• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

[Digital Foundry] Mark Cerny: FSR 4 'Re-Implemented' For PS5 Pro - Next-Gen PSSR Coming 2026

Raster improvements are a dead end at this point. I'm not sure at all how much silicon should be dedicated to beefing up GPUs moving forward. All improvements moving forward are going to come from AI upscaling and being able to natively render at the smallest possible resolution. The major advantage Sony has is that Cerny is intrinsically knowledgable about bottlenecks and finding the right balance the experience he's getting with PSSR and PS5 Pro will pay major dividends for PS6.

It was interesting when he mentioned they considered an NPU but found it wasn't as efficient. If they can find away around that...

I would find it very interesting if the PS6 has a CPU + GPU + NPU. It'll be interesting to see what they do without as many constraints from the base PS5.


Yeah it's pretty clear that PS6 is going to go heavy into AI upscaling, FrameGen, and Ray tracing as the biggest core features. I'm sure people will complain, but extra Raster isn't moving the needle nearly as much as those other features at this point.

Don't be surprised if PS6 only has 67% more compute than the PS5 Pro, just like PS5 Pro had over the base PS5. But the AI and RT components will be more significant than PS5 Pro's AI/RT solutions.

We're looking at a sub 30 TF PS6.
 
Last edited:
It seems that PSSR was litterally the prototype of FSR4. Sony started to develop PSSR alone then brought their experience with the AMD team and together they created FSR4 for the RDNA4 GPUs.

Also I assume AMD would have plenty more available GPUs to train the neural sets which is likely why Cerny knew they needeed AMD for this. Playstation alone can't fight against mighty Nvidia to train their AI upscaling. They'd always be well behind.
 
Last edited:

Mibu no ookami

Demoted Member® Pro™
You feel bad for me? You have to wait 1+ year for competent ai upscaler.

I don't have to wait for anything.

The PS5 Pro is the most powerful console on the market. It plays games better than the base units. I don't need to wait for anything nor do I need to drop a grand plus on PC and deal with headaches. I work in IT the last thing I want to do is tweak settings or configurations on a computer. I had a bad fan on my PC and I waited more than six months to replace it and used my Macbook instead. I probably have a bad ram module too. Doubt I ever replace it.

That the PS5 Pro is only going to get better is even MORE reason to own one. Not only are the experiences better now, but they're only going to get better.

And do you know how I know you know this? Because you keep trying to tell yourself that the PS6 is coming out in 2027 when everything we've seen (court documents) suggests it will come out in 2028 if not later (no incentive for Sony to release the PS6 too early).

I feel bad for you because your narrative is falling apart right before your eyes.
 
Last edited:

Bojji

Member
I don't have to wait for anything.

The PS5 Pro is the most powerful console on the market. It plays games better than the base units. I don't need to wait for anything nor do I need to drop a grand plus on PC and deal with headaches. I work in IT the last thing I want to do is tweak settings or configurations on a computer. I had a bad fan on my PC and I waited more than six months to replace it and used my Macbook instead. I probably have a bad ram module too. Doubt I ever replace it.

That the PS5 Pro is only going to get better is even MORE reason to own one. Not only are the experiences better now, but they're only going to get better.

And do you know how I know you know this? Because you keep trying to tell yourself that the PS6 is coming out in 2027 when everything we've seen (court documents) suggests it will come out in 2028 if not later (no incentive for Sony to release the PS6 too early).

I feel bad for you because your narrative is falling apart right before your eyes.

I don't give a fuck when PS6 drops. There are not many reasons to use my PS5 anyway, Death Stranding 2 in july, ghost of yotei (if it releases this year) and maybe GTA 6 (depending on how playable it is) in 2025/2026. PS6 or fixed PSSR won't change that Sony output is mediocre and I have PC for TP games.
 

winjer

Gold Member
Yeah it's pretty clear that PS6 is going to go heavy into AI upscaling, FrameGen, and Ray tracing as the biggest core features. I'm sure people will complain, but extra Raster isn't moving the needle nearly as much as those other features at this point.

Don't be surprised if PS6 only has 67% more compute than the PS5 Pro, just like PS5 Pro had over the base PS5. But the AI and RT components will be more significant than PS5 Pro's AI/RT solutions.

We're looking at a sub 30 TF PS6.

But consider that each CU on RDNA5 will have higher IPC and clocks.
I just switched from a 6800XT (RDNA2) to an 9070 (RDNA4).
The 6800XT is a 21TFLOP machine. While the 9070 is an 18TFLOP machine. 36 if we count VOPD, but we already know it's not real 36TFLOPS.
And still, the 9070 is faster, even on pure rasterization. And much faster in RT and ML.
So even if a PS6 only has 67% more CUs than the PS5 Pro. It might still be 4 times faster in compute. And in ML and RT.
 

