Support NeoGAF

[Go_Ly_Dow]

Still waiting for those leaks…

Perhaps spring and around GDC.

 

[Feel Like I'm On 42]

In 2016 SSDs weren't extremely common (or cheap) yet, and on top of that CPU speeds were irrelevant to consumer demands. 4K was becoming in demand as well as PSVR was being developed, hence the Pro making sense to create.

It's very interesting reading these replies completely failing to understand marketing. 95% of consumers buying consoles do not care about "oh the CPU is so weak" or "oh it's using dynamic resolution" or "oh the Ray tracing support isn't good". People DID care that they just spent $1000-$2000 on their brand new 4K TV and the PS4 only outputs 1080p at best and "looks blurry". So it serves a purpose even to those not tech-savvy.
On top of that, saying the ray-tracing support on a PS5 Pro would be dramatically better is kidding themselves, RDNA3 cards still have pretty crap RT performance at the lower end of the card lineup (although a step up from RDNA2), it be a minor difference at best, especially considering this would be an APU focused on efficiency and not a dGPU.

You can cope all you want, PS5 Pro would be ultra-niche, which is why I still don't even believe this is a real product. Someone here assumed maybe 20% of PS5 owners owning one, and that's an incredibly generous number. I cannot imagine the general public giving a shit in the slightest about a PS5 Pro, especially considering a good chunk of PS4 owners probably just upgraded with Spider-Man 2.

You talk as if you know either way whether it's happening or not. You don't. How's PS5 going to keep up with the rate at which PC's are moving ahead. Maybe Sony's afraid their really going to lose more customers to PC. That was one of the reasons why we had ps4 pro. Wasn't just 4k TVs. This gen consoles are getting lapped even worse than last and we havn't yet seen the 5070-90 line of Nvidia cards. Maybe Sony thinks there is a major weakness with PS5 with not having machine learning built into the hardware. Maybe the reasons for getting a Pro this gen are just different but equally valid ...lots of leaks and rumors ..where there's smoke there's usually fire.

 

[HeWhoWalks]

I mean, this is coming from someone who says that if the PS5 Pro isn't 4080-tier in GPU performance, it'll be disappointing.

Ah, now it makes sense.

 

[Nvzman]

You talk as if you know either way whether it's happening or not. You don't. How's PS5 going to keep up with the rate at which PC's are moving ahead. Maybe Sony's afraid their really going to lose more customers to PC. That was one of the reasons why we had ps4 pro. Wasn't just 4k TVs. This gen consoles are getting lapped even worse than last and we havn't yet seen the 5070-90 line of Nvidia cards. Maybe Sony thinks there is a major weakness with PS5 with not having machine learning built into the hardware. Maybe the reasons for getting a Pro this gen are just different but equally valid ...lots of leaks and rumors ..where there's smoke there's usually fire.

I could say the same thing about everyone else here.
Also I don't think Sony would retain customers by releasing a product that is arguably a much less valid value proposition for a console. I'm critiquing the absolutely wack thinking going on in this thread. It's very obvious that nobody here has ever worked in the financial sector.

 

[Dorfdad]

Does anyone think Sony could release a much more powerful system than we believe that will #1. Extend the next gen for another 4-5 years, and #2 Force Microsoft to bleed more money on hardware losses as they would need to spend a lot more than a simple spec bump to keep up even the release a year later etc?

I know there are rumors of a new XBOX but nothing confirmed and Microsoft themselves said it wasn't happening any time soon. Maybe they will have to redesign

 

[Mahavastu]

I could say the same thing about everyone else here.
Also I don't think Sony would retain customers by releasing a product that is arguably a much less valid value proposition for a console.

The PS4Pro was a much better device for only 100€ more than the base. I guess the price difference will be about the same this time, maybe 150 more because of inflation.
For that you get better graphic quality, higher resolutions and more ray tracing.
I am sure there will be quite some people wanting to buy this, just because "why not?". I heard the PS4 pro was only like 20% of the overall PS4 family units sold, but these 20% are those, who probably buy more games and have more expensive subscriptions. So even with a rather small number of sold units it is probably worth it for Playstation, but also for those who buy it.

 

[Gaiff]

No they weren't, I remember paying $50-60 just for a 128gb boot SSD. 1TB was just about $100-$120

And this isn't cheap? Are you joking? You could find 500GB for $80.

It's not like in the late 2000s/early 2010s when 128GB was $200 and people only used one as a boot drive but used HDDs for mass storage.

 

[ChiefDada]

Also I don't think Sony would retain customers by releasing a product that is arguably a much less valid value proposition for a console. I'm critiquing the absolutely wack thinking going on in this thread. It's very obvious that nobody here has ever worked in the financial sector.

 

[poppabk]

The PS4Pro was a much better device for only 100€ more than the base. I guess the price difference will be about the same this time, maybe 150 more because of inflation.
For that you get better graphic quality, higher resolutions and more ray tracing.

Diminishing returns. The same %age jump will give you much less back.

 

[Mr.Phoenix]

When will people realize their will never be a high end console.

It's because of the power comsumption, size, thermals, price.

If you want high end go with PC.

If you ask me, I think the problem is that people are comparing apples to oranges. As far as consoles go, if we are looking at the current gen, the PS5/XSX are "high-end". Consoles. The XSS is a low-end/entry-level console. The key word is, "console". Why anyone tries to compare them to PC gaming hardware is ridiculous.

To me, there is PC gaming, and there is console gaming. If you want the most powerful console you can buy, right now, that is the PS5/XSX. That's a high-end console. Because it's possible to get a PC with specs lower than what a console can give you.

I pray ps5 pro rumors about machine learning upscaler are true so we can ditch FSR2. Alan Wake 2 is ruined due to FSR2 and devs insistence upon using it with these really low base resolutions. The aliasing/shimmering/artifacts are SO god damn ugly and distracting. Some areas of the game are not too terrible but most are and some are unbelievably bad like the Oceanview Hotel.

Nothing is really wrong with FSR2. I mean it's not as good as AI-based reconstruction, but it's good enough. The problem is as you said, the ridiculously low resolutions devs are using as a base for it.

If I would ever pray for anything, its that platform holders set certain standards, or minimum performance presets for a game to even get certified. Eg. They should make the minimum base reconstruction rez for 30fps 1440p and minimum for 60fps 1080p.

