm (7nm+) vs 6 nm (based on 7nm with further improvement i.e. 7nm++). AMD wouldn't reject process node improvements.
Oh I know, which is why im theorizing Sony will use either N4 or N4P the later of which has +22 Power Efficiency +11 Power +6 density over N5
I was just clarifying N5,N4 and N4P are all part of the same 5nm node family no 4nm
 
Last edited:
Good luck getting through his head that the limitations are the laws of physics and thermodynamics, not just budget. This kid thinks you could stuff a 4090 inside a PS5 Pro if Sony wanted to sell it for $1000, not considering that they would need a power budget far in excess of 300W to achieve this and unless Sony doesn't mind selling an enormous console with a huge external power brick and 3-4 fans, then that $1000 console cannot exist.

The PS5 Pro will still operate within roughly the same physical and power constraints as the regular PS5.
No one was discussing the 4090. And they probably could make a 4090 equivalent at 1000 if it was on 3nm or the upcoming 2nm
 
I don't think its right to compare with RDNA3 as we know it, for PS5 Pro Sony will either use RDNA4 or a heavily improved RDNA 3.5 that fixes RDNA3 bugs, power consumption and increases RT performance to hopefully match a RTX 4070. I mean 2 years later and on a enhanced node AMD has to match 2022 nvidia, they cant be that far behind


Oh I know, which is why im theorizing Sony will use either N4 or N4P the later of which has +22 Power Efficiency +11 Power +6 density over N5
I was just clarifying N5,N4 and N4P are all part of the same 5nm node family no 4nm
You seem really in the know and quite patient I wanted to ask if when they release the pro they could release 2 versions of it one that a more affordable 500 and a 700 high end one?
 

SonGoku

Member
You seem really in the know and quite patient I wanted to ask if when they release the pro they could release 2 versions of it one that a more affordable 500 and a 700 high end one?
I don't think that would make financial sense and would fragment the userbase too much
I too would like a balls deep Pro at $700 but i think they figure the sweet spot for performance and mass adoption is somewhere around $500 to $600

Plus from the rumor mill RDNA4 will be midrange, what i speculate will happen is AMD will release a midrange RDNA4 GPU based on the work they did for the PS5 Pro
The 7900m is a thing of course I using zen 5 cores stacked with 3d cache maybe an increase to 20-24gb of memory and pcie gen 5 support. That’s the kind of pro I envision for $700
What i think Phoenix is getting at is that jumping to a bigger GPU the price increase might not be linear because for one you will lose the cost savings of APU and I dont know if AMD chiplet technology is mature enough to design a console with GPU and CPU chiplets. Cooling solution will also be more expensive, increased board complexity etc.

With chiplets maybe it would be possible to build such a machine for $700, but that is not ready for console maybe PS6
 
Last edited:

Mr.Phoenix

Member
I agree it will come down to cost which is why im leaning towards N4 alas nothing is set in stone maybe a lot of customers will buy N4P and raise its cost 🤷‍♂️
The biggest improvement however has to come from AMD not the node, N4 is not that big of a improvement

If AMD somehow fixes RDNA3 bugs, improves power consumption, improves RT to makes RDNA4 (or RDNA3.5) as good as Ada Lovelace then it will be possible for the PS5 Pro to match a RTX 4070. On N4 and 2 years later they should be able to at least match Ada Lovelace right?
I have said this too, in total agreement. I am more interested in what AMD actually does to address the gaping issues they have, than what node of how many TF the machine will have.

The 7900m is a thing of course I using zen 5 cores stacked with 3d cache maybe an increase to 20-24gb of memory and pcie gen 5 support. That’s the kind of pro I envision for $700
I am beginning to wonder if its that you just talk for the sake of talking... the 7900m, has a GPU die that on its own is 300mm2. That is like the budget of the entire PS5pro APU. By the time you add MCDs, it balloons up to 500mm2+. Good luck cooling a chip like that in something twice the size of the OG PS5 if you wanna run a chip that size at the OG PS5s GPU clock.

But say you wanna be conservative and stay true to the "m" monicker in its name and run it like a mobile GPU, then guess what? Its less powerful than the 7800XT.

So you are basically advocating that they spend more to get less. And what do you need PCIe5 support for? Are you 13 or something? The orders of magnitude IO advanced from last gen to this gen means that you ought to be looking at everything differently. PS5 is not even saturating PCIE4 SSDs which theoretically should peak at 8GB/s, but you want 12-14GB/s drives?

Anyway, I think I have tried my best to explain what we are likely to get and why. Dont know how to continue saying the same things to you without becoming disrespectful.

