PS5 Pro is getting a big upgrade in 2026 — I asked Mark Cerny what’s coming, and why AMD’s future PC GPUs feel more 'PlayStation' than ever

And should have been an outlined priority from the start, functionality yes but whats the point if performance not measured against prior versions is what blows my mind. That's a main bolded points of revisions, no?
Agreed. I'm just not surprised that whatever R&D group they put together for PSSR hadn't thought of structuring the QA department accordingly from the get go. They may not have even had a QA department of their own to begin with and were sending builds to the one that does firmware QA as a whole, who probably lack the tools or the expertise to judge IQ, which is far more nuanced than judging performance. I'm making a lot of guesses, but given Cerny admits they hadn't thought about it, I'm probably not far off.
 
Last edited:
The Playstation partnership seems to have delivered good AI upscaling for AMD which was an area they were really lacking in and has done wonders for both the console and PC space. What has the Xbox partnership delivered?
VRS.

Straight Face Trying Not To Laugh GIF
 
Seriously, who cares? 2026.. when next gen consoles are 2027 90% locked.

i am not expecting the PlayStation 5 pro to even sell well at all at this current price tag. let alone someone buying it mid-2026, knowing full well that next-gen consoles are 2027.

Also, the best they can do is FSR4. However, the power of PS5 Pro and upscaling from low resolution to 4k isn't gonna look very pretty anyway.

Not to mention, you want the third-party and first-party devs to again patch their game AGAIN to replace PSSR with whatever new tech they are developing. how many games will really support it ?

Just lay the groundwork for PS6 and move on. we already accepted that.
Next-gen consoles are 90% locked for 2027 ? Based on what, magic ? 🧙‍♂️
Sony itself has said PS6 is years away. If we assume a late 2028 launch (which aligns with their typical 7–8 year cycle), and add 2–3 years of inevitable cross-gen support, PS5 Pro will remain a powerhouse well into 2030.

Also:
– Major AI-based upgrades are coming to PS5 Pro in 2026 (Project Amethyst, PSSR 2.0, FSR 4+)
– The PS6 SoC is still in validation phase
– Dev tools for true next-gen aren't ready yet

Reality check: PS5 Pro isn't just a "mid-gen bump", with what's coming, it might age better than PS4 Pro ever did. Enjoy the ride. Real next-gen won't start before 2030...
 
So happy about this AMD collaboration and AI hardware push Sony did with the PS5 Pro. It sets them up perfectly for PS6, i really hope it comes out in 2028, give it time to really feel like an upgrade and enable it to handle Pathtracing.
 
Has mark cerny lost it? I don't get what's happening, pssr was marketed as ps5 pro special sauce, one of the building blocks behind its customed apu design.

It was to dlss vs fsr vs xess vs pssr

Now pssr is just going to get replaced with fsr4? Mark upscaler philosophy defeated in a year? How many resources did he wasted to make pssr? The custom pssr silicon in pro becomes pointless?
The idea that PSSR's low latency and async compute design can be substituted 'as is' with FSR4, DLSS4 or XeSS is your obvious misunderstanding of the various techniques and the key difference to implementing FSR4 algorithms on the Pro hardware as a drop in replacement for PSSR.

The goals of PSSR informed the hardware design and the ability of that hardware to allow for a full FSR4 implementation shows that Cerny's philosophy wasn't in anyway defeated, merely that the algorithm choices didn't yield the results they had hoped for compared to other techniques in scenarios Sony deemed important.

I also doubt that PSSR is getting replaced by FSR4 on Pro but more a merge of the best aspects of both. In PSSR's case I doubt they would want to lose the benefits of it being a native resolution sparse renderer or it being a U-net that exploits the hardware performance and borrows only a tiny fractional async cost per frame....but on the FSR4 side they clearly want those signal processing algorithms that can improve PSSR.
 
That explains a lot.