 

[Haint]

I mean, this is coming from someone who says that if the PS5 Pro isn't 4080-tier in GPU performance, it'll be disappointing.

Ah, now it makes sense.

The 4080 will be over 2 years old by the time Pro launches, and is a smaller die than the 3070 (which was a $499 card), so it's not that preposterous a statement. Nvidia price gouging everything into a different universe was really successful at recalibrating people's expectations and perspective.

 

[Gaiff]

The 4080 will be over 2 years old by the time Pro launches, and is a smaller die than the 3070 (which was a $499 card), so it's not that preposterous a statement. Nvidia price gouging everything into a different universe was really successful at recalibrating people's expectations and perspective.

Yes, it is. The die area of the 4080 is 379mm with a TDP over 300W. You expect AMD, with their N31 at 529mm and 350W to get that down to less than 200W within a single generation? You seriously think a mid-tier chip from RDNA4 can match the 7900 XTX at around half the thermals? Not happening and it would take some serious hopium or a complete moron to believe that.

 

[FireFly]

The 4080 will be over 2 years old by the time Pro launches, and is a smaller die than the 3070 (which was a $499 card), so it's not that preposterous a statement. Nvidia price gouging everything into a different universe was really successful at recalibrating people's expectations and perspective.

5nm dies are substantially more expensive than 7nm dies, and that 379mm^2 chip (already 23% larger than the PS5 APU) doesn't include a CPU either.

 

[Panajev2001a]

5nm dies are substantially more expensive than 7nm dies, and that 379mm^2 chip (already 23% larger than the PS5 APU) doesn't include a CPU either.

Apparently with 5nm and lower, TSMC advices not to go near or above 400mm^2 and to split into chillers instead, the lithography equipment has issue with larger than that dies.

 

[Mr Reasonable]

Just hope this isn’t a small upgrade. Give us a $999.99 pro console; I can’t go back to 30fps console gaming.

I'd be really interested to see how that went. (Edit: sincerely)

If the premium mid gen system matched what the PS6 performance would be, I'd consider that an expensive but justifiable purchase.

If the PS6 is half that price, more powerful and coming in under 3 years..? I think that's probably too high a price to expect people to stomach.

Pricing on a mid Gen system is still the most interesting part off the equation right now. I have no idea how high Sony need to go to justify releasing the system when PS5 sales are still so strong.

 

[Brigandier]

People really think the PS5 Pro is going to have a 4080 equivalent??? Jesus this thread is nuts

It wIlL bE TwO YEarS OlD... Yes it's also incredible hardware that even in a years time will still be top tier.

I'm seeing Zen 4/5 thrown around a lot too... It's not happening guys calm down.

A GPU on the level of a 7800 XT is actually very competitive especially for a console, Zen 2 can work with some tweaks surely

It's a mid gen refresh not a PS6 it doesn't need to be a killer, as for those saying it's pointless and not needed, To you maybe but to others it is definitely needed, Choice is good if you don't like it don't buy it.

Nothing worse than the "it's not needed brigade" speak for yourselves some people like performance.

 

[winjer]

I think people are focusing too much on just one process node.
I wound not be surprised if the SoC for the PS5 Pro would use at least 2 nodes.

The big RDNA3 chips already use 2 nodes, N5 for the GPU and N6 for the L3 cache.
And Zen CPUs have used different processes for the IOD die and the CPU die.

AMD could very well make an SoC with the Zen2 CPU at N6. While having the GPU at N5 or N4.
And considering the GPU will be more powerful, AMD can just use some L3 cache in N6 or N7, instead of making the memory bus wider. Thus saving a bit of power.
And let's remember that sram cells don't scale as well, with newer process nodes, as logic cells.

 

[Mr.Phoenix]

I think people are focusing too much on just one process node.
I wound not be surprised if the SoC for the PS5 Pro would use at least 2 nodes.

The big RDNA3 chips already use 2 nodes, N5 for the GPU and N6 for the L3 cache.
And Zen CPUs have used different processes for the IOD die and the CPU die.

AMD could very well make an SoC with the Zen2 CPU at N6. While having the GPU at N5 or N4.
And considering the GPU will be more powerful, AMD can just use some L3 cache in N6 or N7, instead of making the memory bus wider. Thus saving a bit of power.
And let's remember that sram cells don't scale as well, with newer process nodes, as logic cells.

All that only works if the PS5 APU is not monolithic but a chipset. I doubt that.

 

[HeWhoWalks]

The 4080 will be over 2 years old by the time Pro launches, and is a smaller die than the 3070 (which was a $499 card), so it's not that preposterous a statement. Nvidia price gouging everything into a different universe was really successful at recalibrating people's expectations and perspective.

A $1000 console being more powerful than almost any PC is very much a preposterous statement.

 

[asdasdasdbb]

I think people are focusing too much on just one process node.
I wound not be surprised if the SoC for the PS5 Pro would use at least 2 nodes.

The big RDNA3 chips already use 2 nodes, N5 for the GPU and N6 for the L3 cache.
And Zen CPUs have used different processes for the IOD die and the CPU die.

Unlikely. The PS6, yeah, I would expect that to use chiplets.

 

[ChiefDada]

A $1000 console being more powerful than almost any PC is very much a preposterous statement.

Meh, I'll push back on this just for shits and giggles. Since business objectives are divergent - console manufacturers aim for high volume while gpu manufacturers aim for high margin - I would say it's very possible for $1k console to be more powerful in an imaginary world where game development cost isn't factored into the equation.

 

[M1987]

The 4080 will be over 2 years old by the time Pro launches, and is a smaller die than the 3070 (which was a $499 card), so it's not that preposterous a statement. Nvidia price gouging everything into a different universe was really successful at recalibrating people's expectations and perspective.

Not a chance the Pro will be 4080 level

 

[winjer]

Unlikely. The PS6, yeah, I would expect that to use chiplets.

The tech already exists and AMD is proficient with chiplets.
And with process node prices sky high, this is a great way to save costs.

 

[PeteBull]

Guys, to make ps5pr0 at lvl similar to 4080 sony would have to undervolt/downclock best avaiable chip amd currently has to offer, 7900xtx chip, and it still wouldnt be enough.