What i think Phoenix is getting at is that jumping to a bigger GPU the price increase might not be linear because for one you will lose the cost savings of APU and I dont know if AMD chiplet technology is mature enough to design a console with GPU and CPU chiplets. Cooling solution will also be more expensive, increased board complexity etc.

With chiplets maybe it would be possible to build such a machine for $700, but that is not ready for console maybe PS6
Exactly!!!

You go with a higher-spec GPU, and EVERYTHING goes up. Now you need more cache, and more bandwidth, which means you need more RAM, wider bus, bigger chip, higher TDP, more cooling, bigger PSU, bigger case...etc. He seems to think you can just throw in a bigger GPU and more powerful CPU and everything else remains the same running at 200W in a chassis the size of the OG PS5 cooled with one measly fan.
 
Last edited:

FireFly

Member
I don't think its right to compare with RDNA3 as we know it, for PS5 Pro Sony will either use RDNA4 or a heavily improved RDNA 3.5 that fixes RDNA3 bugs, power consumption and increases RT performance to hopefully match a RTX 4070. I mean 2 years later and on a enhanced node AMD has to match 2022 nvidia, they cant be that far behind
The 7800 XT is a 250W part though, so if the Pro budgets 180W for the GPU, that 12% clock speed up requires a 56% improvement in performance per watt. Which the post I was replying to was expecting even on 5nm, with no real information about the efficiency improvements in RDNA 3.5/4.
 

Mr.Phoenix

Member
The 7800 XT is a 250W part though, so if the Pro budgets 180W for the GPU, that 12% clock speed up requires a 56% improvement in performance per watt. Which the post I was replying to was expecting even on 5nm, with no real information about the efficiency improvements in RDNA 3.5/4.
First off, I don't even see the PS5pro using all 60CU which would make it a direct 7800XT equivalent expect it to use 54CU like the 7700XT. And second, while that is also a 250W card, I don't see the PS5pro running its clocks at 2400Mhz+ like that card can do. 2300MHz seems more likely. I also think the CPU in the PS5pro, clocked up to 4.4Ghz and all that, will be an under 30W CPU.
 

FireFly

Member
First off, I don't even see the PS5pro using all 60CU which would make it a direct 7800XT equivalent expect it to use 54CU like the 7700XT. And second, while that is also a 250W card, I don't see the PS5pro running its clocks at 2400Mhz+ like that card can do. 2300MHz seems more likely. I also think the CPU in the PS5pro, clocked up to 4.4Ghz and all that, will be an under 30W CPU.
Right but the person I was replying to was expecting 2.7 GHz on 5nm. I agree that circa PS5 clocks are more likely.
 

Perrott

Member
They wouldn't show it off in CES anyways. They just launched the PS5 refresh a few months ago and the Pro will (most likely) be coming in the Holidays.
Well, you never know.

Prior to 2019, they had never unveiled the existence of a next generation console through an interview with Wired out of all publications.
 

PaintTinJr

Member
...

I believe the 7700XT is bottlenecked by bandwidth, 90% of the 7800XT CUs but only 70% (30% less) the memory bandwidth and 75% (25% less) the amount of infinity cache

....
But the bandwidth per shader unit (StreamProcessor) actually increases on the 7700XT compared to the 6800XT - which in Alan Wake Raster and RT it has similar performance to.

In RT I don't think the bandwidth plays a big part because the low frame-rates provide far more bandwidth(on the PCIe bus) per frame than for high frame-rate Raster, and typically with deeper(more complex) compute like RT the compute to data ratio goes up, not down, so reading the same data recursively is more common. The vector units are twice as wide on RDNA3 to RDNA2, so if the work is batched optimally on both cards then half the reads and writes should occur on the RDNA3 memory, and with it far less latency for half the total memory operations. So I'm not even sure the inferior cache sizes on the 7700XT to 6800XT would be a big problem with optimised RT code.
 

PaintTinJr

Member
Prove PS5 RT beats RX 6700 XT.
How did you get to that pivot? Clearly not with a quote, as it is a complete strawman.

