Just 9 FPS, 54 to 63 moving all the way down from Quality to Performance, from FSR 3.1 to FSR 4.1, with frame-gen being enabled on both (base frame rate around 30, meaning the actual base FPS improvement is probably more like 5 before FG almost doubles the output FPS) is pretty nuts TBH. I would have hoped they could at least use a switch from Quality down to Balanced due to the heavier model on compatibility mode, while holding the same frame rate. I'm guessing FSR 4.1 Balanced is still a performance regression from FSR 3.1 Quality on pre-RDNA 4 hardware. Unfortunate.

RDNA 1-2-3-3.5 really paying for not having that RDNA 4 AI hardware FSR 4.x needs to run natively it seems. At least it can run, which wasn't the case not long ago.

For reference, the internal resolution is 50% that of native with Performance upscaling, for a 2x scale factor, whereas with Quality you're at 67% and 1.5x. Balanced is 59% 1.7x. UItra Quality is 77% 1.3x. Quality generally has a 20-30% performance bonus, with Performance being up to 70% bonus FPS due to literally rendering half of native. I'm honestly shocked that the raw FPS before frame gen on FSR 4.1 Performance is in the ~30 FPS range on hardware as on-paper powerful as an XSX. That's 540p internal resolution assuming output is 1080p, and it's only just holding ~30 FPS. Damn.

Can definitely see why that lighter model is required for handhelds. Lets hope that even with a cut down model the IQ improvements are still as notable as the jump from (typically not worth using as it's impossible to ignore the IQ issues) FSR 1-3.x to FSR 4.x (worth using and on par with DLSS 3.x). The Switch 2 had a pretty good result even with cut-down DLSS on obsolete handheld hardware, so in theory, it's possible. Fingers crossed. If not, at least there's still Intel XeSS, which even in compatibility mode is still much better than FSR 1-3.x, and works pretty well on those handhelds.

RDNA 4/5 handheld/console chips can't come soon enough, even if the raw raster is barely improved, these upscalers are essential in modern games, and struggling to hit 30 FPS before FG with the upscaler almost as low as it can go, internal resolution wise, isn't good.

Hey, at least MS/Sony are trying to eke out every last bit of oomph these consoles have. Kudos there.

Interesting given Sony restricted to a new Pro console?

From the pictures that are shown, I can conclude comparing apples against oranges ain't a correct way to do.

Hopefully they'll implement it correctly and do a better comparison in the future. I'm no console user anymore, but still interested in the market.

chrcolukInteresting given Sony restricted to a new Pro console?

Xbox Series is usual RDNA 2. Meanwhile the original PS5 is an hybrid of RDNA 1 with some features of RDNA 2, but more towards RDNA 1.

@dgianstefani
Did you expect anything different? FSR 4.x was known to be heavier than FSR 3.x and would therefore yield lower performance on RDNA2 hardware. The question was simply whether FSR 4.x on performance/balanced settings would offer a better, more stable image compared to FSR 3.x on quality. And that was the whole point of releasing FSR 4.x for RDNA2 cards: to give the user more options—options that might offer image quality improvements that FSR 3.x couldn't deliver. In no way were we expecting higher performance out of RDNA2.

john_@dgianstefani
Did you expect anything different? FSR 4.x was known to be heavier than FSR 3.x and would therefore yield lower performance on RDNA2 hardware. The question was simply whether FSR 4.x on performance/balanced settings would offer a better, more stable image compared to FSR 3.x on quality. And that was the whole point of releasing FSR 4.x for RDNA2 cards: to give the user more options—options that might offer image quality improvements that FSR 3.x couldn't deliver. In no way were we expecting higher performance out of RDNA2.

I expected going down one upscaling quality preset to maintain similar/same FPS would be the compromise, with image quality seeing an improvement from the long overdue decent FSR implementation (4.x). Similar to how DLSS Transformer at Balanced, sometimes even at Performance still looks better than DLSS CNN at Quality. You get better results, similar FPS, and one step lower level upscaling is needed to achieve that similar FPS.

As has been said, this isn't tested on the base PS5 with its RDNA 1.5 esque GPU, this is full RDNA 2 on a fairly powerful config on the series X. Sure, it's not a PS5 Pro with a newer arch, but two quality levels? Damn.

While we knew it's a heavy model especially if the dedicated hardware isn't present (why wasnt it AMD/Xbox/Sony, you had both Turing and Ampere to learn from by the time the console chips shipped), I am quite surprised that the move from Quality to Performance is necessary to end up roughly at the same output FPS (considering the 9 improvement is after frame gen, the actual improvement is roughly half that). A whole 5 FPS gained at ~25 vs 30 base FPS while jumping down two quality settings is pretty nuts to me.