AMD Radeon RX 7900 XTX Specs

AMD Navi 31, 2498 MHz, 6144 Cores, 384 TMUs, 192 ROPs, 24576 MB GDDR6, 2500 MHz, 384 bit

www.techpowerup.com

Now lets not even bring up price(which likely would make console cost 999usd), but focus on its 355W tdp, made on 5nm process node, and its avaiability, its 96CU and 529mm die, so relatively big, way too big for console chip especially that its just gpu alone there, for it to be apu u would need to put cpu on it too.

And another thing to remember, its topend of amd has to offer, aka it cant be produced like massproduced console at 10-15milion units/year, i bet at max it can be maybe 1m units yearly production, thats the biggest problem that scratches this idea off right away, coz even as big techenthusiast im fully aware sony wont make ps5pr0 with a cap of 1m units production/year, they will want it to have at the very least 4-5m units/year for midgen console.

 

[Mr.Phoenix]

The tech already exists and AMD is proficient with chiplets.
And with process node prices sky high, this is a great way to save costs.

Its not about the tech not being there. Its about it not being necessary in a console. At this time.

Again, the PS5pro is going to be based off the 7xxx series RDNA3 GPU. That GPU peaks out at 60CU. So right off the bat, there is no way that the PS5 has more than 64CU at best. The 7xxx series uses a chiplet design, it has the entire GPU (GCD) on one chiplet (200mm2), and then memory controllers and cache (MCD) on separate chiplets (~40mm2 each). The reason they do this is that memory controllers and cache are harder to shrink, and the MCD is interchangeable, i.e. you can put say 2 MCDs in a 7600xt, 3 in a 7700xt and 4 in a 7800xt...etc.

Ideally, a PS chiplet would have the CPU+GPU be on one chiplet (say an APD) and the MCD be on separate chiplets. All that isn't going into a PS5pro. PS6 sure, but not the PS5pro.

Especially when you consider that even if they made a PS5pro monolithic APU smaller than the OG PS5 APU (that was 320mm2 on 7nm), so say around 300mm2 on a 4/5nm process, they would still be able to fit those 60/64CU, an 8c zen 2 CPU, 8x32-bit memory PHY, I/O complex, and likely even increase cache from 8MB to 16MB.

So why would they opt to make an all round more expensive chip that does the exact same thing all because they are trying to use chiplets?

 

[winjer]

Its not about the tech not being there. Its about it not being necessary in a console. At this time.

Again, the PS5pro is going to be based off the 7xxx series RDNA3 GPU. That GPU peaks out at 60CU. So right off the bat, there is no way that the PS5 has more than 64CU at best. The 7xxx series uses a chiplet design, it has the entire GPU (GCD) on one chiplet (200mm2), and then memory controllers and cache (MCD) on separate chiplets (~40mm2 each). The reason they do this is that memory controllers and cache are harder to shrink, and the MCD is interchangeable, i.e. you can put say 2 MCDs in a 7600xt, 3 in a 7700xt and 4 in a 7800xt...etc.

Ideally, a PS chiplet would have the CPU+GPU be on one chiplet (say an APD) and the MCD be on separate chiplets. All that isn't going into a PS5pro. PS6 sure, but not the PS5pro.

Especially when you consider that even if they made a PS5pro monolithic APU smaller than the OG PS5 APU (that was 320mm2 on 7nm), so say around 300mm2 on a 4/5nm process, they would still be able to fit those 60/64CU, an 8c zen 2 CPU, 8x32-bit memory PHY, I/O complex, and likely even increase cache from 8MB to 16MB.

So why would they opt to make an all round more expensive chip that does the exact same thing all because they are trying to use chiplets?

But it is necessary to keep costs down. Process node costs are increasingly higher. And this already includes the N5/N4 nodes.
And with sram and analog scaling a lot worse than logic density, there is a great incentive to have these things in cheaper nodes.
Another reason is having to create new masks, which are very expensive. If using the old masks on N7/N6 for the CPU and caches, it would mean AMD would not need to make as many new masks.
They would only need to make new masks for the GPU chiplets.

BTW, the 7600 does not have separate dies for cache.

 

[Gaiff]

Meh, I'll push back on this just for shits and giggles. Since business objectives are divergent - console manufacturers aim for high volume while gpu manufacturers aim for high margin - I would say it's very possible for $1k console to be more powerful in an imaginary world where game development cost isn't factored into the equation.

If it was strictly limited to price under the current market conditions, you'd be correct. However, there are other factors such as size, noise, power consumption, heat, etc. Good luck fitting that into a single APU when the GPU alone is 609mm.

The profit margins of NVIDIA are nothing short of disgusting whereas Sony and Microsoft with the economy of scale at play get a massive discount on their chips compared to consumer ones.

 

[Mr.Phoenix]

But it is necessary to keep costs down. Process node costs are increasingly higher. And this already includes the N5/N4 nodes.
And with sram and analog scaling a lot worse than logic density, there is a great incentive to have these things in cheaper nodes.
Another reason is having to create new masks, which are very expensive. If using the old masks on N7/N6 for the CPU and caches, it would mean AMD would not need to make as many new masks.
They would only need to make new masks for the GPU chiplets.

BTW, the 7600 does not have separate dies for cache.

Think about it...

PS5pro on 4/5nm node would have a ~300mm2 monolithic chip.

PS5pro on chiplet will have 1x 4/5nm APD (CPU+GPU)@~240mm2 + 4x 6nm MCD (each having 2x32bit mem PHY, 2-4MB cache)@~20mm2 + 1x interposer (substrate that both APD + MCDs are on)

And you really think the second option would be cheaper for the PS5pro? The reason chipletrs work for AMD GPUs, outside the GPU alone costing more than a PS5pro, is because those GPUs use over 64-96MB of cache, each MCD usually has anywhere between 12-16MB of cache in them (at least in the 7xxx series). That's primarily what gets relegated to a separate die due to how they don't shrink that well. And you want them to do that for a PS5pro, with MCDs that have what? 2MB or 4MB of cache in them??? There is more cache in 1 MCD than there would likely be in the whole PS5pro.

And you shouldn't mention the 7600... its monolithic. Not using chiplets. Has 32CU, 32MB of cache, 128bit bus....etc and all in a chip size of ~200mm2.