But either way, PS5's design being more efficient than vanilla RDNA2 in using TMUS and BVH intersection testing accelerators doesn't mean anything in a PS5 vs GPU faceoff, because the entire PS5 is using a laptop class CPU and operating on less power entirely than the GPU alone, a game on PS5 might also pick 8K or 4K texturing over 2K texturing with superior RT on the PC, so the efficiency in itself is only meaningful to PS5 devs utilising the hardware to punch even further above its weight without any restrictions on measuring up exactly to just RT features on a specific PC GPU.
 

hlm666

Member
What do you guys think the equivalent gpu is to the current PS5?
It depends what the game is doing, I think we have examples with some coop games and ue5 games which have a bigger cpu load look to be under performing but it could very well be the none linear cost on bandwidth like the ps4 had. The xsx split memory pools may have been an attempt to improve on this but may have had more cons than pros in the end. Can probably look at teardown for another heavy cpu scenario. The faster cpu in these new socs may have a bigger impact on this aswell.

RT work loads gobble up bandwidth aswell as increasing cpu load in current implementations and we can see with spidey 2 they have had to drop under 1080p in heavy moments. Again maybe the increased cpu load from the traffic density increase eats into the bandwidth enough that the resolution has to give.

So heavy RT and/or cpu load ps5 sits around a 2070, no RT with light cpu load the gpu can get enough bandwidth to punch around 2080s level.

I've never seen this addressed in any xsx or ps5 tech talks/slides, so i'm going to assume it's still something to consider especially with how some heavy cpu titles (like ue5 ones) are dropping frame rate and resolution. I can't think of any other logical reason why the current consoles are hitting as low as 720p and sub 60 frame rates. You can see below with the ps4 ~10GB/s cpu bandwidth cost the gpu ~25GB/s. If it also impacts latency cpus arn't going to like that much either.
PS4-GPU-Bandwidth-140-not-176.png
 

Zathalus

Member
Can't compare the two, one is an APU the other is a GPU that consumes more power than the entire console
Try to find something in their mobile GPUs
You can compare it to the 6700 10GB (non XT version). Exact same amount of shaders, TMU's, ROPs, with the only major difference being the higher bandwidth of the PS5 which is offset somewhat by the Infinity Cache and slightly higher GPU clock of the 6700.

Power usage is likely the same as well, the 6700 10GB only uses about 140w or so. Performance is roughly around a 2080 (higher in some games, worse in others). RT performance in Miles Morales is about equal to a PS5. You can do PS5 settings + PS5 RT at 4k (FSR balanced/quality) at a locked 40FPS.
 
Last edited:
You can compare it to the 6700 10GB (non XT version). Exact same amount of shaders, TMU's, ROPs, with the only major difference being the higher bandwidth of the PS5 which is offset somewhat by the Infinity Cache and slightly higher GPU clock of the 6700.

Power usage is likely the same as well, the 6700 10GB only uses about 140w or so. Performance is roughly around a 2080 (higher in some games, worse in others). RT performance in Miles Morales is about equal to a PS5. You can do PS5 settings + PS5 RT at 4k (FSR balanced/quality) at a locked 40FPS.
There is no FSR in PS5 version. Resolution near native 4K most of the time at 40fps.
 

Killjoy-NL

Member
Leaks where?

pulp-fiction-wtf.gif
Just thinking out loud, but would it be possible that they are working on PS6 behind closed doors and send out kits for devs for feedback?

Hence no leaks.

Not sure if those things work like that though. If not, disregard this post.
 
Last edited:

Zathalus

Member
There is no FSR in PS5 version. Resolution near native 4K most of the time at 40fps.
At 40fps it drops under 4k quite often, FSR 2 also has a performance overhead. You can lock to 4k on the PC version as well with a 40fps cap with dynamic resolution if you want to, just obviously not at max settings, you need to use PS5 equivalent.

Its not a perfect match, but they are obviously in the same ballpark. Which is not surprising, the two are almost the exact same architecture.
 
At 40fps it drops under 4k quite often, FSR 2 also has a performance overhead. You can lock to 4k on the PC version as well with a 40fps cap with dynamic resolution if you want to, just obviously not at max settings, you need to use PS5 equivalent.

Its not a perfect match, but they are obviously in the same ballpark. Which is not surprising, the two are almost the exact same architecture.
I said close to native 4K. FSR "4K" is way different than native 1800p / 1900p in rendering cost. And Insomniac TAA (temporal injection) from ~1900p to 4K has also a performance cost.
 
Last edited:

jroc74

Phone reception is more important to me than human rights
Just thinking out loud, but would it be possible that they are working on PS6 behind closed doors and send out kits for devs for feedback?

Hence no leaks.

Not sure if those things work like that though. If not, disregard this post.
I'm still of the mindest that mid gen refreshes are tests for next gen consoles. Instead of keeping the hardware in house for dev purposes release it and get some money out of it.

The performance and quality profiles from last gen mid gen refreshes had to help for current gen development.