KeplerL2

Member
Don't be surprised if PS6 only has 67% more compute than the PS5 Pro, just like PS5 Pro had over the base PS5.

maxresdefault.jpg
 

Elios83

Member
I don't give a fuck when PS6 drops. There are not many reasons to use my PS5 anyway, Death Stranding 2 in july, ghost of yotei (if it releases this year) and maybe GTA 6 (depending on how playable it is) in 2025/2026. PS6 or fixed PSSR won't change that Sony output is mediocre and I have PC for TP games.

Emm honestly the more you talk the more you're taking Ls left and right.

"Once nVidia releases the 5000 series the value offered by Pro compared to a PC will be destroyed just a few months after launch" -> nVidia fucks up completely with the 5000 series

"Sony and AMD co-developing FSR4 has no value and meaning for PlayStation" -> a few day later Cerny announces FSR4 will come to PS5 Pro

And now this gem..."I only won't be able to play DS2, GTA6 and Ghost of Yotei"....lol you know only the best games that will be released this year :messenger_tears_of_joy:

Maybe it's time to stop with the attitude and the obvious insecurity complex that leads you to try to downplay consoles at all costs.
 

Bojji

Member
Emm honestly the more you talk the more you're taking Ls left and right.

"Once nVidia releases the 5000 series the value offered by Pro compared to a PC will be destroyed just a few months after launch" -> nVidia fucks up completely with the 5000 series

"Sony and AMD co-developing FSR4 has no value and meaning for PlayStation" -> a few day later Cerny announces FSR4 will come to PS5 Pro

And now this gem..."I only won't be able to play DS2, GTA6 and Ghost of Yotei"....lol you know only the best games that will be released this year :messenger_tears_of_joy:

Maybe it's time to stop with the attitude and the obvious insecurity complex that leads you to try to downplay consoles at all costs.

I don't remember words like that about 5xxx series. But who knows, those cards at MSRP would be not too bad BUT they are not available (right now). I know for sure that value of Pro was weak compared to available GPUs when it launched.

"FSR4" that will land on Pro somewhere in 2026 you mean? We don't even know if this version will be often used by developers if it's more heavy than standard PSSR. They still use FSR1 in some games...

Oh, I will play those games on old ass PS5 that I own, it even has disc drive!


It means PS6 won't be a big update in raw power.
 
Last edited:

Bojji

Member
Because about RDNA5 we know very little. Even don't know how much CU's top RDNA5. Late thise year or early 2026 first info about RDNA5.
We only know, info for now it's SWC and Matrix Cores

I should have added "on paper". RDNA4 already (suprisingly) has nice IPC jump vs. older versions (that's why my calculations were wrong, this was the missing factor). So PS6 might look weak on paper but it will still deliver decent jump vs. PS5/Pro.
 

Nankatsu

Gold Member
I'm gonna wait a while to see AMD's position on this.

It's nice to see PSSR getting tweaked, but only for 2026, that gives it what, two more years before PS6?
 
Yeah it's pretty clear that PS6 is going to go heavy into AI upscaling, FrameGen, and Ray tracing as the biggest core features. I'm sure people will complain, but extra Raster isn't moving the needle nearly as much as those other features at this point.

Don't be surprised if PS6 only has 67% more compute than the PS5 Pro, just like PS5 Pro had over the base PS5. But the AI and RT components will be more significant than PS5 Pro's AI/RT solutions.

We're looking at a sub 30 TF PS6.
In terms of rasterization it will depend on what they can get in 3nm or 2nm for the end of 2027/2028 with a TDP of 230 watts if Sony stays with the same power consumption as the PS5 Pro for the PS6.

It's hard to say but maybe the PS6 in terms of rasterization will be around a 4080 but as Cerny said the biggest gains for the next gen will come from ML and RT.
 

John Wick

Member
It is clear that PS5 pro and PSSR will improve and evolve over the months as Cerny says, reaching the quality of FSR4 even though some said it was not possible.
I'm surprised you haven't realised there are many on here who think they know better than the actual engineers and programmers. It's usually the ones that have expensive overpriced GPU's..........
 

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
It was so smart by Cerny to wait for analysists to judge FSR4 before stating it was their API.

DF 5 days ago.


Cerny 5 hour ago. PS5 Pro 300 TOPs is without sparsity, Pro won't need sparsity for FSR4, which BTW we co-designed with AMD.


