With PlayStation Spectral Super Resolution (PSSR), Sony has joined the machine learning-based image reconstruction race that Nvidia started almost six years ago. At Digital Foundry, we’re excited to see this kind of technology come to the console space, so for our latest exclusive PS5 Pro photo capture opportunity, we spent some time using the PS5 Pro and Ratchet and Clank: Rift Apart to get a series of shots based on reconstruction, which we know from previous experience, really tests reconstruction technologies. To what extent can PSSR be superior to the FSR-level technology typically used on consoles? How close is PSSR to the established state-of-the-art solution: Nvidia DLSS?
First of all, let’s discuss the caveats and limitations. We compared PS5 Pro to PC at approximate quality settings, but it’s impossible to get a complete match. Secondly, Rift Apart uses dynamic resolution scaling, and the implementation on PC is very, very different – so we counted the pixels of every single test photo on Pro, and then fine-tuned the resolution on DLSS using a tool called DLSSTweaks, while limiting ourselves to FSR 3.1 quality mode, which typically uses a slightly lower base resolution than Pro mode. Additionally, while we tried turning off motion blur on PS5 Pro to increase clarity, turning it off only seemed to reduce the intensity of the effect, so we had to match it to our desktop footage.
Finally – and perhaps the biggest caveat of all – this is just one game, and the quality of scaling implementation varies from title to title. We haven’t seen enough of PSSR to draw more robust conclusions in the way we can with FSR and DLSS, but there’s a good chance that what we see here with PSSR may be indicative. For now, though, it’s largely a Ratchet and Clank showdown.
Sony’s PS5 Pro upscaler module – PSSR – is undergoing testing against FSR 3.1 (the technology it will most likely replace) alongside Nvidia’s DLSS, currently an established state-of-the-art ML upscaler. Watch on YouTube
I highly recommend watching the video above, as the very nature of the exercise means that video is a better means of conveying the differences, and as you’ll see when you watch, the first thing that had to be done was comparisons with FSR 3.1. AMD’s compute-based upscaler feature is heavily used in the console space, and in many scenarios it is the technology that PSSR aims to replace. I started by looking for “disocclusion.” A third-person game like Rift Apart is a great place to test this: it’s basically about the quality of newly revealed details that have little or no “story” from previous frames.
This is my main complaint about the quality of FSR 3.1 – it does not effectively smooth out newly revealed details that “get stuck” in motion. Fortunately, like DLSS and XeSS, PSSR does not suffer from this problem – and it is plain as day. Ratchet and Clank runs at relatively high resolutions on PS5 Pro, so this quality difference may become more pronounced at much lower input pixel ranges – we’ll test this in due course.
FSR also has issues with particles, whether it’s particles when Ratchet collects his first gun or confetti flying in the background during the opening phase. The particles have a subtle, ghost-like appearance when processed with FSR, to the point that confetti can appear and disappear. PSSR does a relatively better job and seems to contain more confetti. In reality, this is not the case: PSSR particles are simply more time-coherent and remain on-screen, which is a stark contrast to FSR 3.1. Hologram signs also look better with PSSR. FSR 3.1 lacks anti-aliasing, and while PSSR isn’t perfect, the image is less jagged around the edges.
Manage cookie settings
Finally, processing the grass in the first stage challenges the FSR to the point that it can apparently create liquid artifacts that look almost like liquid. Machine learning scalers tend to discriminate better, and PSSR tries to address this problem. It’s still not perfect, but I’d say it manages to keep the grass from looking as weird as it does with FSR. Given that this is the first publicly visible iteration of PSSR, we can safely assume that it manages to do the job required of it – it offers significant improvements in scaling quality, but keep in mind that we’re still dealing with relatively high base resolutions. It’s at 1080p – and even lower – that PSSR really needs to prove its worth.
Comparisons with Nvidia DLSS may be less practical. Unlike FSR 3.1, you’ll never see a Sony console using this scaling technology. However, when viewed more holistically, DLSS retains its crown as the best machine learning-based scaling tool on the market – a position that Nvidia has earned through nearly six years of iterations – so we can have some idea of how close Sony came with its debut iteration. I noticed a sort of moving hissing sound in all of my test clips. I’m not entirely sure what it is – perhaps some kind of meta instability – but it’s visible in every image of the game I’ve looked at so far. To be honest, this is not in FSR 3.1 either, I also found that in some areas DLSS resolves moving geometry with less aliasing.
This is also confirmed by comparing a fast moving object. PSSR performs better in traffic than FSR, but in the case of DLSS it also does not anti-alias. Another thing you’ll notice about PSSR is that it produces a softer image than DLSS. Nvidia’s scaling tool no longer has a post-process sharpening tool built into the model, instead games only add it when they want it. Even with post-process sharpening turned off, I think we can see the PSSR resolution become noticeably smaller. This could be seen as a disadvantage, but I think it’s definitely better than cases where Ratchet’s end-game sharpening value is maxed out. Of course, the resolution of the reconstruction technique is essentially subjective and a matter of personal taste.
Comparing with FSR 3.1 and DLSS is one thing, but how about PS5 Pro PSSR compared to the standard IGTI upscaling module in the PS5 – a topic we covered here. Watch on YouTube
On a less subjective note, there’s something interesting I noticed – ray tracing ghosting in the PS5 Pro version of Ratchet and Clank, visible in RT reflections. Combining ray tracing with image reconstruction is complex, and the noise pattern introduced into the reconstruction can have a huge impact on the output image. Insomniac says the sampling pattern is adjusted to PSSR to better connect with it. Even if motion reflections have negative aspects, improved stability may occur in some aspects. For example, at lower quality settings, RT checkerboard reflections clearly show checkerboard artifacts using DLSS. PSSR combines the checkerboard correctly, providing higher resolution.
Beyond these differences, we’ll need more time with PSSR on this – and other – titles to make a more final verdict. Today’s testing shows that there are areas where PSSR is significantly better than FSR 3.1, which is the technology it most often replaces and is certainly preferred. Assuming these differences translate to other games, PS5 Pro owners may appreciate much better scaling quality. Stability issues are something to keep an eye out for in the future to see if they show up in other games, and if not, whether they’ll be fixed in future iterations of PSSR.
But perhaps games like Alan Wake 2 will prove their worth best: an internal 864p resolution in 60fps performance mode, upscaled to full 4K, is a tall order for the technology – and we’re looking forward to testing it out alongside other titles that may pushed FSR technology too far on standard PS5 and Xbox consoles. The November 7 PS5 Pro release date is getting closer and closer, and we’ll have more information as soon as possible.