Digital Foundry Article Technical Discussion [2024]

[LiveGamer]

I haven't thought about Too Human for a long time! Are there any good write ups on that development drama?

At the time, I always thought it was a bit odd, given there were plenty of UE3 titles that seemed just fine. Mass Effect was a out a year earlier, to pick something comparable.

Yeah I had forgotten about them too! That was a crazy time in the game industry. I'm not sure if there's a good writeup out there but here's a Google synopsis:




Essentially Denis (and other developers) at the time publically complained how poorly supported the engine was. So the engine was causing huge delays for games meanwhile EPIC was developing the huge hit Gears of War. It was so bad that Silicon Knights used the incomplete UE3 code as the foundation to create their own engine. Denis said their contract allowed them to be able to do that. EPIC disagreed.

Mass Effect had far higher resources to brute force any issues.

UE3 did wind up good but it was a long road to get there.

 

[Karamazov]

Mod edit accompanying article:

PS5 Pro's PSSR upscaler tested against FSR 3.1 and Nvidia DLSS 3.7 in Ratchet and Clank: Rift Apart

Sony's PS5 Pro PSSR vs FSR 3.1 vs DLSS 3.7. Digital Foundry puts Sony's upscaler through its paces.

www.eurogamer.net

Finally - and perhaps the biggest caveat of all - this is just one game and the quality of upscaling implementations vary from title to title. We've not seen enough of PSSR to draw more solid conclusions in the way we can with FSR and DLSS, but there's a good chance that what we're seeing here with PSSR may be indicative. For now though, it's very much just a Ratchet and Clank head-to-head.

 

[snc]

Very good results especialy taking into account its just first iteration of pssr

 

[homerdog]

Mod edit accompanying article:

PS5 Pro's PSSR upscaler tested against FSR 3.1 and Nvidia DLSS 3.7 in Ratchet and Clank: Rift Apart

Sony's PS5 Pro PSSR vs FSR 3.1 vs DLSS 3.7. Digital Foundry puts Sony's upscaler through its paces.

www.eurogamer.net

I've been waiting for this video

 

[Scott_Arm]

PSSR looks great. Really well done for a first version. It'll only get better. I imagine future games will have newer versions with nice improvements. Great offering from Sony.

 

[Charlietus]

Those videos are really important as feedback for the developers of PSSR. If they just launched this with PS6, maybe they wouldn't have catched some things.

 

[Remij]

Very VERY impressive in Ratchet and Clank. Though I think maybe it would have been better to just compare it to Insomniac's IGTI instead of FSR. The game never used FSR on console, and let's be honest, FSR just does not compete with the big boys.

 

[QPlayer]

and let's be honest, FSR just does not compete with the big boys.

And then here comes this:

 

[Globalisateur]

Very VERY impressive in Ratchet and Clank. Though I think maybe it would have been better to just compare it to Insomniac's IGTI instead of FSR. The game never used FSR on console, and let's be honest, FSR just does not compete with the big boys.

yes but It's great to finally see how bad FSR really is compared to the others solution. But it's bad versus everything really as long as stuff are in motion + alphas anyways. It's garbage, worse than anything we had before, including quincux, and it should have never existed.

The feedback from those videos are really important as feedback for the developers of PSSR. If they just launched this with PS6, maybe they wouldn't have catched some things.

For the same reason they released PS4 Pro with the first ever smart reconstruction upscaling using dedicated hardware. Before DLSS was ever a thing. Many developers started to have fun with motion vectors and custom hardware because of CBR. Now applying PSSR (and DLSS) is much easier for them as they already have a production ready pipeline with motion vectors and such. They released PS4 Pro to prevent people going the PC route but also to test et get developers used to new upscaling technologies. Same thing with PS5 Pro.

 

[Remij]

I think I may have noticed a new bug with the PC version of Ratchet and Clank while watching Alex's newest video. At about 13m56s in the video, while Alex is speaking about the somewhat "laggy" reflections with PSSR, I noticed that on the PS5 version, Clank's vents on his body will animate as they open and close randomly, his arms will also extend and animate as he waves to the crowds. On the PC version, in this video you never see either happen on the PC side even once. I think that animation may either be bugged, or maybe it's just a coincidence in the footage that he captured.