I also have to believe next gen consoles have to be in development now, not after a mid gen refresh releases.
 

Mahavastu

Member
Just thinking out loud, but would it be possible that they are working on PS6 behind closed doors and send out kits for devs for feedback?
I am sure they already working on designing the PS6, but it is way too early to send devkits out.
I expect the PS6 in 2027, maybe (less likely) 2028.
For the PS5 the devkits were send to 3rd party during late summer 2019, 1st party maybe got it a bit earlier. Release was November 2020.

In the past they asked the devs a few years after launch about their experiences with the current consoles, what is good, what should be improved, what are the wishes for next gen.
 
Last edited:

Killjoy-NL

Member
I'm still of the mindest that mid gen refreshes are tests for next gen consoles. Instead of keeping the hardware in house for dev purposes release it and get some money out of it.

The performance and quality profiles from last gen mid gen refreshes had to help for current gen development.

I also have to believe next gen consoles have to be in development now, not after a mid gen refresh releases.
Possibly.

Although, PS5-architecture is completely different from PS4 Pro, if I'm not mistaken.
PS4 Pro was nothing but an extra GPU and a bit of extra RAM.

I'm just wondering if these dev kits are actual PS5 Pro devkits, or something in the works for next-gen.
 

jroc74

Phone reception is more important to me than human rights
Possibly.

Although, PS5-architecture is completely different from PS4 Pro, if I'm not mistaken.
PS4 Pro was nothing but an extra GPU and a bit of extra RAM.

I'm just wondering if these dev kits are actual PS5 Pro devkits, or something in the works for next-gen.
Between Kepler, HeisenbergFX4 and Tom Henderson, I'll lean towards it being PS5 Pro until confirmed its not.

I do remember many insiders said Series S was coming, then scrapped, then its coming. So yeah, everything thats a rumor is a rumor until it isnt.
 

Zathalus

Member
I said close to native 4K. FSR "4K" is way different than native 1800p / 1900p in rendering cost. And Insomniac TAA (temporal injection) from ~1900p to 4K has also a performance cost.
As I said, the 6700 can do 4k/40fps with dynamic resolution in place. From what I recall a .6 scaling on each axis was enough to lock to 40. Not quite as good, but definitely in the same ballpark. Returnal does much better but that was a rather straightforward UE4 port.
 

Killjoy-NL

Member
Between Kepler, HeisenbergFX4 and Tom Henderson, I'll lean towards it being PS5 Pro until confirmed its not.

I do remember many insiders said Series S was coming, then scrapped, then its coming. So yeah, everything thats a rumor is a rumor until it isnt.
Fair enough.

But given that only 10% of the PS4 installbase bought a PS4 Pro, at the time PSVR was new and needed a little performance-boost and this gen there is no competition for Sony, nor 'need' for extra power for VR, I'm going to remain sceptic until officially confirmed by Sony.
There is hardly any incentive for Sony to release one.

There should've been concrete leaks already, if Sony themselves also were supposedly expecting full leaks last month.
 
Last edited:
It depends what the game is doing, I think we have examples with some coop games and ue5 games which have a bigger cpu load look to be under performing but it could very well be the none linear cost on bandwidth like the ps4 had. The xsx split memory pools may have been an attempt to improve on this but may have had more cons than pros in the end. Can probably look at teardown for another heavy cpu scenario. The faster cpu in these new socs may have a bigger impact on this aswell.

RT work loads gobble up bandwidth aswell as increasing cpu load in current implementations and we can see with spidey 2 they have had to drop under 1080p in heavy moments. Again maybe the increased cpu load from the traffic density increase eats into the bandwidth enough that the resolution has to give.

So heavy RT and/or cpu load ps5 sits around a 2070, no RT with light cpu load the gpu can get enough bandwidth to punch around 2080s level.

I've never seen this addressed in any xsx or ps5 tech talks/slides, so i'm going to assume it's still something to consider especially with how some heavy cpu titles (like ue5 ones) are dropping frame rate and resolution. I can't think of any other logical reason why the current consoles are hitting as low as 720p and sub 60 frame rates. You can see below with the ps4 ~10GB/s cpu bandwidth cost the gpu ~25GB/s. If it also impacts latency cpus arn't going to like that much either.
PS4-GPU-Bandwidth-140-not-176.png
I’ve always said the cpu is the biggest issue with the current consoles followed by rt capabilities. Thanks for affirming that
 
You can compare it to the 6700 10GB (non XT version). Exact same amount of shaders, TMU's, ROPs, with the only major difference being the higher bandwidth of the PS5 which is offset somewhat by the Infinity Cache and slightly higher GPU clock of the 6700.