Why are they always FUDing and lying about anything Playstation, always imagining the worst, never even once assuming FSR4 was co-designed by Sony etc. Just watch the tone of their articles about anything Xbox, always assuming the box would overperform when it almost always disappoint. I remember one video about them fantasizing like fanboys about a 2025 mid-gen Xbox that would magically performs like a 2027 console... They are MS propagandists!

You would think they'd stop doing this after their FUD about variable profile speeds of the PS5 and the whole 9 TF hate campaign where they were wrong. But nope.
 

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
Raster improvements are a dead end at this point. I'm not sure at all how much silicon should be dedicated to beefing up GPUs moving forward. All improvements moving forward are going to come from AI upscaling and being able to natively render at the smallest possible resolution. The major advantage Sony has is that Cerny is intrinsically knowledgable about bottlenecks and finding the right balance the experience he's getting with PSSR and PS5 Pro will pay major dividends for PS6.

It was interesting when he mentioned they considered an NPU but found it wasn't as efficient. If they can find away around that...

I would find it very interesting if the PS6 has a CPU + GPU + NPU. It'll be interesting to see what they do without as many constraints from the base PS5.

Any thoughts on the PS6 not being over 40 TFs in 2028?
 

Zathalus

Member
You would think they'd stop doing this after their FUD about variable profile speeds of the PS5 and the whole 9 TF hate campaign where they were wrong. But nope.
The real FUD here appears to be that quote, it starts off with what DF said in the FSR 4 video, but the rest from the x4 less part is not something they claimed.
 

Mibu no ookami

Demoted Member® Pro™
Any thoughts on the PS6 not being over 40 TFs in 2028?

I couldn't care less how many TFs it is.

I think we get too caught up with specs. What I'm more interested in is the design philosophy and the limitations of bottlenecks in order to maximize the value of the system to deliver the best performance for the cost. I'm interested to see what approach they take to solve the problem of diminishing returns.
 

mckmas8808

Mckmaster uses MasterCard to buy Slave drives
I doubt it will even be over 30. High 20s would be my guess.

Bro.....LOL!!! The PS4 saw a 5x jump to PS5 (1.8TFs to 10TFs). No way the PS5-to-PS6 will be less than a 3x jump.

- The PS5 Pro has 64% more TFs than the PS5. Giving it 16.7 TFs.
- If the exact same 64% is added to the PS6 (over the PRO), that gives us 27 TFs.
- But clearly they'll be more efficient. If they are 40% more efficient, the PS6 would have the equivalent of 38 TFs of power.

That's how I'm getting to the PS6 having 40 TFs of PS5 equivalent raster power.
 

Lysandros

Member
Bro.....LOL!!! The PS4 saw a 5x jump to PS5 (1.8TFs to 10TFs). No way the PS5-to-PS6 will be less than a 3x jump.

- The PS5 Pro has 64% more TFs than the PS5. Giving it 16.7 TFs.
- If the exact same 64% is added to the PS6 (over the PRO), that gives us 27 TFs.
- But clearly they'll be more efficient. If they are 40% more efficient, the PS6 would have the equivalent of 38 TFs of power.

That's how I'm getting to the PS6 having 40 TFs of PS5 equivalent raster power.
What did this generation teach us about teraflops and power?...
 
Last edited:

omegasc

Member
Bro.....LOL!!! The PS4 saw a 5x jump to PS5 (1.8TFs to 10TFs). No way the PS5-to-PS6 will be less than a 3x jump.

- The PS5 Pro has 64% more TFs than the PS5. Giving it 16.7 TFs.
- If the exact same 64% is added to the PS6 (over the PRO), that gives us 27 TFs.
- But clearly they'll be more efficient. If they are 40% more efficient, the PS6 would have the equivalent of 38 TFs of power.

That's how I'm getting to the PS6 having 40 TFs of PS5 equivalent raster power.
I wonder what the power budget will be for next gen consoles.
 

viveks86

Member
Bro.....LOL!!! The PS4 saw a 5x jump to PS5 (1.8TFs to 10TFs). No way the PS5-to-PS6 will be less than a 3x jump.

- The PS5 Pro has 64% more TFs than the PS5. Giving it 16.7 TFs.
- If the exact same 64% is added to the PS6 (over the PRO), that gives us 27 TFs.
- But clearly they'll be more efficient. If they are 40% more efficient, the PS6 would have the equivalent of 38 TFs of power.

That's how I'm getting to the PS6 having 40 TFs of PS5 equivalent raster power.
That's the nature of current GPU scaling. There's a reason Nvidia 50 series wasn't that big a jump over 40 series. And it's not just because they are greedy (probably a factor though). The same factors are going to force Sony to keep the costs down. They will look for AI and RT gains to act as bigger multipliers instead.