You can see it here.

I don't have the game installed, and honestly it's not a big enough deal for me to want to reinstall the game and check myself, but if someone who does have the game installed could check it out and see if it happens or not, that would be cool.

 

[QPlayer]

yes but It's great to finally see how bad FSR really is compared to the others solution. But it's bad versus everything really as long as stuff are in motion + alphas anyways. It's garbage, worse than anything we had before, including quincux, and it should have never existed.

For the same reason they released PS4 Pro with the first ever smart reconstruction upscaling using dedicated hardware. Before DLSS was ever a thing. Many developers started to have fun with motion vectors and custom hardware because of CBR. Now applying PSSR (and DLSS) is much easier for them as they already have a production ready pipeline with motion vectors and such. They released PS4 Pro to prevent people going the PC route but also to test et get developers used to new upscaling technologies. Same thing with PS5 Pro.

Watch the video I linked earlier and your opinion will change. Practice shows that well-applied FSR is a very effective technique. Of course, in 1440p native need the best image quality.

 

[davis.anthony]

yes but It's great to finally see how bad FSR really is compared to the others solution.

This isn't the first time that has been shown, we've seen and known how bad FSR is for a few years now

 

[pjbliverpool]

I think I may have noticed a new bug with the PC version of Ratchet and Clank while watching Alex's newest video. At about 13m56s in the video, while Alex is speaking about the somewhat "laggy" reflections with PSSR, I noticed that on the PS5 version, Clank's vents on his body will animate as they open and close randomly, his arms will also extend and animate as he waves to the crowds. On the PC version, in this video you never see either happen on the PC side even once. I think that animation may either be bugged, or maybe it's just a coincidence in the footage that he captured.

Haha well spotted. My kids were watching the video while I was, treating it as a "spot the difference" and my 5yo noticed this which I totally missed.

 

[Globalisateur]

The RT reflections have also been significantly improved compared to PS5. It was already very noticeable on the Spider-man 2 footage BTW. Big difference there actually comparing PS5 Pro vs PS5 quality 30fps in 2 fast past scenes.

They are almost as good looking as the full resolution reflections on PC and look even...better in motion somehow? Sony is aiming at something different here, clearly.

Also while Alex was talking (rightly) about the more instable aliasing in motion on Pro side the most noticeable stuff for me without zooming was the aliasing on the PC side in still images. Great job by Alex about his work and the very closely matched comparisons. Overall PSSR is already super close to DLSS 3.7 in that game. The closer than any other technology from Intel or AMD (LOL)!

It's like the AI upscaling implementations are quite different in that game. Plenty of aliasing in stills with DLSS and plenty of aliasing in motion with PSSR. Pick your poison! I found those GIFs on GAF:

 

[iroboto]

Mod edit accompanying article:

PS5 Pro's PSSR upscaler tested against FSR 3.1 and Nvidia DLSS 3.7 in Ratchet and Clank: Rift Apart

Sony's PS5 Pro PSSR vs FSR 3.1 vs DLSS 3.7. Digital Foundry puts Sony's upscaler through its paces.

www.eurogamer.net

Great first showing from PSSR here, and quite ideal I think. Proof that you can run without dedicated tensor cores as an effective cost savings compromise here for consoles in the future. If somehow frame generation is also added to this mix (which I suspect, thinking out loud, will be a touch harder without dedicated units - on the flip side FSR FG improves sufficiently), then we have a pretty strong direction of where console hardware is headed in the future.

We will need to see true multiplatform comparisons however to get a real idea of what's happening here. Exclusive and marquee titles tends to contain significant optimizations for the platform that may not be available on PC. But otherwise, all things equal, great showing and really good to see an answer exists for the non-dedicated silicon route.\

I think there was a small point in time where we did see DLSS also go this route, IIRC, where it skips the tensor cores. It may really just come down to putting in the investment to create the model.