Anyway. Glad I've not really recommended any console besides the Switch 2 and PS5 Pro when I've been asked by friends what to pick if not going PC. Besides being roughly 2x the performance GPU wise of the base PS5, not being stuck in RDNA 1/2 hell is worth the premium.

dgianstefaniI expected going down one upscaling quality preset to maintain similar/same FPS would be the compromise, with image quality seeing an improvement from the long overdue decent FSR implementation (4.x). Similar to how DLSS Transformer at Balanced, sometimes even at Performance still looks better than DLSS CNN at Quality. You get better results, similar FPS, and one step lower level upscaling is needed to achieve that similar FPS.

As has been said, this isn't tested on the base PS5 with its RDNA 1.5 esque GPU, this is full RDNA 2 on a fairly powerful config on the series X. Sure, it's not a PS5 Pro with a newer arch, but two quality levels? Damn.

While we knew it's a heavy model especially if the dedicated hardware isn't present (why wasnt it AMD/Xbox/Sony, you had both Turing and Ampere to learn from by the time the console chips shipped), I am quite surprised that the move from Quality to Performance is necessary to end up roughly at the same output FPS (considering the 9 improvement is after frame gen, the actual improvement is roughly half that). A whole 5 FPS gained at ~25 vs 30 base FPS while jumping down two quality settings is pretty nuts to me.

Anyway. Glad I've not really recommended any console besides the Switch 2 and PS5 Pro when I've been asked by friends what to pick if not going PC.

That's likely true. FSR 4.x Balanced might give similar performance to FSR 3.x Quality. But don't forget we are talking about marketing slides/videos, which means you have to present bigger numbers to the public. And since 7(?) out of 10 people—especially console users—don't know what FSR quality/balanced/performance is, you use performance figures in your presentation. Now, regarding DLSS, it has the advantage of running on RTX hardware, which means that generally speaking, the hardware is available, even on RTX 20 cards. RDNA 2 lacks dedicated hardware, so it inherently requires more work in terms of compatibility and has a higher chance of underperforming.

Now, concerning the absence of corresponding hardware in RDNA 1/2 implementations that existed in Nvidia's GPUs, I imagine Nvidia didn't design that hardware in a month—let alone test it and build software and drivers for it. AMD would needed even more time, being something new to them. And of course, additional hardware means extra cost and potentially higher power consumption. With consoles being 6 years on the market already, the fact that the need for such hardware surfaced likely somewhere in their 4th, 5th, or 6th year of life suggests that choosing not to complicate the architecture with extra hardware—which would have increased both complexity and cost—was probably the right call.

Going from 25 fps to 30 fps is a 20% increase. Sounds good to me, maybe even better than good.

As for the last point, since you're a PC user who is deeply immersed in technology and its news, and since you're a PC enthusiast who finds themselves caring about things like "what technology is being used here", I think the wisest choice is to go for hardware that won't feel obsolete to you, because psychologically it won't sit well with you. It's like buying an RX 6950 today. It might still perform very well in games, but the moment you get it in your hands, you've already built a mental list of everything it can't do. This is a classic case of "ignorance is bliss".

john_As for the last point, since you're a PC user who is deeply immersed in technology and its news, and since you're a PC enthusiast who finds themselves caring about things like "what technology is being used here", I think the wisest choice is to go for hardware that won't feel obsolete to you, because psychologically it won't sit well with you. It's like buying an RX 6950 today. It might still perform very well in games, but the moment you get it in your hands, you've already built a mental list of everything it can't do. This is a classic case of "ignorance is bliss".

Agreed. But relying on your customers to be ignorant is not a good long term strategy, for console or GPU makers. Seems to be backfiring considering the state of general opinions regarding consoles, even amongst "normies", per se. I also think it shouldn't be necessary for AMD to see NVIDIA do something first, before starting work on their version. If NVIDIA can figure out a tech can be used in a beneficial way, and start R&D from that analysis, AMD can certainly do that on their own, too.

dgianstefaniAgreed. But relying on your customers to be ignorant is not a good long term strategy, for console or GPU makers. Seems to be backfiring considering the state of general opinions on consoles, even amongst "normies", per se. I also think it shouldn't be necessary for AMD to see NVIDIA do something first, before starting work on their version. If NVIDIA can figure out a tech can be used in a beneficial way, and start R&D from that analysis, AMD can certainly do that on their own, too.