Fucking 9060XT is faster than Pro, that's the slowest GPU AMD offers this gen

fm8ufdr.jpeg

D4hNcvA.png

Using PC centric games such as Wukong and Alan Wake 2 is certainly a choice....

Completely unrelated: there's a reason DF never bothered to do Spider-Man 2 PC vs PS5/Pro Performance video.
 
Been playing fast fusion on S2, I hope wipeout omega collection comes back 1 last time with full dualsense support, FSR4, RT, native Dolby atmosphere, PSVR2, 8K 120hz support on the PRO along with a PC version that scale even higher like fully uncapped frame rate, ultrawide for future proofing. The final evolution of the game. 🙏 Good technical showcase game like franchise has always been.
 
Just bring over FSR4 and FG/AFMF and be done with it. PSSR seems kind of pointless, unless PS5 Pro just can't handle FSR4. 🤷‍♂️
As I understand it (and some of this is probably wrong):

The PS5 Pro was always meant to use FSR4 as the upscaler. However, AMD wanted / needed more time to work on it, and so Sony decided to either have a crack at it themselves, or shove PSSR in, knowing it would be removed in future.

I believe that the PS5 Pro has most, but not all, of the hardware features that FSR4 will be using, so there will be some differences between the two at the end of the day.
 
As I understand it (and some of this is probably wrong):

The PS5 Pro was always meant to use FSR4 as the upscaler. However, AMD wanted / needed more time to work on it, and so Sony decided to either have a crack at it themselves, or shove PSSR in, knowing it would be removed in future.

I believe that the PS5 Pro has most, but not all, of the hardware features that FSR4 will be using, so there will be some differences between the two at the end of the day.

Pro has very custom ML hardware not found on RDNA3 or RDNA4. FSR4 on RDNA4 uses hardware acceleration not available on Pro as far as I'm aware.

What Cerny implies is that we will see FSR4 level quality done on Pro hardware, different version but similar (or same) results.
 
Has mark cerny lost it? I don't get what's happening, pssr was marketed as ps5 pro special sauce, one of the building blocks behind its customed apu design.

It was to dlss vs fsr vs xess vs pssr

Now pssr is just going to get replaced with fsr4? Mark upscaler philosophy defeated in a year? How many resources did he wasted to make pssr? The custom pssr silicon in pro becomes pointless?
There is no custom PSSR silicon. There is just custom ML silicon. FSR4 builds upon a lot of the work already done by Sony on PSSR. FSR4 is better as it uses FP8 acceleration on RDNA4, which the Pro does not accelerate as the Pro was developed before RDNA4 was done.

Now what Cerny and Sony are doing is to get FSR4 working efficiently on INT8, which the Pro does have hardware support for. Hence it taking so long, as it likely requires a lot of work to do and isn't trivial. The PS6 would almost certainly have FP8 support.
 
As I understand it (and some of this is probably wrong):

The PS5 Pro was always meant to use FSR4 as the upscaler. However, AMD wanted / needed more time to work on it, and so Sony decided to either have a crack at it themselves, or shove PSSR in, knowing it would be removed in future.

I believe that the PS5 Pro has most, but not all, of the hardware features that FSR4 will be using, so there will be some differences between the two at the end of the day.
I beg to disagree. ps5 pro was designed before FSR4 was a thing and probably their own upscaler was already on the plan.
 
There is no custom PSSR silicon. There is just custom ML silicon. FSR4 builds upon a lot of the work already done by Sony on PSSR. FSR4 is better as it uses FP8 acceleration on RDNA4, which the Pro does not accelerate as the Pro was developed before RDNA4 was done.

Now what Cerny and Sony are doing is to get FSR4 working efficiently on INT8, which the Pro does have hardware support for. Hence it taking so long, as it likely requires a lot of work to do and isn't trivial. The PS6 would almost certainly have FP8 support.
Pretty sure that info is wrong about FP8 and the Pro's capability - from my memory of his AI ML talk on the pro - but will need to go back to the video to be 100%. I think he might have made comment that PSSR specifically didn't benefit from certain formats, not that the formats were missing from the hardware.

