Digital Foundry Article Technical Discussion [2025]

[Cyan]

 

[DavidGraham]

0:00:48 News 1: Version of FSR 4 planned for PS5 Pro
0:13:46 News 2: RDNA 4 review reaction!
0:39:28 News 3: RDNA 4 exhibits mixed path tracing performance
0:49:23 News 4: Lossless Scaling introduces adaptive frame gen
1:01:01 News 5: Apple releases new Mac Studio… with last gen chip
1:09:36 News 6: Tony Hawk’s Pro Skater 3 + 4 remakes announced
1:18:20 News 7: Scalable Video Switch offers hassle-free analog gaming
1:30:04 Supporter Q1: What CPU should you pair with a 5090?
1:36:53 Supporter Q2: Should AMD or Nvidia release driver-level ML upscaling tech?
1:42:53 Supporter Q3: Why don’t PCs move towards unified RAM setups like Macs?
1:48:56 Supporter Q4: Is Nvidia going to neglect their gaming audience?
1:55:07 Supporter Q5: This gen, are we going to be able to build a console-like PC for a console-like price?

 

[Karamazov]

PSFSR4

 

[chris1515]

Mark Cerny: FSR 4 for PS5 Pro is the "next evolution of PSSR"

Last week, AMD released its new wave of graphics hardware, developed around the new RDNA 4 architecture. Machine learni…

www.eurogamer.net

Good news

 

[mr magoo]

i kinda died inside a little bit when i realised that i own GPU that Alex "deeply dislikes" ...

 

[DavidGraham]

0:39:28 News 3: RDNA 4 exhibits mixed path tracing performance

@Dictator One possible reason for the poor path tracing results is the shared nature of RDNA4's ray accelerators, they still share resources with shader cores, which also share resources with the machine learning cores. When everything is shared among 3 sections it becomes difficult to balance things properly when the workload is heavily favoring a certain section.

This design will be ok when the workload is balanced among raster/ray tracing/machine learning, but when ray tracing/machine learning is pushed hard (such as in cases of path tracing with machine learning denoising/upscaling) performance will take a huge hit, in these cases performance favors designs with dedicated cores for each workload.

 

[Cyan]

i kinda died inside a little bit when i realised that i own GPU that Alex "deeply dislikes" ...

oh man, I really like the RTX 4060, conceptually at least, I find it super efficient, but Alex sure has strong reasons to dislike it -maybe the VRAM dunno-.

On a different note, at that point in time they are talking about Lossless Scaling. It's kinda curious what that guy that left AMD to work on Intel said about the future being frame generation outputting super high framerates to achieve the max framerate of your monitor regardless of the base framerate, so knowing the framerate of your monitor would be enough for FG to do the rest.

In addition, Lossless Scaling now can also reduce the framerate to 0.4X, 0.5X, 0.3X etc, so you can run a game internally at 60fps and LS will run it at 24fps (40% of 60fps), or 30fps (1/2 or 50% of 60fps) or 18fps (60 x 0.3), etc etc . Dunno how useful that can be though...

 

[Charlietus]

Mark Cerny: FSR 4 for PS5 Pro is the "next evolution of PSSR"

Last week, AMD released its new wave of graphics hardware, developed around the new RDNA 4 architecture. Machine learni…

www.eurogamer.net

Good news

Sony deserves to get dragged even more for PS5 Pro. They are selling what is pretty much a early access product for an absurd price. That comment on how PS5 Pro was a "wonderful learning experience" tells you all you need to know

 

[mr magoo]

Sony deserves to get dragged even more for PS5 Pro. They are selling what is pretty much a early access product for an absurd price. That comment on how PS5 Pro was a "wonderful learning experience" tells you all you need to know

Thats how you get tech-junkies to spend their hard earned cash. Its brilliant bussines strategy

 

[Shifty Geezer]

Sony deserves to get dragged even more for PS5 Pro. They are selling what is pretty much a early access product for an absurd price. That comment on how PS5 Pro was a "wonderful learning experience" tells you all you need to know

He's said that about everything he's made, "we learned a lot", because he's a genuine engineer and not a corporate mouthpiece. You get any capable engineer and they'll be able to talk about what they learned from their last project, and what went wrong, and how they can do it better. Either they should never try anything new, or they should never learn from what did try, or, as many would do and as it seems you'd expect Sony to do, they should hide the reality of product development and just talk vacuous PR, contributing to the empty language of everything, devoid of truth and insight and just dressed up to extract the most positive sentiments from consumers.