AMD can rarely push new technologies on its own. Granted, where it did succeed (x86-64, Mantle), those achievements helped it tremendously to survive or remain competitive. On top of that, they're foolish. They seem to have solved the shader compilation stutter issue in their drivers, yet they didn't issue half a press release—whereas if Nvidia were in their shoes, they would have flooded the internet with press releases. Fortunately, they have Sony primarily and MS secondarily pushing them, because if they keep making stupid mistakes there, the next-next generations of PlayStation and Xbox will feature hardware from another manufacturer—be it Nvidia or Intel (I could mention Qualcomm here too, but they are way too far from being a serious competitor to Intel, AMD, or Nvidia in gaming hardware of any form factor that doesn't fit in a pocket).

john_AMD can rarely push new technologies on its own. Granted, where it did succeed (x86-64, Mantle), those achievements helped it tremendously to survive or remain competitive. On top of that, they're foolish.

Agreed. I mean they have evidence that actual innovation sells. Yet here we are, year after year, either too little, or too late. Hence my point, "at least copy fast if you're not going to lead".

john_They seem to have solved the shader compilation stutter issue in their drivers, yet they didn't issue half a press release—whereas if Nvidia were in their shoes, they would have flooded the internet with press releases.

The shoot-in-foot AMD marketing at work. Not to be harsh, but sometimes I wonder if they even talk to the engineering dept or if they've blocked each other's numbers.

john_Fortunately, they have Sony primarily and MS secondarily pushing them, because if they keep making stupid mistakes there, the next-next generations of PlayStation and Xbox will feature hardware from another manufacturer—be it Nvidia or Intel (I could mention Qualcomm here too, but they are way too far from being a serious competitor to Intel, AMD, or Nvidia in gaming hardware of any form factor that doesn't fit in a pocket).

I wouldn't be surprised. Honestly. Not surprised at all if eventually consoles break away from AMD hardware. I could understand why Intel bothered to make a bid for the consoles, their tech and upscaler is rather excellent. Even the Switch 2 with its multiple gen old NVIDIA IP/Arm cores, ancient process, and cut down DLSS is extremely compelling, besides Nintend-no's business practices.

dgianstefaniI wouldn't be surprised. Honestly. Not surprised at all if eventually consoles break away from AMD hardware. I could understand why Intel bothered to make a bid for the consoles, their tech and upscaler is rather excellent. Even the Switch 2 with its multiple gen old NVIDIA IP/Arm cores, ancient process, and cut down DLSS is extremely compelling, besides Nintend-no's business practices.

The truth is, I expected MS to switch to Nvidia starting as early as this generation. Yes, more expensive hardware, and any incompatibility arising from the transition from x86 to ARM might cost them, but given that they are repeatedly losing the battle against PS, and given the reputation Nvidia has managed to build, I was almost certain they would go for the first Nvidia-based Xbox. After all, they are trying to push Windows on the ARM platform, and Apple showed that the transition can happen smoothly, with people not caring about old software as long as they focus on the new software being developed. But they are probably terrified of the idea of such a transition failing. We're talking about a 7-year product cycle. If the Nvidia logo wasn't enough to drive sales and gain market share against PS, MS would either have to lose billions every year selling below cost, or the Xbox division would have to close down completely.

john_The truth is, I expected MS to switch to Nvidia starting as early as this generation. Yes, more expensive hardware, and any incompatibility arising from the transition from x86 to ARM might cost them, but given that they are repeatedly losing the battle against PS, and given the reputation Nvidia has managed to build, I was almost certain they would go for the first Nvidia-based Xbox. After all, they are trying to push Windows on the ARM platform, and Apple showed that the transition can happen smoothly, with people not caring about old software as long as they focus on the new software being developed. But they are probably terrified of the idea of such a transition failing. We're talking about a 7-year product cycle. If the Nvidia logo wasn't enough to drive sales and gain market share against PS, MS would either have to lose billions every year selling below cost, or the Xbox division would have to close down completely.

First gen Xbox was NVIDIA, so there's precedent.

dgianstefaniFirst gen Xbox was NVIDIA, so there's precedent.

- And rumor had it that Nvidia wanted to retain too much control in the process and were tough to work with, so MS went with AMD from then on, who were much more willing to cede control of certain elements of that hardware/software stack to MS.

GodisanAtheist- And rumor had it that Nvidia wanted to retain too much control in the process and were tough to work with, so MS went with AMD from then on, who were much more willing to cede control of certain elements of that hardware/software stack to MS.

Maybe they didn't want the console enshittified by poor MS decision making? IDK, but seems to me like the hardware choices in current gen consoles aren't great. 30 FPS blurry mess targets are laughable in $800 machines useless for anything else except gaming.

Huang tried to rip MS off. They refused to pass savings from chip dieshrinks to MS, which was in those generations the way to be profitable in console business. Launch the hardware with silicon litography N, where chips are huge and you are selling the console at loss. Then shrink to N+1 when available, hardware gets cheaper, lower power, you break even. With die shrink N+2 or even N+3 later, you turn profit or can lower the console's price.