 

[winjer]

Think about it...

PS5pro on 4/5nm node would have a ~300mm2 monolithic chip.

PS5pro on chiplet will have 1x 4/5nm APD (CPU+GPU)@~240mm2 + 4x 6nm MCD (each having 2x32bit mem PHY, 2-4MB cache)@~20mm2 + 1x interposer (substrate that both APD + MCDs are on)

And you really think the second option would be cheaper for the PS5pro? The reason chipletrs work for AMD GPUs, outside the GPU alone costing more than a PS5pro, is because those GPUs use over 64-96MB of cache, each MCD usually has anywhere between 12-16MB of cache in them (at least in the 7xxx series). That's primarily what gets relegated to a separate die due to how they don't shrink that well. And you want them to do that for a PS5pro, with MCDs that have what? 2MB or 4MB of cache in them??? There is more cache in 1 MCD than there would likely be in the whole PS5pro.

And you shouldn't mention the 7600... its monolithic. Not using chiplets. Has 32CU, 32MB of cache, 128bit bus....etc and all in a chip size of ~200mm2.

The PS5 Pro can't be just more CUs. It will also have to have more memory bandwidth. GDDR6 speeds have improve a bit, but probably not enough for a bigger GPU. Especially when it has to contend with the CPU.
The option is to have more channels, which means more power usage and more used die space. Or add a chunk of L3 cache on a separate node. 64MB probably would work great, both for performance and for power usage.

How much it would cost, neither of us has numbers for these things. But wafer costs are already very high. And chiplets are a great way to save money.

 

[Mr.Phoenix]

The PS5 Pro can't be just more CUs. It will also have to have more memory bandwidth. GDDR6 speeds have improve a bit, but probably not enough for a bigger GPU. Especially when it has to contend with the CPU.

The PS5pro isnt just more CUs. Its more RDNA3.5 CUs, which means that each CU has VOPD, 2 AI units, 3rd gen RT cores.... all these things make that CU much better than what was in the OG PS5 RDNA 1.5 GPU, it will have an upclocked CPU (3.5Ghz > 4.4Ghz), will have more cache than what was in the OG PS5 which had 8Mb, so it would have maybe 12-16MB at least. And it does have more mem bandwidth because its using 16gbs mem chips instead of the 14 found in the OG PS5 taking memory bandwidth up from 448GB/s to 512GB/s. So yeah, there are improvements across the board. None of which warrants having a bigger die than 320mm2 on a 4/5nm node though.

The option is to have more channels, which means more power usage and more used die space. Or add a chunk of L3 cache on a separate node. 64MB probably would work great, both for performance and for power usage.

Nope, they aren't doing any of that. It will be the same 8 channels and a very small uptick in cache. 64MB? Lmao. We would be lucky if they even take it up to 20/24MB. My guess puts it at 12-16MB... for reference the current PS5 has 8MB.

Also, you (people) really need to understand exactly what the PS5pro is fixing to do.

How much it would cost, neither of us has numbers for these things. But wafer costs are already very high. And chiplets are a great way to save money.

Chiplets only save you money depending on what size of chip you would otherwise have to make to begin with. The PS5pro, simply doesn't need that.

 

[winjer]

The PS5pro isnt just more CUs. Its more RDNA3.5 CUs, which means that each CU has VOPD, 2 AI units, 3rd gen RT cores.... all these things make that CU much better than what was in the OG PS5 RDNA 1.5 GPU, it will have an upclocked CPU (3.5Ghz > 4.4Ghz), will have more cache than what was in the OG PS5 which had 8Mb, so it would have maybe 12-16MB at least. And it does have more mem bandwidth because its using 16gbs mem chips instead of the 14 found in the OG PS5 taking memory bandwidth up from 448GB/s to 512GB/s. So yeah, there are improvements across the board. None of which warrants having a bigger die than 320mm2 on a 4/5nm node though.

Nope, they aren't doing any of that. It will be the same 8 channels and a very small uptick in cache. 64MB? Lmao. We would be lucky if they even take it up to 20/24MB. My guess puts it at 12-16MB... for reference the current PS5 has 8MB.

Also, you (people) really need to understand exactly what the PS5pro is fixing to do.

Chiplets only save you money depending on what size of chip you would otherwise have to make to begin with. The PS5pro, simply doesn't need that.

Sorry, but all this is mere speculation of what the chip will be.
What whatever the new PS5 Pro will use, if it doubles compute and adds more functions like dedicated AI and RT cores, then an improvement of 14% in GDDR6 speed will not be enough.
It will have to have more memory channels or more cache to make up for it.
Consider that the 7800XT has a 256 bit bus, doesn't have to share bandwidth with a CPU. And has 64MB of L3 cache.

AMD has been using chiplets since Zen1, made in N14. From the lowest end CPU SKUs, to the high end workstation CPUs.
And the reason is that they are cost saving. In part because it uses cheaper nodes for some parts, but it also improves yields a lot. And binning with chiplets is also improved.
Whatever are the cost for packaging and interposers, it is lower than having a monolithic chip.
AMD started with this tech with Zen CPUs, but it has matured enough, that we are already seeing it in GPUs, like RDNA3.

Now mind you, I'm not sure that the PS5 Pro is using chiplets or not. I'm just speculating.

 

[Mr.Phoenix]

Sorry, but all this is mere speculation of what the chip will be.
What whatever the new PS5 Pro will use, if it doubles compute and adds more functions like dedicated AI and RT cores, then an improvement of 14% in GDDR6 speed will not be enough.
It will have to have more memory channels or more cache to make up for it.
Consider that the 7800XT has a 256 bit bus, doesn't have to share bandwidth with a CPU. And has 64MB of L3 cache.

Sigh... that is not how more works. Unlike the 7800... the PS5 is not trying to push upwards of 200fps in games at 1080p. Is not having to accommodate for "extreme" settings on PC... no, this is why I said you need to understand what the PS5 is designed to do.

The PS5 is going to take that 30fps PS5 game running at 1280-1440p before reconstruction to 4K, and lock that to 1440p then take that up to 60fps. You do not need significantly more bandwidth to do that.

And you are dismissing my speculations as "just speculation" then making yours? Not how that works...