Power usage is likely the same as well, the 6700 10GB only uses about 140w or so. Performance is roughly around a 2080 (higher in some games, worse in others). RT performance in Miles Morales is about equal to a PS5. You can do PS5 settings + PS5 RT at 4k (FSR balanced/quality) at a locked 40FPS.
What the situation seems to be is the ps5 is a prettt close match for the 6700 in raw spec but on average performs more like a 6700xt occasionally 6750xt
 
I am sure they already working on designing the PS6, but it is way too early to send devkits out.
I expect the PS6 in 2027, maybe (less likely) 2028.
For the PS5 the devkits were send to 3rd party during late summer 2019, 1st party maybe got it a bit earlier. Release was November 2020.

In the past they asked the devs a few years after launch about their experiences with the current consoles, what is good, what should be improved, what are the wishes for next gen.
Ps6 is 2028-2029 (it’s already kind of confirmed to the ratchet link
 
At 40fps it drops under 4k quite often, FSR 2 also has a performance overhead. You can lock to 4k on the PC version as well with a 40fps cap with dynamic resolution if you want to, just obviously not at max settings, you need to use PS5 equivalent.

Its not a perfect match, but they are obviously in the same ballpark. Which is not surprising, the two are almost the exact same architecture.
Spider-Man remastered and 2 yes not miles
 

foamdino

Member
I'm still of the mindest that mid gen refreshes are tests for next gen consoles. Instead of keeping the hardware in house for dev purposes release it and get some money out of it.

The performance and quality profiles from last gen mid gen refreshes had to help for current gen development.

I also have to believe next gen consoles have to be in development now, not after a mid gen refresh releases.
My honest opinion is that mid-gen refresh consoles allow Sony to retain good designers/engineers. If you have a large gap between console gens then there's the opportunity for there to be a period where people aren't engaged in meaningful (this is a shipping device vs this is a prototype) work and they get bored and leave.

To mitigate this problem you give the engineers interesting stuff to work on (mid-gen console) which allows them to try out some new ideas that may make it into the next-gen design. You give them more practice at designing to fit within constraints (not this fantasy $700 machine), you have a project which you can put less senior engineers on to train them up on how you design a device etc etc.

Many positives from a workforce management point of view - the fact that it ships and earns revenue is (in my opinion) hardly relevant beyond the fact they can prove out ideas/tech.
 

rnlval

Member
It depends what the game is doing, I think we have examples with some coop games and ue5 games which have a bigger cpu load look to be under performing but it could very well be the none linear cost on bandwidth like the ps4 had. The xsx split memory pools may have been an attempt to improve on this but may have had more cons than pros in the end. Can probably look at teardown for another heavy cpu scenario. The faster cpu in these new socs may have a bigger impact on this aswell.

RT work loads gobble up bandwidth aswell as increasing cpu load in current implementations and we can see with spidey 2 they have had to drop under 1080p in heavy moments. Again maybe the increased cpu load from the traffic density increase eats into the bandwidth enough that the resolution has to give.

So heavy RT and/or cpu load ps5 sits around a 2070, no RT with light cpu load the gpu can get enough bandwidth to punch around 2080s level.

I've never seen this addressed in any xsx or ps5 tech talks/slides, so i'm going to assume it's still something to consider especially with how some heavy cpu titles (like ue5 ones) are dropping frame rate and resolution. I can't think of any other logical reason why the current consoles are hitting as low as 720p and sub 60 frame rates. You can see below with the ps4 ~10GB/s cpu bandwidth cost the gpu ~25GB/s. If it also impacts latency cpus arn't going to like that much either.
PS4-GPU-Bandwidth-140-not-176.png

NVIDIA reveals Turing RTX, Ampere, and ADA RT TFLOPS vs shader TFLOPS ratio, hence the major reason for AMD's RT is inferior.

xwuc5zm.jpg


According to Microsoft's Xbox Series X, RDNA 2 has shader TFLOPS nearly matching RT TFLOPS i.e. 12 TFLOPS shader with 13 TFLOPS raytracing.

RDNA 3 CU has a 1.5X RT instructions in flight increase.

AMD needs to substantially increase RT's raw TFLOPS performance. AMD needs to treat RT seriously when hardware RT affects professional apps and gaming use cases. Hint: There's NO mobile Radeon RX 7000 series shown on 2024-era laptops during CES 2024.
 
Last edited:
Top Bottom