The main reason as I understand it why FSR4 doesn't just run on PS5 Pro 'as is' comes from FSR4, DLSS and XeSS all relying on far more L3/L2 cache and far more latency to accommodate much lower L1 cache bandwidth than the PS5 Pro which relies effectively on northbridge bandwidth straight into the register cache, where as the others have huge caches to store the tensor images for long durations without needing to go out to VRAM and without impacting other graphics work operating in parallel - better mimicking a fused neural network than the Pro.
 
Last edited:
Pretty sure that info is wrong about FP8 and the Pro's capability - from my memory of his AI ML talk on the pro - but will need to go back to the video to be 100%. I think he might have made comment that PSSR specifically didn't benefit from certain formats, not that the formats were missing from the hardware.

The main reason as I understand it why FSR4 doesn't just run on PS5 Pro 'as is' comes from FSR4, DLSS and XeSS all relying on far more L3/L2 cache and far more latency to accommodate much lower L1 cache bandwidth than the PS5 Pro which relies effectively on northbridge bandwidth straight into the register cache, where as the others have huge caches to store the tensor images for long durations without needing to go out to VRAM and without impacting other graphics work operating in parallel - better mimicking a fused neural network than the Pro.
I'm quite sure in the video (as well as the interview with DF) he mentioned only INT8 and FP16. INT8 is 300 TOPS as we know, and FP16 is 67 TFLOPS. It makes some sense, FP8 is brand new to RDNA4 and the Pro work was done way in advance of it, probably a lot of what went into the Pro helped the development of RDNA4. We know FSR4 heavily uses FP8, as when it is emulated on RDNA3 the performance is terrible. Converting from FP8 to INT8 probably isn't something trivial.
 
Last edited:
I'm quite sure in the video (as well as the interview with DF) he mentioned only INT8 and FP16. INT8 is 300 TOPS as we know, and FP16 is 67 TFLOPS. It makes some sense, FP8 is brand new to RDNA4 and the Pro work was done way in advance of it, probably a lot of what went into the Pro helped the development of RDNA4. We know FSR4 heavily uses FP8, as when it is emulated on RDNA3 the performance is terrible. Converting from FP8 to INT8 probably isn't something trivial.
Still pretty sure it is in the register extensions they added to the CUs on the Pro - even if it turns out when I check if he doesn't explicitly state because the experimental nature of the ML AI field would make it a bit of a gamble to leave accelerated formats out that are the bedrock of unquantized ML data in PC desktop upscaling inference.

Also I found this reference in an article from yesterday.


Sony does claim that the PS5 can do 300 TOPs INT8/FP8,
 
Still pretty sure it is in the register extensions they added to the CUs on the Pro - even if it turns out when I check if he doesn't explicitly state because the experimental nature of the ML AI field would make it a bit of a gamble to leave accelerated formats out that are the bedrock of unquantized ML data in PC desktop upscaling inference.

Also I found this reference in an article from yesterday.

That comment links to the following article:



Which is speculation from the article writer:

claims a "fully custom" AI accelerator capable of "300 TOPS 8-bit" (FP8+BF8 I guess).

A later comment has a pretty good rundown:

"PS5 Pro is technically an RDNA3 design with RDNA4 RT engines, so it has matrix FP16/INT8/INT4 instructions. The shader ISA, however, remains on RDNA2 (gfx10) to prevent work doubling between PS5 and PS5 Pro. So, no dual-issue FP32 support.
- The "AI cores" in RDNA3 aren't actually cores. They're instructions that execute in the CUs (sound familiar?). 2 AI ops can be issued per CU, one per SIMD32; RDNA has 2xSIMD32s per CU. So, 60 CUs = 120 "AI cores"

Thus, to access the matrix instructions in PS5 Pro, a separate SDK is used, in this case, PSSR. Devs also have to do a little coding to optimize for the new RDNA4 RT engines in PS5 Pro as well.