 

[Scott_Arm]

Great first showing from PSSR here, and quite ideal I think. Proof that you can run without dedicated tensor cores as an effective cost savings compromise here for consoles in the future. If somehow frame generation is also added to this mix (which I suspect, thinking out loud, will be a touch harder without dedicated units - on the flip side FSR FG improves sufficiently), then we have a pretty strong direction of where console hardware is headed in the future.

We will need to see true multiplatform comparisons however to get a real idea of what's happening here. Exclusive and marquee titles tends to contain significant optimizations for the platform that may not be available on PC. But otherwise, all things equal, great showing and really good to see an answer exists for the non-dedicated silicon route.\

I think there was a small point in time where we did see DLSS also go this route, IIRC, where it skips the tensor cores. It may really just come down to putting in the investment to create the model.

I didn't follow the ps5 pro specs closely, but I'm pretty sure they added ML hardware for running PSSR. Not sure if it's a separate unit or just more cpu instructions/capability.

 

[Kaotik]

I think there was a small point in time where we did see DLSS also go this route, IIRC, where it skips the tensor cores. It may really just come down to putting in the investment to create the model.

It already did once with DLSS "1.9" in Control. Not sure if there was ever any performance comparisons, but apparently it was still worth it without matrix crunchers

I didn't follow the ps5 pro specs closely, but I'm pretty sure they added ML hardware for running PSSR. Not sure if it's a separate unit or just more cpu instructions/capability.

It was rumored, but even rumors didn't say if it would be something like XDNA or something like matrix units in CDNA or what, and Sony hasn't said it has or doesn't have 'em.
Edit: apparently Cerny did confirm but not what type

 

[Dictator]

The RT reflections have also been significantly improved compared to PS5. It was already very noticeable on the Spider-man 2 footage BTW. Big difference there actually comparing PS5 Pro vs PS5 quality 30fps in 2 fast past scenes.

They are almost as good looking as the full resolution reflections on PC and look even...better in motion somehow? Sony is aiming at something different here, clearly.

Also while Alex was talking (rightly) about the more instable aliasing in motion on Pro side the most noticeable stuff for me without zooming was the aliasing on the PC side in still images. Great job by Alex about his work and the very closely matched comparisons. Overall PSSR is already super close to DLSS 3.7 in that game. The closer than any other technology from Intel or AMD (LOL)!

It's like the AI upscaling implementations are quite different in that game. Plenty of aliasing in stills with DLSS and plenty of aliasing in motion with PSSR. Pick your poison! I found those GIFs on GAF:

I think the reflections thing that is really evident between the versions is that the sample position/jitter has been tweaked to match PSSR by insomniac, something that Nixxes did not do for the upscalers on PC. Insomniac told Oliver that they had to change them to make PSSR look better.

 

[Shifty Geezer]

I didn't follow the ps5 pro specs closely, but I'm pretty sure they added ML hardware for running PSSR. Not sure if it's a separate unit or just more cpu instructions/capability.

Cerny called it ML hardware of some sort, but the juries still out on whether that was adding some form of ML hardware or just using what AMD provide to MLise on the compute units.

 

[iroboto]

I didn't follow the ps5 pro specs closely, but I'm pretty sure they added ML hardware for running PSSR. Not sure if it's a separate unit or just more cpu instructions/capability.

IMO, there is sufficient evidence that the ML hardware is support for sparsity formats and supporting dual issue on the CUs for this type of function, one can also assume this includes dp4a support. Using the leaks provided, we can quickly math out how they came to 300 TOPs using the above customizations. If they had tensor TOPs, it would be magnitudes faster. Tensor silicon are type of hardware you want for training models, but aren't necessary for running models - otherwise we wouldn't have any form of AI on IoT devices.

To be clear, I'm not trying to downplay the 5Pro here, it's not a discussion of haves and have nots. But imo, it's actually quite advantageous here if it weren't running dedicated silicon. There will always be tensor cores on nvidia hardware going forward, but that is a premium every nvidia gamer must pay, as they are sharing the market with the AI industry. By keeping dedicated silicon out, you can reduce the price for those just wanting to use the GPU to only play games.