All this talk about more channels, more cache, chiplets... its simply just more expensive than just going with a tried and true 300-320mm2 monolithic die. And what I am saying is in accordance with what Sony has done in the past, so there is precedent... what you are saying is just based on what you wish they would do.

AMD has been using chiplets since Zen1, made in N14. From the lowest end CPU SKUs, to the high end workstation CPUs.
And the reason is that they are cost saving. In part because it uses cheaper nodes for some parts, but it also improves yields a lot. And binning with chiplets is also improved.
Whatever are the cost for packaging and interposers, it is lower than having a monolithic chip.
AMD started with this tech with Zen CPUs, but it has matured enough, that we are already seeing it in GPUs, like RDNA3.

Stop it, using chipsets on the CPU and then using it on the GPU is not the same thing. And they only started using it on the GPuy with RDNA3. All RDNA2 GPUs were monolithic. And their chiplet tech on the GPU is so underdeveloped that they do not even have a high-end 7xxx series GPU. But what's crazy is that you are not even just asking for chiplets, you are saying they should make a chiplet that encompasses the CPU. That's not something that's been done before.

Now mind you, I'm not sure that the PS5 Pro is using chiplets or not. I'm just speculating.

Exactly, as am I...

I feel you are not being practical, not just of what can be done, but considering what sony is likely to do. You can't just make speculations based on whatever cutting-edge tech is out there, that's just not how these things work. At that point, we might as well bring HBM3e into the conversation.

 

[winjer]

Sigh... that is not how more works. Unlike the 7800... the PS5 is not trying to push upwards of 200fps in games at 1080p. Is not having to accommodate for "extreme" settings on PC... no, this is why I said you need to understand what the PS5 is designed to do.

The PS5 is going to take that 30fps PS5 game running at 1280-1440p before reconstruction to 4K, and lock that to 1440p then take that up to 60fps. You do not need significantly more bandwidth to do that.

And you are dismissing my speculations as "just speculation" then making yours? Not how that works...

All this talk about more channels, more cache, chiplets... its simply just more expensive than just going with a tried and true 300-320mm2 monolithic die. And what I am saying is in accordance with what Sony has done in the past, so there is precedent... what you are saying is just based on what you wish they would do.

It doesn't matter the frame rate or resolution of upscaling. What matters is how many instructions are passing through the GPU.
And if the memory is not feeding all those CUs, then what is the point of having more of them, if many will be idle.
The point of having more cache or bandwidth is to be able to feed more and more units.

And you don't know the costs of using chiplets. The reality is that AMD has been using chiplets to make cheaper CPUs for several years, and it now has the tech to do it also with GPUs.
The reality is that chiplets are cheaper to make than monolithic chips.

Yes, your speculations are just speculations. Just like mine and anyone else.
Until we have concrete data about the PS5 Pro, speculations is all we have.

Stop it, using chipsets on the CPU and then using it on the GPU is not the same thing. And they only started using it on the GPuy with RDNA3. All RDNA2 GPUs were monolithic. And their chiplet tech on the GPU is so underdeveloped that they do not even have a high-end 7xxx series GPU. But what's crazy is that you are not even just asking for chiplets, you are saying they should make a chiplet that encompasses the CPU. That's not something that's been done before.

The reason for RDNA3 being a weak GPU arch is not about chiplets. It's about how the GPU is designed.
Be it the big RDNA3 or small RDNA3, the memory subsystem is not an issue for them.

 

[Mr.Phoenix]

It doesn't matter the frame rate or resolution of upscaling. What matters is how many instructions are passing through the GPU.
And if the memory is not feeding all those CUs, then what is the point of having more of them, if many will be idle.
The point of having more cache or bandwidth is to be able to feed more and more units.

That's not how this works at all. Your resolution, your framerate, size of your textures all time into your RAM bandwidth. Cache size is what has to do with keeping your processors fed with data. And that even has more to do with L0/L1 cache than L2 cache... and yeah, the PS5 doesn't even use L3 cache. Each CU has its own cache, L0, then they have an SA fallback, L1, and then a cache shared between the GPU and CPU, L2. This isn't rocket science, if they are increasing internal L2 cache from 8MB to 16MB, that is doubling its throughput considering the PS5pro is ultimately still just running PS5 games.

And you don't know the costs of using chiplets. The reality is that AMD has been using chiplets to make cheaper CPUs for several years, and it now has the tech to do it also with GPUs.
The reality is that chiplets are cheaper to make than monolithic chips.

at this point, we are just going round in circles... there are lots of obvious and known reasons as to why AMD does what they do and why Sony won't. I am done however arguing, so lets just wait and see what happens with the PS5pro. I have told you they would not just use a monolithic chip, but that they would be using the same 16GB of RAM (albeit faster modules) and the same 256-bit bus. And 24MB/less of cache or at least some small increment over the current 8MB but definitely nowhere near 32/64MB.

We will see who is right at the end of the day. Till then, we can agree to disagree.

Yes, your speculations are just speculations. Just like mine and anyone else.
Until we have concrete data about the PS5 Pro, speculations is all we have.

agreed.

The reason for RDNA3 being a weak GPU arch is not about chiplets. It's about how the GPU is designed.
Be it the big RDNA3 or small RDNA3, the memory subsystem is not an issue for them.

Never said RDNA3 was a weak GPU. I said that their GPU chiplet design is underdeveloped (or at least still in its infancy), evident in the fact that they didn't make a high-end 7xxx GPU at that series launch. Hence why there is no 7900xtx. Basically, the 7xxx GPU is the first-gen chiplet-based GPU, kinda like how first-gen RDNA peaked at 5700xt.

Precedent and/or history matters. It helps you better predict what is to come.

 

[winjer]

That's not how this works at all. Your resolution, your framerate, size of your textures all time into your RAM bandwidth. Cache size is what has to do with keeping your processors fed with data. And that even has more to do with L0/L1 cache than L2 cache... and yeah, the PS5 doesn't even use L3 cache. Each CU has its own cache, L0, then they have an SA fallback, L1, and then a cache shared between the GPU and CPU, L2. This isn't rocket science, if they are increasing internal L2 cache from 8MB to 16MB, that is doubling its throughput considering the PS5pro is ultimately still just running PS5 games.