So, there are a couple of ways FSR4 can be ported, and in PS5 Pro, emulating FP8 through matrix FP16 is a possibility (or using 8-bit integers for most data, then FP16). PSSR 1.0 is heavy on INT8, and that can continue in PSSR 2.0/FSR4 along with emulated FP8 for the parts of the algorithm that require it.

PS5 Pro does not support FP8 natively."

I can't say with any certainty that FP8 is not supported, but I cannot find any source from Sony that says they do, despite talking about INT8 or FP16.
 
Last edited:
That's a Tom Morgan review, not an Alex Battaglia review, y'know, the guy who covers the PC ports. He cited repeatedly how his videos became useless after a week because the game had been patched.

C'mon man I understand Alex is the PC guy but they're all DF.

I also think it's a nonsensical policy to not cover a game at launch because you anticipate patches down the road. At the very least, they should cover all games at their launch state and afterwards they can make a decision whether or not a patch warrants further coverage. It's especially silly when a game (SM2) is proven to run very well on at least one platform (PS5). It's these specific scenarios where analysis should be conducted, as sensible reasoning would suggest it is more likely than not differences in hardware design philosophy and how a certain game may or may not be taking advantage of said hardware design, that's contributing to performance delta. Especially when the developer of said game explicitly states that they are leveraging platform specific tech to execute fundamental gameplay mechanics.

But no, DF decided to be hypocritical and do a console analysis on MindsEye, a game that by their own admission runs poorly on all platforms.
 
C'mon man I understand Alex is the PC guy but they're all DF.
Yes, and most of the time, they actually do stuff they want to do unless there are big games they need to cover. Hell, Alex straight-up avoided covering Hogwarts Legacy on PC because of his friends at era and politics.
I also think it's a nonsensical policy to not cover a game at launch because you anticipate patches down the road.
I disagree and think this is perfectly sensible. You spend 1-2 weeks producing a video that is rendered obsolete in a day, making your work worthless. Those productions don't take 1 or 2 hours to do, the GOWR PC coverage was apparently two weeks of work.
At the very least, they should cover all games at their launch state and afterwards they can make a decision whether or not a patch warrants further coverage. It's especially silly when a game (SM2) is proven to run very well on at least one platform (PS5). It's these specific scenarios where analysis should be conducted, as sensible reasoning would suggest it is more likely than not differences in hardware design philosophy and how a certain game may or may not be taking advantage of said hardware design, that's contributing to performance delta. Especially when the developer of said game explicitly states that they are leveraging platform specific tech to execute fundamental gameplay mechanics.

But no, DF decided to be hypocritical and do a console analysis on MindsEye, a game that by their own admission runs poorly on all platforms.
That would be pointless and just a bunch of speculation. You think hardware and design philosophies differences are the reason for the performance disparities, but ultimately, they would just be guesses. DF doesn't have the profiling tools to explain why a game runs well on X platform compared to Y. Either the dev tells them, they can see the obvious differences with the publicly available tools, or they just do guesswork.

The bottom line is, Spider-Man 2's release on PC was disastrous, it got patched like 5 times in the first month, so Alex covering would have made his video quickly pointless. Could he do it now? I suppose so, but even now, the reviews are "mostly positive" at 79% compared to overwhelmingly positive at 96% for the first one. There are still performance issues that perhaps are the result of hardware design or perhaps not.
 
Last edited:
C'mon man I understand Alex is the PC guy but they're all DF.

I also think it's a nonsensical policy to not cover a game at launch because you anticipate patches down the road. At the very least, they should cover all games at their launch state and afterwards they can make a decision whether or not a patch warrants further coverage. It's especially silly when a game (SM2) is proven to run very well on at least one platform (PS5). It's these specific scenarios where analysis should be conducted, as sensible reasoning would suggest it is more likely than not differences in hardware design philosophy and how a certain game may or may not be taking advantage of said hardware design, that's contributing to performance delta. Especially when the developer of said game explicitly states that they are leveraging platform specific tech to execute fundamental gameplay mechanics.