Slightly bonkers that on B3D, the response to honest insight into the workings of complex graphics choices and developments is...commercial positioning. No interest at all in Sony's bespoke ML solution, how it can adapt with the potential for FSR4, how collaboration has informed progress in ML techniques and how that'll feed back not only into AMD but also Sony and other platforms.


Edit:

AMD started to work on AI FSR at the start of 2024 or so, as said by a AMD rapresentative. Apparently FSR 4 is the result of a year of collaboration with Sony, which is really fast for such good results.

So you see the value of it! I don't understand your position. Tech has to start somewhere. The first version of anything, Minimum Viable Product, is in essence an 'early access' or testcase for future tech. iPhone's Gorilla Glass screen was in essence a crappy early-access version of later, better screens that learnt from iterations of the technology. PSSR is a good solution where none existed, and from that greater solutions will come, and Cerny will talk honestly about it giving us a rare, honest insight. Everything about this is Good!

 

[mr magoo]

He's said that about everything he's made, "we learned a lot", because he's a genuine engineer and not a corporate mouthpiece. You get any capable engineer and they'll be able to talk about what they learned from their last project, and what went wrong, and how they can do it better. Either they should never try anything new, or they should never learn from what did try, or, as many would do and as it seems you'd expect Sony to do, they should hide the reality of product development and just talk vacuous PR, contributing to the empty language of everything, devoid of truth and insight and just dressed up to extract the most positive sentiments from consumers.

Slightly bonkers that on B3D, the response to honest insight into the workings of complex graphics choices and developments is...commercial positioning. No interest at all in Sony's bespoke ML solution, how it can adapt with the potential for FSR4, how collaboration has informed progress in ML techniques and how that'll feed back not only into AMD but also Sony and other platforms.


Edit:

So you see the value of it! I don't understand your position. Tech has to start somewhere. The first version of anything, Minimum Viable Product, is in essence an 'early access' or testcase for future tech. iPhone's Gorilla Glass screen was in essence a crappy early-access version of later, better screens that learnt from iterations of the technology. PSSR is a good solution where none existed, and from that greater solutions will come, and Cerny will talk honestly about it giving us a rare, honest insight. Everything about this is Good!

Why? I think this is a valid point thou, nobody is dissmissing or neglecting the impact of collab Sony,MSFT or whoever has on future of the gfx. Progress is progress and progress is good! Questioning the product and value to end user consumer is still valid question or opinion. I am happy they learned a lot, and happy that it helped to shape fssr4 or whatever they have in pipeline for future imrpvement, now do i want to be part of theirs QA and for what price is still valid question.

 

[Charlietus]

He's said that about everything he's made, "we learned a lot", because he's a genuine engineer and not a corporate mouthpiece. You get any capable engineer and they'll be able to talk about what they learned from their last project, and what went wrong, and how they can do it better. Either they should never try anything new, or they should never learn from what did try, or, as many would do and as it seems you'd expect Sony to do, they should hide the reality of product development and just talk vacuous PR, contributing to the empty language of everything, devoid of truth and insight and just dressed up to extract the most positive sentiments from consumers.

Slightly bonkers that on B3D, the response to honest insight into the workings of complex graphics choices and developments is...commercial positioning. No interest at all in Sony's bespoke ML solution, how it can adapt with the potential for FSR4, how collaboration has informed progress in ML techniques and how that'll feed back not only into AMD but also Sony and other platforms.


Edit:

So you see the value of it! I don't understand your position. Tech has to start somewhere. The first version of anything, Minimum Viable Product, is in essence an 'early access' or testcase for future tech. iPhone's Gorilla Glass screen was in essence a crappy early-access version of later, better screens that learnt from iterations of the technology. PSSR is a good solution where none existed, and from that greater solutions will come, and Cerny will talk honestly about it giving us a rare, honest insight. Everything about this is Good!

