PlayStation 6 to utilize AMD's 3D stacked chips; AMD UDNA Flagship GPU revived for 2026, Zen 6 Halo with 3D stacking technology, and Zen 6 all on TSMC

The 3D vcache chips are still being made in N6. So they are not as expensive as cutting edge chips in N5 or N3 nodes.
If it was cheap enough and wasn't going to eat lots of die space I could see them taking a 3D vcache mostly for a superior memory hierarchy for the GPUs ML AI, and if it improved CPU latency using GDDR, and reduced the need for a x4 increase over the PS5 for north bridge bandwidth and provided a nice store cupboard for the IO complex. but I still have my doubts...if they were stacking two APUs and the 3d cache was mainly a substitute SLI/Crossfire connection then I think that would be the most logical reason they'd use one.
 
If it was cheap enough and wasn't going to eat lots of die space I could see them taking a 3D vcache mostly for a superior memory hierarchy for the GPUs ML AI, and if it improved CPU latency using GDDR, and reduced the need for a x4 increase over the PS5 for north bridge bandwidth and provided a nice store cupboard for the IO complex. but I still have my doubts...if they were stacking two APUs and the 3d cache was mainly a substitute SLI/Crossfire connection then I think that would be the most logical reason they'd use one.

We have already seen that Zen CPUs benefit greatly from that cache. Mostly with games.
And RDNA2 to 4 also use cache to reduce memory accesses. So much so, that even Nvidia greatly increased their L2 cache after Ada.
So it does work very well. I was very surprised that Sony didn't use this cache. Even if it was just a 32MB chunk.
The Pro is very limited by memory bandwidth and this would have help them a lot in reducing memory accesses and pressure on the vram bus.
 
Every cents count, 9800X3D may be faster than 7800X3D, but you look at the TDP used, it goes up to 150W and more in games, that is almost my 7950x3d in full stress load (180W~200W i set it)

I think the 64MB extra cache will be less useful if the CPU is able to address to faster unified memory
 
Every cents count, 9800X3D may be faster than 7800X3D, but you look at the TDP used, it goes up to 150W and more in games, that is almost my 7950x3d in full stress load (180W~200W i set it)

I think the 64MB extra cache will be less useful if the CPU is able to address to faster unified memory

The reason that the 3Dvcache is effective is because it improves locality.
A unified memory system improves communication between CPU and GPU.
But it does not improve the connecting between memory and CPU. In fact, it makes it worse, as gaming companies prefer to have memory controllers for high bandwidth, such as GDDR, instead of latency.
Remember that going to memory for a CPU, that has high levels of instruction dependency and branching, is a huge performance hit. This is why CPUs use over a third of their die space just for the front end and another third for caches, not counting 3dv caches.
 
We have already seen that Zen CPUs benefit greatly from that cache. Mostly with games.
And RDNA2 to 4 also use cache to reduce memory accesses. So much so, that even Nvidia greatly increased their L2 cache after Ada.
So it does work very well. I was very surprised that Sony didn't use this cache. Even if it was just a 32MB chunk.
The Pro is very limited by memory bandwidth and this would have help them a lot in reducing memory accesses and pressure on the vram bus.
But in the PS4 era they were cache snooping the other CCD (IIRC) with hand optimisation to reduce latencies so a lot of the easy benefits of any L3 cache don't necessarily apply to a console unless the software is as generic and unoptimized as the PC.

Yes a 3D cache will help, but more than other areas for the same cost? Only in the case of an SLI/Crossfire connection substitute would be my guess.

In the event a vcache would have been planned for the PS6 when finalizing the PS5 Pro the obvious reason why they wouldn't have did a 32mb vcache is the same reason why they didn't put a ssd in the PS4 Pro, they need to keep a substantial difference between the half-gen upgrade and the next-gen
 
But in the PS4 era they were cache snooping the other CCD (IIRC) with hand optimisation to reduce latencies so a lot of the easy benefits of any L3 cache don't necessarily apply to a console unless the software is as generic and unoptimized as the PC.

Yes a 3D cache will help, but more than other areas for the same cost? Only in the case of an SLI/Crossfire connection substitute would be my guess.

In the event a vcache would have been planned for the PS6 when finalizing the PS5 Pro the obvious reason why they wouldn't have did a 32mb vcache is the same reason why they didn't put a ssd in the PS4 Pro, they need to keep a substantial difference between the half-gen upgrade and the next-gen

As much as try, there are limits to branch speculation and prefetching data.
And even on a CPU, where latency is not as important, just not having to go as often to vram, helps a lot with performance.
 