The ps5 pro is not giving a 64% raw uplift, despite the raw teraflops. This is likely due to other bottlenecks, such as memory bandwidth. If that bottleneck is removed with a generational shift, there is already more to gain in terms of actual performance. In another 3 years, I just can't see them doing a bigger uplift than another 60-70% without blowing out the costs. My guess is only on the raw TFs, around 27 to 30. No idea about gains in raster efficiency until we have a better idea of UDNA and if that yields anything at all. Though I can't imagine it being 40%. I think those days are long gone as well.
 
Last edited:

HeWhoWalks

Gold Member
I couldn't care less how many TFs it is.
I really wish more thought like this. But, here we are.

Cerny said it best: teraflop numbers matter little. I’ve gone over it myself. In 2022, for example, I didn’t buy a 3090 Ti because it was a 41 teraflop card. It was because it was a considerable upgrade over the TITAN RTX in my rig and could afford me much more of everything (save for RAM) to work with.

Who cares if PS6 is 25 teraflops or 40? The games will, as usual, do the talking. We’re also seeing a rise in useful upscalers. Much more important.
 
Last edited:
2026? LoL! It´s just better to wait for PS6! PS5 Pro it´s nonsense. We need more games that push regular PS5 and, 5 years later, get rid of the cross gen stuff already.
 

Mibu no ookami

Demoted Member® Pro™
I really wish more thought like this. But, here we are.

Cerny said it best: teraflop numbers matter little. I’ve gone over it myself. In 2022, for example, I didn’t buy a 3090 Ti because it was a 41 teraflop card. It was because it was a considerable upgrade over the TITAN RTX in my rig and could afford me much more of everything (save for RAM) to work with.

Who cares if PS6 is 25 teraflops or 40? The games will, as usual, do the talking. We’re also seeing a rise in useful upscalers. Much more important.

It all happens in cycles

It use to be how many bit was the console. Now we're 16-bit! Now we're 32-bit! Now we're 64-bit! Now we're 128-bit... ummm this doesn't actually make sense...

Okay well, how many polygons are we pushed? Wait the polygon totals aren't relevant anymore?

Well now we're in HD. Okay, but what about 4K?

Okay, how many gigaflops? Oh wow... that many? How many teraflops?

And now we're yet at new measure with diminishing returns.

IQ matters more but people don't have a convenient way of measuring the quality of upscaling. There's no marketing terms that have been pushed to the masses. What happens with a 4K game is natively 1080p but looks worse than a game that is natively 920p because the 920p game uses a better upscaler? You see people nervous about this and you see them still bringing up native resolution, because it's all they know/understand.

Next generation, it's all going to be about raytracing and path tracing and upscaling. TOPS will probably be the most relevant measure, but even then misleading.
 

viveks86

Member
It all happens in cycles

It use to be how many bit was the console. Now we're 16-bit! Now we're 32-bit! Now we're 64-bit! Now we're 128-bit... ummm this doesn't actually make sense...

Okay well, how many polygons are we pushed? Wait the polygon totals aren't relevant anymore?

Well now we're in HD. Okay, but what about 4K?

Okay, how many gigaflops? Oh wow... that many? How many teraflops?

And now we're yet at new measure with diminishing returns.

IQ matters more but people don't have a convenient way of measuring the quality of upscaling. There's no marketing terms that have been pushed to the masses. What happens with a 4K game is natively 1080p but looks worse than a game that is natively 920p because the 920p game uses a better upscaler? You see people nervous about this and you see them still bringing up native resolution, because it's all they know/understand.

Next generation, it's all going to be about raytracing and path tracing and upscaling. TOPS will probably be the most relevant measure, but even then misleading.
I remember Jensen tried to define a metric when they originally introduced RTX, called RTX-OPS, which combined raster, AI and ray tracing to give a single number. It was mostly marketing spin to downplay the non-RT performance uplift, so it never caught on and nobody cared for it. They seemed to have dropped the idea entirely after that generation. I do wish there was one number that could give us a clear representation of the hardware's capability.

It's a bit ironic that we are actually becoming more qualitative and less quantitative with advancements in graphics. Like... we can talk about TFs and AI TOPs and RT intersections per second, but there is no sense of what that translates to. When 3Dmark used to be a great way to compare cards, the manufacturers found ways to tune and cheat the benchmarks, so that couldn't be used either. Now we need to benchmark a dozen games to gauge relative performance, but even then it's not clear as the final output muddies the water with different upscalers and image quality :/
 
Top Bottom