Now you are making stuff up when you say that the memory subsystem does not keep processors feed with data. This is obviously false.
Data for a GPU or CPU to process, has to come from somewhere.
If we double the amount of CUs and maybe even their clock speed, we need to increase the data troughput to keep those units feed.
Be it with more caches, more channels, faster memory, but it has to be done.

Never said RDNA3 was a weak GPU. I said that their GPU chiplet design is underdeveloped (or at least still in its infancy), evident in the fact that they didn't make a high-end 7xxx GPU at that series launch. Hence why there is no 7900xtx. Basically, the 7xxx GPU is the first-gen chiplet-based GPU, kinda like how first-gen RDNA peaked at 5700xt.

AMD has fewer resources than Intel or NVidia, especially a few years ago.
They can't develop chiplets for everything at the same time. They started with CPUs, which have higher profit margins than GPUs.
Only recently with RDNA3 were they able to have some of the chiplet tech transferred to a GPU. The big RDNA3 is just the start.
But RDNA3 was released 2 years ago, so it's likely that AMD has made moves to improve chiplet tech for GPUs.
This is why I say it's a possibility that AMD could use chiplets for something like a PS5 Pro.

 

[Mr.Phoenix]

Now you are making stuff up when you say that the memory subsystem does not keep processors feed with data. This is obviously false.
Data for a GPU or CPU to process, has to come from somewhere.
If we double the amount of CUs and maybe even their clock speed, we need to increase the data troughput to keep those units feed.
Be it with more caches, more channels, faster memory, but it has to be done.

Pls take the time to read what is being said and not just read to argue.

The more you talk the more I wonder if you even understand exactly what the cache's role is and what RAM is for.

But hey, I said lets agree to disagree.

AMD has fewer resources than Intel or NVidia, especially a few years ago.
They can't develop chiplets for everything at the same time. They started with CPUs, which have higher profit margins than GPUs.
Only recently with RDNA3 were they able to have some of the chiplet tech transferred to a GPU. The big RDNA3 is just the start.
But RDNA3 was released 2 years ago, so it's likely that AMD has made moves to improve chiplet tech for GPUs.
This is why I say it's a possibility that AMD could use chiplets for something like a PS5 Pro.

RDNA3 was released this year. Not two years ago. THIS YEAR.

smh...

 

[winjer]

Pls take the time to read what is being said and not just read to argue.

The more you talk the more I wonder if you even understand exactly what the cache's role is and what RAM is for.

But hey, I said lets agree to disagree.

You don't think that a GPU that has twice the compute units, will need a more robust memory subsystem?
Because that is what has been happening with every generation. Be it more caches, faster memory, more channels, better compression, or a mix and match of any of these.

RDNA3 was released this year. Not two years ago. THIS YEAR.

smh...

We are both wrong. It was released at the end of 2022.

AMD Radeon RX 7900 XT Specs

AMD Navi 31, 2394 MHz, 5376 Cores, 336 TMUs, 192 ROPs, 20480 MB GDDR6, 2500 MHz, 320 bit

www.techpowerup.com

Release DateNov 3rd, 2022
AvailabilityDec 13th, 2022

 

[rsouzadk]

Ps5 pro at the level of a 4080?

Yes guys.. keep dreaming

 

[Mr.Phoenix]

You don't think that a GPU that has twice the compute units, will need a more robust memory subsystem?
Because that is what has been happening with every generation. Be it more caches, faster memory, more channels, better compression, or a mix and match of any of these.

Nope, I don't.

The PS4pro had twice the CU to the PS4, the PS5pro is not even twice the CU, and the PS4pro still maintained the same RAM quantity and bus width and only increased bandwidth from 176GB/s to 218GB/s. That's less than a 30% boost in bandwidth, no changes were made to L0 and L1 caches. Oh, and the pixel jump the PS4pro was attempting to make was more than what the PS5pro is going o be making for the PS5.

As I said, precedent is important.

And I didn't say it wasn't even getting a memory sub-system upgrade. I said bandwidth is increasing from 445GB/s to 512Gb/s and that L2 cache will increase from 8 to 12/16MB at best. And besides, increasing the GPU clock also increases the cache bandwidth. You really aren't listening and just trying to win an argument lol.

What I am saying in addition to what I already said. Is that you are just flat-out wrong. They will NOT be using chiplets. They will NOT have 64MB of cache. They will NOT be increasing channels...etc.

We are both wrong. It was released at the end of 2022.

AMD Radeon RX 7900 XT Specs

AMD Navi 31, 2394 MHz, 5376 Cores, 336 TMUs, 192 ROPs, 20480 MB GDDR6, 2500 MHz, 320 bit

www.techpowerup.com

Release DateNov 3rd, 2022
AvailabilityDec 13th, 2022

Same difference really, it sure as hell wasn't 2 years ago.

 

[winjer]

Nope, I don't.

The PS4pro had twice the CU to the PS4, the PS5pro is not even twice the CU, and the PS4pro still maintained the same RAM quantity and bus width and only increased bandwidth from 176GB/s to 218GB/s. That's less than a 30% boost in bandwidth, no changes were made to L0 and L1 caches. Oh, and the pixel jump the PS4pro was attempting to make was more than what the PS5pro is going o be making for the PS5.

As I said, precedent is important.

The PS4 Pro had several improvements to it's memory subsystem. For starter, it has almost 25% higher memory bandwidth than the PS4. And it has bigger L2 caches.
It also has Delta Color Compression. Something that Polaris based GPUs have, but earlier GCN GPUs didn't have. AMD claims it improves memory bandwidth b 35%.

Same difference really, it sure as hell wasn't 2 years ago.

Sure as hell it wasn't this year.

 

[Mr.Phoenix]

The PS4 Pro had several improvements to it's memory subsystem. For starter, it has almost 25% higher memory bandwidth than the PS4. And it has bigger L2 caches.
It also has Delta Color Compression. Something that Polaris based GPUs have, but earlier GCN GPUs didn't have. AMD claims it improves memory bandwidth b 35%.

Dude just stop. Like just take the time to read and understand before you talk.