But no, DF decided to be hypocritical and do a console analysis on MindsEye, a game that by their own admission runs poorly on all platforms.
The issue isn't just poorly performing games, the issue is games they had no prior access to and getting one or two patches weekly. By the time the video comes out they would already be a few patches behind. Look at Ragnarok, they had early access to that and the video was already out of date in the very week it came out, with issues being patched.

Now an argument can be made they should cover it when it is fully patched, but DF is a business, not a charity, and looking back at a older release (that is not retro) won't generate nearly as much interest and views as something fresh, something they have talked about a few times. But he may or may not come back to it.
 
That comment links to the following article:



Which is speculation from the article writer:

claims a "fully custom" AI accelerator capable of "300 TOPS 8-bit" (FP8+BF8 I guess).

A later comment has a pretty good rundown:

"PS5 Pro is technically an RDNA3 design with RDNA4 RT engines, so it has matrix FP16/INT8/INT4 instructions. The shader ISA, however, remains on RDNA2 (gfx10) to prevent work doubling between PS5 and PS5 Pro. So, no dual-issue FP32 support.
- The "AI cores" in RDNA3 aren't actually cores. They're instructions that execute in the CUs (sound familiar?). 2 AI ops can be issued per CU, one per SIMD32; RDNA has 2xSIMD32s per CU. So, 60 CUs = 120 "AI cores"

Thus, to access the matrix instructions in PS5 Pro, a separate SDK is used, in this case, PSSR. Devs also have to do a little coding to optimize for the new RDNA4 RT engines in PS5 Pro as well.

So, there are a couple of ways FSR4 can be ported, and in PS5 Pro, emulating FP8 through matrix FP16 is a possibility (or using 8-bit integers for most data, then FP16). PSSR 1.0 is heavy on INT8, and that can continue in PSSR 2.0/FSR4 along with emulated FP8 for the parts of the algorithm that require it.

PS5 Pro does not support FP8 natively."

I can't say with any certainty that FP8 is not supported, but I cannot find any source from Sony that says they do, despite talking about INT8 or FP16.

I can see why you are saying that from what you quoted, but what you quoted doesn't say anything about what alterations were made to the CU registers to provide the uncontested 300TOPs at INT8 from extensions that RDNA3 doesn't have that Cerny explained in detail in his PS5 Pro technical breakdown video.

Now that Cerny is quoted as saying - something like - it will be the full fat FSR4 on PS5 Pro, either the Pro does have a register extension to do FP8 at 300TOPs, or the mixed precision layers - quantised and unquantized - in all these DLSS/FSR4/XeSS allow for FP16 to be used in place on the Pro where latency hiding - with orders less latency to begin with -makes it a non-issue.

Given that the main layers of use for FP8/FP16 in DLSS/FSR4.XeSS is motion vector inferencing and detail extraction layers AFAIK, and PSSR is already superior in many ways at detail extraction using INT8 (quantisation) and it (uses or) can use sparse rendering at full resolution from just 1/4 of the output resolution it might be a case of other solutions have to infer the whole final resolution motion layer, whereas the sparsity would maybe provide enough final resolution motion quality for interpolating the 3/4 of missing pixels in each output tile effectively reducing the amount of motion inference done - in the lower tensors - lowering use of FP8(FP16) to something like a 1/3 of other solutions for similar quality.
 