Sony wont use 3D, with costs of making chips increasing non-stop.

They just need a new zen 6 architecture.

Just take a 9600X and compare to 5800X3D, the newer architecture is better in most games
We never thought Sony would put an SSD in the PS5 either, with how expensive NAND Flash is.
 
As much as try, there are limits to branch speculation and prefetching data.
And even on a CPU, where latency is not as important, just not having to go as often to vram, helps a lot with performance.
Completely agree which is why I think everything will be compromised in preference to getting a 120MB combined L1/L2 vector register memory across 96 WGPs for the PS6 because if they can get PSSR2 to be fully fused as a neural network every system they offload to TOPs processing within the GPU next-gen will see huge multi factor performance gains, and if they fully fuse PSSR2 then rendering even lower native 540p (270p) would again provide multi factor headroom for rendering native with more fx for each tier lower they were able to reach.
 
Last edited:
The question I would be asking is why would Sony need a 64MB L3 performance cache in the PS6- when the critical performance cache - vector register memory - in the PS5 Pro is way beyond L3 performance AFAIK and assuming there is double the WGPs of the PS5 Pro in the PS6 that cache will double by default to 30MB?
I was thinking in can double as Infinity Cache as well.
 
I was thinking in can double as Infinity Cache as well.
Which would be great but with so many constraints by BoM by TDP limit, by die space and vector register amount for PSSR I don't think it fits all possible designs to justify its inclusion.

If they opt for stacking two APUs and having a large vcache, I think that makes the most sense for a design that justifies one.
 
We never thought Sony would put an SSD in the PS5 either, with how expensive NAND Flash is.

Yes it was a surprised. But circumstances are different. They were lucky NAND was overproduced. And they did sacrifised VRAM and using older zen2 APU.

Having a SSD is also forward looking useful, cant say the same for 64MB of extra cache

Zen6 with bigger infinity fabric to VRAM and give us 32GB of GDDR7.
 
Last edited:
Which would be great but with so many constraints by BoM by TDP limit, by die space and vector register amount for PSSR I don't think it fits all possible designs to justify its inclusion.

If they opt for stacking two APUs and having a large vcache, I think that makes the most sense for a design that justifies one.
Guess it depends if the 3D stacking leak is true.

If true, other than 3D V-Cache design, AMD has RDNA5/CDNA5 (UDNA) 3D stacking design.
cdLWn0Z.jpeg


Based on the image above, PS6 can be made with 1 AID, 1 CCD, and 2 SEDs, where the CCD and SEDs are stacked on top of the AID.
 
Yes it was a surprised. But circumstances are different. They were lucky NAND was overproduced. And they did sacrifised VRAM and using older zen2 APU.

Having a SSD is also forward looking useful, cant say the same for 64MB of extra cache

Zen6 with bigger infinity fabric to VRAM and give us 32GB of GDDR7.

There is one misconception there. Zen3 released on PC, a month before the PS5.
That means there was no time to make it into the PS5 SoC, unless Sony was willing to delay the PS5 release date by many months.
Just consider that AMD always releases the APUs around a year after the CPUs with the same arch.
At that time, the most up to date APUs from AMD, ready for production were Zen2. Zen3 APUs were far from being ready.

A similar thing might happen with the PS6. If it releases at the same time frame as Zen6, then it will probably use Zen5 APU.
 
There is one misconception there. Zen3 released on PC, a month before the PS5.
That means there was no time to make it into the PS5 SoC, unless Sony was willing to delay the PS5 release date by many months.
Just consider that AMD always releases the APUs around a year after the CPUs with the same arch.
At that time, the most up to date APUs from AMD, ready for production were Zen2. Zen3 APUs were far from being ready.

A similar thing might happen with the PS6. If it releases at the same time frame as Zen6, then it will probably use Zen5 APU.
If they had used Zen3, the PS5 has been more expensive
 
I believe L3 on the CPU takes up quite a lot of die space, so if you can shrink down your bespoke APU (or CPU CCD?) quite a bit by removing half or more of the L3 on-die; and then use a generic, off-the-shelf L3 V-cache die that's already in production for other desktop/laptop parts (and is on a much older, cheaper node), I'd expect that could come with considerable cost savings.