The PS4pro has higher memory bandwidth? 176GB/s to 218GB/s. So will the PS5pro. PS5 has 448GB/s and the pro will have at least 512GB/s if using 16gbs modules or 576GB/s if using the rumored 18gbs modules. So that's between ~20% or almost 35% increase in RAW main bandwidth. I also pointed out that the PS5pro could have double the cache, which unfortunately for you just means it goes up from 8MB to 16MB. Thats the L2 cache... and don't even get me started on the fact that that L2 cache I have been talking about is for the CPU not even the GPU because if you really wanna be anal about it, the PS5 GPU only has 4MB of L2 cache.

Sure as hell it wasn't this year.

In truth when I said this year, I was talking about the 7800xt. But yeah, same difference, I don't care if I am right on that.

 

[Salty Pickle]

Nope, I don't.

The PS4pro had twice the CU to the PS4, the PS5pro is not even twice the CU, and the PS4pro still maintained the same RAM quantity and bus width and only increased bandwidth from 176GB/s to 218GB/s. That's less than a 30% boost in bandwidth, no changes were made to L0 and L1 caches. Oh, and the pixel jump the PS4pro was attempting to make was more than what the PS5pro is going o be making for the PS5.

As I said, precedent is important.

And I didn't say it wasn't even getting a memory sub-system upgrade. I said bandwidth is increasing from 445GB/s to 512Gb/s and that L2 cache will increase from 8 to 12/16MB at best. And besides, increasing the GPU clock also increases the cache bandwidth. You really aren't listening and just trying to win an argument lol.

What I am saying in addition to what I already said. Is that you are just flat-out wrong. They will NOT be using chiplets. They will NOT have 64MB of cache. They will NOT be increasing channels...etc.

Same difference really, it sure as hell wasn't 2 years ago.

But one of the major weak points in the PS4 Pro was the lacking mem bandwith, do you think they would repeat that same mistake?

 

[Mr.Phoenix]

But one of the major weak points in the PS4 Pro was the lacking mem bandwith, do you think they would repeat that same mistake?

Yes. And that is because sony doesn't design their hardware the way some people seem to think of it. They take everything into account. And I mean EVERYTHING. Eg. Sony went with lower bandwidth on the PS4pro because they knew they were never really trying to push native 4K but instead using CBR which only renders half the pixels of 4K.

And they are going to do the same thing now. They are not making the PS5pro to be this magical native 4K pushing 60fps machine that even a 4080 can't do. They are making it to be a PS5+.

And that is the way anyone should look at this before making any assumptions or speculations.

So here is the question, the PS5 in quality mode is internally running games at 1440p@35-45fps and then reconstructed to 4K. (yes, you need to be running a game at that framerate range if you want to have a locked 30fps).

So what do you need to add to the PS5, to let you take the exact same game above, and run that at 60fps in quality mode? So basically, you are trying to go from 35fps minimum to around 65fps minimum. The internal resolution remains unchanged.

 

[winjer]

The PS4pro has higher memory bandwidth? 176GB/s to 218GB/s. So will the PS5pro. PS5 has 448GB/s and the pro will have at least 512GB/s if using 16gbs modules or 576GB/s if using the rumored 18gbs modules. So that's between ~20% or almost 35% increase in RAW main bandwidth. I also pointed out that the PS5pro could have double the cache, which unfortunately for you just means it goes up from 8MB to 16MB. Thats the L2 cache... and don't even get me started on the fact that that L2 cache I have been talking about is for the CPU not even the GPU because if you really wanna be anal about it, the PS5 GPU only has 4MB of L2 cache.

Why are you being so obtuse about this. It has always been that faster GPUs, require more data throughput to keep their units feed.
The PS4 Pro didn't just increase the memory bandwidth, it also had bigger caches and DCC. All together it was a significant improvement to the PS4 pro memory subsystem. And even then some would argue it wasn't enough.

The PS5 Pro will need more than just a 20-30% improvement in memory bandwidth.
You can be certain that it will have bigger caches and/or more channels.

In truth when I said this year, I was talking about the 7800xt. But yeah, same difference, I don't care if I am right on that.

We were talking about RDNA3. Not the 7800XT specifically.
The reality is that I was wrong in saying it was 2 years ago. And you were wrong in saying it was released this year.

 

[Haint]

Yes, it is. The die area of the 4080 is 379mm with a TDP over 300W. You expect AMD, with their N31 at 529mm and 350W to get that down to less than 200W within a single generation? You seriously think a mid-tier chip from RDNA4 can match the 7900 XTX at around half the thermals? Not happening and it would take some serious hopium or a complete moron to believe that.

A $1000 console being more powerful than almost any PC is very much a preposterous statement.

The fact that AMD's utterly incompetent and generations behind Nvidia doesn't mean it won't be a dissapointment though. That statement doesn't inherently imply or suggest he believes it will be around a 4080 (it absolutely will not), it means it should and is technically possible.

Again, the 4080 will be over 2 years old and is actually a XX70 series die. It is objectively true that it will be a dissapointment if a 2 year newer and expected premium priced console cannot come close to a 2 year old 70 series GPU. That is like the minimum people should be expecting and demanding, and is historically what modern consoles have achieved (rough parity with the prior gen 70 series cards).

 

[Mr.Phoenix]

Why are you being so obtuse about this. It has always been that faster GPUs, require more data throughput to keep their units feed.
The PS4 Pro didn't just increase the memory bandwidth, it also had bigger caches and DCC. All together it was a significant improvement to the PS4 pro memory subsystem. And even then some would argue it wasn't enough.

The PS5 Pro will need more than just a 20-30% improvement in memory bandwidth.
You can be certain that it will have bigger caches and/or more channels.



We were talking about RDNA3. Not the 7800XT specifically.
The reality is that I was wrong in saying it was 2 years ago. And you were wrong in saying it was released this year.

Ok sir... lets agree to disagree.

My hypothetical PS5 specs are as follows

• Zen2-based 8c/16t CPU @4.4Ghz + 12-16MB of cache (increased from 8MB)
• RDNA3 based 56/60CU GPU @2.35Ghz + 8MB of L2 cache (increased from 4MB) (16.8TF/18TF)
• 16GB GDDR6 on 256-bit bus. 18Gbs speed. Which translates to 576GB/s of bandwidth (128GB more bandwidth than PS5)
• APU (monolithic die)
  
  Special mention
• AI cores present in each CU (already present in RDNA3)
• using 3rd gen RT units (would be in RDNA4)
• VOPD support (also present in RDNA3) and as such the TF number would be marketed as 33/35TF

So that's my take on it. rather than we continue arguing about what could and couldn't be, how about you list how you think it would be and we see which of us is closer when it drops or we at least have more concrete agreed-upon information. And how about we make this more interesting, lets make it a 1-month avatar bet?