Last edited:
The thing about dlss on pc that makes it magical is you can force dlss 4 for games that have inferior version of the upscaller without the devs coming in and doing a dedicated patch. Like Jedi Survivor will still look that blurry on ps5 pro when this latest pssr update hits. Its only worth it for future games of which how many is likely gonna be next gen anyway.
Jedi Survivor (PS5 Pro) is not blurry, but rather has a crisp 4K image quality.
The only issue it has is "shimmering of vegetation in bright daylight."
I hope next year's PSSR update will be applied to this game.
ExoztrBQ_o.png
 
Jedi Survivor (PS5 Pro) is not blurry, but rather has a crisp 4K image quality.
The only issue it has is "shimmering of vegetation in bright daylight."
I hope next year's PSSR update will be applied to this game.
ExoztrBQ_o.png
I believe it also introduced a texture streaming issue iirc? Not sure if they were hitting a bottleneck…
 
As I understand it (and some of this is probably wrong):

The PS5 Pro was always meant to use FSR4 as the upscaler. However, AMD wanted / needed more time to work on it, and so Sony decided to either have a crack at it themselves, or shove PSSR in, knowing it would be removed in future.

I believe that the PS5 Pro has most, but not all, of the hardware features that FSR4 will be using, so there will be some differences between the two at the end of the day.
Sony started working with AMD on the PS5 Pro before AMD decided to move FSR to an ML model. When AMD did finally decided to make FSR4 using ML, PS5 Pro design was already done and it did not support FP8 which means it can't run the same model as RDNA4 PC GPUs.
 
Sony started working with AMD on the PS5 Pro before AMD decided to move FSR to an ML model. When AMD did finally decided to make FSR4 using ML, PS5 Pro design was already done and it did not support FP8 which means it can't run the same model as RDNA4 PC GPUs.
Given your excellent inside knowledge on things like this I would say that pretty much settles mine and Zathalus Zathalus difference of opinion on whether FP8 at 300TOPs was implemented as an extension or not - clearly with Zathalus opinion matching your info.

Based on what I said in my previous post in regards of the AI layers FP8(and FP16) are used, how do you think the PS5 Pro will run the full fat FSR4 solution given that "lite" versions on older RDNA already fallback to INT8?

Do you have any knowledge of any possible PS5 Pro ASIC for inferencing of motion vectors - faster than FP16 workloads - or like I suggested do you think the drop of 4.5 to 1 (from INT8 at 300TOPs) to FP16 - 67(H)TFLOPs to do FP8 - is a much lower hit in performance & total latency than the drop automatically implies of 18TOPs @INT8 down to 4(H)TFLOPs@FP16 - assuming 1ms of 16.6ms is the upscaler running time?
 
Seriously, who cares? 2026.. when next gen consoles are 2027 90% locked.

i am not expecting the PlayStation 5 pro to even sell well at all at this current price tag. let alone someone buying it mid-2026, knowing full well that next-gen consoles are 2027.

Also, the best they can do is FSR4. However, the power of PS5 Pro and upscaling from low resolution to 4k isn't gonna look very pretty anyway.

Not to mention, you want the third-party and first-party devs to again patch their game AGAIN to replace PSSR with whatever new tech they are developing. how many games will really support it ?

Just lay the groundwork for PS6 and move on. we already accepted that.
Sony has never mentioned 2027 for anything regarding the PS6.
In fact Cerny seems to imply in this interview that PS5 is going to be the priority for a while.
Even if PS6 releases in 2027 it is crazy to just pretend that 2026 is "too late" for anything regarding the PS5 era. Cross gen period will be big as hell (if we are getting PS4 versions of games 4 years later, imagine how it's gonna be for the PS5). This gen will be having games well past 2030. A substantial update in 2026 for the Pro model is more than welcomed, specially with GTAVI releasing that same year.

They already are laying the groundwork for the PS6...but wanting them to literally ignore their most expensive piece of hardware because "2026 is too late" is crazy lol.
 
It would be great if Sony was able to simply make this a system update that transparently worked with all games that use the existing PSSR tech. Sadly, I doubt it, since it tends to never work that way.
You are assuming the inputs the new PSSR need they are the exact same the current one, DLSS and FSR works simply because they had iterations for years until they finished the api. PS5 Pro is not even a one year old.
 
Last edited:
Top Bottom