I respect Cerny and his team, they are some of the brightest minds in his industry, and how he explains complicated technology in an easy to understand way is second to none.

The Amethyst collaboration is an amazingly positive thing for the industry, and I'm glad that PSSR exists as a first stepping stone going in to the future.

But still, PS5 Pro remains a bad product, with a bad price, that right now barely ever keeps the promise of quality mode at performance framerates. It's not Cerny's fault, he didn't decide the price and he is just doing his job with AMD, but a first draft upscaler will be judged for what it is, not for the efforts that it took to get there.

 

[techuse]

I'm curious to see how Sony will manage to implement something that's quite expensive to process for a 9070xt on the much slower Pro.

 

[Subtlesnake]

@Dictator One possible reason for the poor path tracing results is the shared nature of RDNA4's ray accelerators, they still share resources with shader cores, which also share resources with the machine learning cores. When everything is shared among 3 sections it becomes difficult to balance things properly when the workload is heavily favoring a certain section.

This design will be ok when the workload is balanced among raster/ray tracing/machine learning, but when ray tracing/machine learning is pushed hard (such as in cases of path tracing with machine learning denoising/upscaling) performance will take a huge hit, in these cases performance favors designs with dedicated cores for each workload.

In the case of DF's benchmarks, there was no upscaling and RR was disabled on the Nvidia side.

 

[Shifty Geezer]

Questioning the product and value to end user consumer is still valid question or opinion.

But still, PS5 Pro remains a bad product, with a bad price, that right now barely ever keeps the promise of quality mode at performance framerates.

The value of PS5Pro isn't on point for this thread, but even then, I don't think being an 'early access' implementation of a new tech affects its value, As I say, all new tech is early access from a certain POV. Tech has to start somewhere. If PSSR sucks, PS5Pro can be criticised for having a sucky upscaler, but the fact it's new and soon to be superseded by some other tech shouldn't inherently devalue the product.

In short, I think Cerny's comment about learning from PSSR completely tangential to the PS5Pro value discussion. There just shouldn't be a correlation between this tech interview and 5Pro's pricing.

 

[Shifty Geezer]

I'm curious to see how Sony will manage to implement something that's quite expensive to process for a 9070xt on the much slower Pro.

Indeed. They might not and Cerny is honest about that, but he's also optimistic in their ML hardware. Somewhere along the line, the numbers have matched up within a 'feasible' target. I can only conclude that FSR4 included forward-thinking for lower spec hardware, so side-by-side with Sony as it was developed, the limit of 300 Tops was being considered. Amethyst is also keeping an eye on scalability. I don't know how ML solutions scale down. Is there a simple "power ∝ quality" equation?

 

[Karamazov]

PS5/XsX were early access for RT effects on consoles by that logic.
Totally useless !

 

[techuse]

Indeed. They might not and Cerny is honest about that, but he's also optimistic in their ML hardware. Somewhere along the line, the numbers have matched up within a 'feasible' target. I can only conclude that FSR4 included forward-thinking for lower spec hardware, so side-by-side with Sony as it was developed, the limit of 300 Tops was being considered. Amethyst is also keeping an eye on scalability. I don't know how ML solutions scale down. Is there a simple "power ∝ quality" equation?

I wish I knew. In the DF interview some months back, Cerny said they could not reach their target utilization of the Pro hardware due to memory bottlenecks. I would think this more advanced model would only exacerbate that issue.

 

[Charlietus]

Looking at the Death stranding 2 trailer on my tv, the image quality is top notch. I wonder if they are using the new Guerrilla tech for reconstruction that was mentioned to DF even on a base PS5. Even hair strands don't look bad at all.

 

[mr magoo]

PS5/XsX were early access for RT effects on consoles by that logic.
Totally useless !

Well yes they were. Not only on consoles but on amd cards in general and we know that msft collabed with amd to get RT working on AMD cards, whats your point? Why this is totally useless? If you are buying consoles solely for RT effects its not the best bang for bucks. We were early adopters on RT on amd cards, however imo they have fair price/performance ratio in general.