I guess there's nothing stopping them removing all of it if it's a bespoke chip and using only V-Cache for L3. And if they do chiplets, you'd have a tiny CPU-CCD, especially with reduced FPU again and even moreso with a "C" Zen Compact variant (you're probably not gonna need absolute top end clocks on a console anyway).

A custom ZEN 6C CPU CCD @ ~4.2GHz with a 12C / 24T on N3E + 64MB of L3 3D V-Cache Die on N6 would be pretty sweet (they could even use lower yield chips to save a few extra pennies and go with ~48MB active, which'll be a 6x step up from PS5 & PS5 Pro).

The packaging / stacking process is mature and the only real concern is thermals, but seeing as this device will be provisioned for a sizeable GPU die (or APU), then I expect it'll be well within the realm of possibilities. And Sony also have significant economies of scale on their side, for eg. they sold 22m PS5s before the stealth N6 refresh, so that's 20m of a given chip/config there, then another 40m+ since of another chip/config.
 
Last edited:
The NextBox and PS6 Need to achieve the following:

-Skip PCIE 5.0 and be PCIE 6.0
-NVME SSD speeds 15GB/Sec to 30 GB/Sec or moar
-Wifi7
-Latest Bluetooth
-GPU Clock Speed slightly past 3Ghz threshold
-CPU Clock Speed slightly past 5Ghz
-2nm+ process
-25GB-32GB GDDR7 RAM available for games with bandwidth slightly past 1 TB/sec
-Multiple USB 4.2 ports (80gbps)
-Raytracing/Path tracing (with recent patents and being realistic) up to 5080 performance
-8K 120fps with checkerboarding and FSR4.0+
-neural rendering and compression
-30-50 TitieFLOP range
-500-1000 TOPS+ with NPU in CPU + GPU
-Super effecient and has brute force at same time.
 
The NextBox and PS6 Need to achieve the following:

-Skip PCIE 5.0 and be PCIE 6.0
-NVME SSD speeds 15GB/Sec to 30 GB/Sec or moar
-Wifi7
-Latest Bluetooth
-GPU Clock Speed slightly past 3Ghz threshold
-CPU Clock Speed slightly past 5Ghz
-2nm+ process
-25GB-32GB GDDR7 RAM available for games with bandwidth slightly past 1 TB/sec
-Multiple USB 4.2 ports (80gbps)
-Raytracing/Path tracing (with recent patents and being realistic) up to 5080 performance
-8K 120fps with checkerboarding and FSR4.0+
-neural rendering and compression
-30-50 TitieFLOP range
-500-1000 TOPS+ with NPU in CPU + GPU
-Super effecient and has brute force at same time.
Stacking Artificial Intelligence GIF by Krater.ai
 
The NextBox and PS6 Need to achieve the following:

-Skip PCIE 5.0 and be PCIE 6.0
-NVME SSD speeds 15GB/Sec to 30 GB/Sec or moar
-Wifi7
-Latest Bluetooth
-GPU Clock Speed slightly past 3Ghz threshold
-CPU Clock Speed slightly past 5Ghz
-2nm+ process
-25GB-32GB GDDR7 RAM available for games with bandwidth slightly past 1 TB/sec
-Multiple USB 4.2 ports (80gbps)
-Raytracing/Path tracing (with recent patents and being realistic) up to 5080 performance
-8K 120fps with checkerboarding and FSR4.0+
-neural rendering and compression
-30-50 TitieFLOP range
-500-1000 TOPS+ with NPU in CPU + GPU
-Super effecient and has brute force at same time.

So you want a $2,000 console? They are gonna have to get creative. This will probably be Cerny's last hurrah as well so I expect something good.
 
The NextBox and PS6 Need to achieve the following:

-Skip PCIE 5.0 and be PCIE 6.0
-NVME SSD speeds 15GB/Sec to 30 GB/Sec or moar
-Wifi7
-Latest Bluetooth
-GPU Clock Speed slightly past 3Ghz threshold
-CPU Clock Speed slightly past 5Ghz
-2nm+ process
-25GB-32GB GDDR7 RAM available for games with bandwidth slightly past 1 TB/sec
-Multiple USB 4.2 ports (80gbps)
-Raytracing/Path tracing (with recent patents and being realistic) up to 5080 performance
-8K 120fps with checkerboarding and FSR4.0+
-neural rendering and compression
-30-50 TitieFLOP range
-500-1000 TOPS+ with NPU in CPU + GPU
-Super effecient and has brute force at same time.
That would be rubbish, as you would have added 4-6ms to PSSR2 with an NPU solution. and would have lost real 60fps to fake frame-gen with 33ms latency baked in, so it isn't super efficient unless just doing native rendering - which in itself is highly inefficient.
 