 

[winjer]

Ok sir... lets agree to disagree.

My hypothetical PS5 specs are as follows

• Zen2-based 8c/16t CPU @4.4Ghz + 12-16MB of cache (increased from 8MB)
• RDNA3 based 56/60CU GPU @2.35Ghz + 8MB of L2 cache (increased from 4MB) (16.8TF/18TF)
• 16GB GDDR6 on 256-bit bus. 18Gbs speed. Which translates to 576GB/s of bandwidth (128GB more bandwidth than PS5)
• APU (monolithic die)
  
  Special mention
• AI cores present in each CU (already present in RDNA3)
• using 3rd gen RT units (would be in RDNA4)
• VOPD support (also present in RDNA3) and as such the TF number would be marketed as 33/35TF

So that's my take on it. rather than we continue arguing about what could and couldn't be, how about you list how you think it would be and we see which of us is closer when it drops or we at least have more concrete agreed-upon information. And how about we make this more interesting, lets make it a 1-month avatar bet?

You disagree with me, but then increase caches for the CPU and GPU like I said it was needed.....

 

[Mr.Phoenix]

You disagree with me, but then increase caches for the CPU and GPU like I said it was needed.....

You see what I mean... you have been reading to argue as opposed to reading to understand.

Its not about the tech not being there. Its about it not being necessary in a console. At this time.

Again, the PS5pro is going to be based off the 7xxx series RDNA3 GPU. That GPU peaks out at 60CU. So right off the bat, there is no way that the PS5 has more than 64CU at best. The 7xxx series uses a chiplet design, it has the entire GPU (GCD) on one chiplet (200mm2), and then memory controllers and cache (MCD) on separate chiplets (~40mm2 each). The reason they do this is that memory controllers and cache are harder to shrink, and the MCD is interchangeable, i.e. you can put say 2 MCDs in a 7600xt, 3 in a 7700xt and 4 in a 7800xt...etc.

Ideally, a PS chiplet would have the CPU+GPU be on one chiplet (say an APD) and the MCD be on separate chiplets. All that isn't going into a PS5pro. PS6 sure, but not the PS5pro.

Especially when you consider that even if they made a PS5pro monolithic APU smaller than the OG PS5 APU (that was 320mm2 on 7nm), so say around 300mm2 on a 4/5nm process, they would still be able to fit those 60/64CU, an 8c zen 2 CPU, 8x32-bit memory PHY, I/O complex, and likely even increase cache from 8MB to 16MB.

So why would they opt to make an all round more expensive chip that does the exact same thing all because they are trying to use chiplets?

From the beginning of this I had made it clear that there will be a cache increase.

What I have disagreed with from you was all that nonsense about chiplet and as much as 64MB cach3e.

Again... pls READ. or do you need me to bold where I said increase cache from 8-16MB? Or should I quote myself again from atr least 2 other times I have mentioned the cache increase?

In this second post I even bolded where I said it.

The PS5pro isnt just more CUs. Its more RDNA3.5 CUs, which means that each CU has VOPD, 2 AI units, 3rd gen RT cores.... all these things make that CU much better than what was in the OG PS5 RDNA 1.5 GPU, it will have an upclocked CPU (3.5Ghz > 4.4Ghz), will have more cache than what was in the OG PS5 which had 8Mb, so it would have maybe 12-16MB at least. And it does have more mem bandwidth because its using 16gbs mem chips instead of the 14 found in the OG PS5 taking memory bandwidth up from 448GB/s to 512GB/s. So yeah, there are improvements across the board. None of which warrants having a bigger die than 320mm2 on a 4/5nm node though.

Nope, they aren't doing any of that. It will be the same 8 channels and a very small uptick in cache. 64MB? Lmao. We would be lucky if they even take it up to 20/24MB. My guess puts it at 12-16MB... for reference the current PS5 has 8MB.

Also, you (people) really need to understand exactly what the PS5pro is fixing to do.

Chiplets only save you money depending on what size of chip you would otherwise have to make to begin with. The PS5pro, simply doesn't need that.

 

[MikeM]

The fact that AMD's utterly incompetent and generations behind Nvidia doesn't mean it won't be a dissapointment though. That statement doesn't inherently imply or suggest he believes it will be around a 4080 (it absolutely will not), it means it should and is technically possible.

Again, the 4080 will be over 2 years old and is actually a XX70 series die. It is objectively true that it will be a dissapointment if a 2 year newer and expected premium priced console cannot come close to a 2 year old 70 series GPU. That is like the minimum people should be expecting and demanding, and is historically what modern consoles have achieved (rough parity with the prior gen 70 series cards).

I wouldn’t go as far to say that AMD is incompetent. They still push far better frames per dollar and FSR isn’t the hot garbage ot used to be.

 

[Gaiff]

The fact that AMD's utterly incompetent and generations behind Nvidia doesn't mean it won't be a dissapointment though. That statement doesn't inherently imply or suggest he believes it will be around a 4080 (it absolutely will not), it means it should and is technically possible.

Again, the 4080 will be over 2 years old and is actually a XX70 series die. It is objectively true that it will be a dissapointment if a 2 year newer and expected premium priced console cannot come close to a 2 year old 70 series GPU. That is like the minimum people should be expecting and demanding, and is historically what modern consoles have achieved (rough parity with the prior gen 70 series cards).

And I'm sure AMD could do that if the TDP of the RTX 4080 was 215W like the 2070S...but it's rated at 320W, almost 50% higher. There's a reason new architectures take around 24 months to develop and not 12 months like at the beginning of the 8th generation of consoles. Look at how out of whack the performance scaling of the 4090 vs 4080 is compared to their SM count. Now do the same with the 4080 and 4070.

If Sony says "fuck that" and doesn't mind a huge, power hungry monster, then RTX 4080/7900 XTX in rasterization might be be feasible (at a large premium and I actually didn't check if they ever had an APU like that) but I somehow doubt that the Pro will be pulling 300W.

Never mind that Turing sacrificed die space for tensor and RT cores.