Nvidia either refused to do dieshrinks or kept charging as much as before, harming Microsoft's first Xbox generation. There may have been technological dissatisfaction issues too (Sony dropped Nvidia after PS3 over those a generation later).

dgianstefaniMaybe they didn't want the console enshittified by poor MS decision making? IDK, but seems to me like the hardware choices in current gen consoles aren't great. 30 FPS blurry mess targets are laughable in $800 machines useless for anything else except gaming.

You realize Xbox was Microsoft console? It was neither Nvidia console, nor joint Nvidia-MS project where Nvidia was in place to push some ideas or vision and be in position to mind alleged "enshitifications" or whatever. They were asked (and paid) to supply a GPU and a chipset.

Ols-HolHuang tried to rip MS off. They refused to pass savings from chip dieshrinks to MS, which was in those generations the way to be profitable in console business. Launch the hardware with silicon litography N, where chips are huge and you are selling the console at loss. Then shrink to N+1 when available, hardware gets cheaper, lower power, you break even. With die shrink N+2 or even N+3 later, you turn profit or can lower the console's price.

Nvidia either refused to do dieshrinks or kept charging as much as before, harming Microsoft's first Xbox generation. There may have been technological dissatisfaction issues too (Sony dropped Nvidia after PS3 over those a generation later).

You realize Xbox was Microsoft console? It was neither Nvidia console, nor joint Nvidia-MS project where Nvidia was in place to push some ideas or vision and be in position to mind alleged "enshitifications" or whatever. They were asked (and paid) to supply a GPU and a chipset.

OK, and now here we are with the (better?) AMD alternative that launches consoles already using two generation old GPU IP, that was four generations outdated if you looked at the competition (no AI cores until PS5P).

This is the result of that decision making by both AMD/MS, and predictable (many people here on TPU have been talking about this for years). MS/Sony want to add non-shit upscalers, and it's a struggle, because to get good image quality you need AI, and if you need AI, it's going to be slow on anything without (decent) dedicated AI hardware. So, Pre RDNA 4 or Pre Turing. One of those debuted in 2018, the other in 2025. If you are making a console in 2019, which do you pick?

Obviously a last gen/two generation old combo you can get cheaply from AMD, right? Well here we are. 20-30 base FPS after upscaling from 540p. What can you do with that? Very little. They've boxed themselves in. Can't even sell the consoles cheaply to make up for it because of that same AI train they missed the boat on.

Only good thing I can say about the non-PS5P consoles is at least they weren't using the *shudder* Jaguar cores for CPU. That decision set back game dev what, 5 years? The Zen 2 cores in consoles are at least workable, even in 2026.

The question now becomes how many years using RDNA 1/2 GPU IP in the current gen of consoles set back the game dev industry.

Now we're at the point where the rumours are "next xbox is PC with UDNA 5", so, decent GPU/AI IP, presumably. Downside? ~1.5-2k "console". What's the alternative? PS5P (if you can find), or 20-25 FPS last gen console.

Good luck John Consumer, definitely some great choices set before you. Hope you built a PC a year ago, though you probably didn't.

chrcolukInteresting given Sony restricted to a new Pro console?

Given the GPU architecture the base PS5 and both XBOXes use, the only way to get both decent upscaling and decent performance is to run it on PS5P, it seems. Sony released a worse IQ and lighter upscaler for the base PS5 recently IIRC, to throw them a bone.

I think poor dev practices, RT and UE4/5 have set back the game industry more, alongside the obsession with photo realistic graphics, the upscaling technologies, are a band aid trying to fix that mess.

dgianstefaniMaybe they didn't want the console enshittified by poor MS decision making? IDK, but seems to me like the hardware choices in current gen consoles aren't great. 30 FPS blurry mess targets are laughable in $800 machines useless for anything else except gaming.

- If I recall they wanted to retain ownership over some proprietary shaders and other elements of the console design that MS wasn't happy about.

Part of the reason the old Gearbox Halo PC port way back in the day didn't have the same reflective surfaces and some other fancy shading effects as the console version despite PC parts being more than powerful enough to run the game with all shaders enabled.

I mean the issue you point out is across the board, Sony is in the same boat. But there is a reason both console devs have stuck exclusively with AMD after doing one console gen with Nvidia (Xbox for MS and PS3 for Sony) despite AMD's flagging popularity and install base.

Nintendo is the odd man out, going from AMD/ATI to Nvidia, but that's pretty obviously because Nvidia had an ARM based chip for their form factor that AMD couldn't match in power/thermals.