There is one misconception there. Zen3 released on PC, a month before the PS5.
That means there was no time to make it into the PS5 SoC, unless Sony was willing to delay the PS5 release date by many months.
Just consider that AMD always releases the APUs around a year after the CPUs with the same arch.
At that time, the most up to date APUs from AMD, ready for production were Zen2. Zen3 APUs were far from being ready.

A similar thing might happen with the PS6. If it releases at the same time frame as Zen6, then it will probably use Zen5 APU.
Zen6 launches a year before PS6
 
The NextBox and PS6 Need to achieve the following:

-Skip PCIE 5.0 and be PCIE 6.0
-NVME SSD speeds 15GB/Sec to 30 GB/Sec or moar
-Wifi7
-Latest Bluetooth
-GPU Clock Speed slightly past 3Ghz threshold
-CPU Clock Speed slightly past 5Ghz
-2nm+ process
-25GB-32GB GDDR7 RAM available for games with bandwidth slightly past 1 TB/sec
-Multiple USB 4.2 ports (80gbps)
-Raytracing/Path tracing (with recent patents and being realistic) up to 5080 performance
-8K 120fps with checkerboarding and FSR4.0+
-neural rendering and compression
-30-50 TitieFLOP range
-500-1000 TOPS+ with NPU in CPU + GPU
-Super effecient and has brute force at same time.
So a $2000 console?

Give I Want One GIF
Kevin Hart GIF by Complex
 
599 USD is the limit, I'd say, for the base console. Any more than that and the casual gamer will just stick with a PS5 or move to another platform. So there will definitely be cost cutting somewhere, I very much doubt it will have the latest version of everything. IMO I can see Zen 5 (maybe somewhat upgraded Zen 5+) and 20-24 GB GDDR6. Most of the "upgrade budget" will likely go to getting the latest UDNA chip and AI features.
 
599 USD is the limit, I'd say, for the base console. Any more than that and the casual gamer will just stick with a PS5 or move to another platform. So there will definitely be cost cutting somewhere, I very much doubt it will have the latest version of everything. IMO I can see Zen 5 (maybe somewhat upgraded Zen 5+) and 20-24 GB GDDR6. Most of the "upgrade budget" will likely go to getting the latest UDNA chip and AI features.
When launching new hardware, Sony more times than not utilize the latest hardware, sometimes even unreleased tech and with the XBSX already going for $599, I highly doubt that'll be the limit.
 
599 USD is the limit, I'd say, for the base console. Any more than that and the casual gamer will just stick with a PS5 or move to another platform. So there will definitely be cost cutting somewhere, I very much doubt it will have the latest version of everything. IMO I can see Zen 5 (maybe somewhat upgraded Zen 5+) and 20-24 GB GDDR6. Most of the "upgrade budget" will likely go to getting the latest UDNA chip and AI features.

What other platform? Switch 2? Because Xbox sure as hell doesn't seem interested in producing consoles at a loss anymore. They've been throwing around some wild specs for their next console, and that won't come cheap.

PC parts are more expensive than ever, so what other platforms are left? PlayStation has built such a strong ecosystem that many users won't jump ship. I agree that people may stick with their PS5s for longer, and the adoption of a PS6 might be slower, but pricing it above $599 wouldn't be a disaster.
 
599 USD is the limit, I'd say, for the base console. Any more than that and the casual gamer will just stick with a PS5 or move to another platform. So there will definitely be cost cutting somewhere, I very much doubt it will have the latest version of everything. IMO I can see Zen 5 (maybe somewhat upgraded Zen 5+) and 20-24 GB GDDR6. Most of the "upgrade budget" will likely go to getting the latest UDNA chip and AI features.
The only way for the next ps6 to be 599 is having two skus... in this case the 599 would be ps6 "s" and we would get a 799/899 ps6 "x". And I think this would be the best option, in this case there would be no mid gen refresh.

If they opt for 1 sku then 599 its not happening. I would easily bet my account.
 
Last edited:
The only way for the next ps6 to be 599 is having two skus... in this case the 599 would be ps6 "s" and we would get a 799/899 ps6 "x". And I think this would be the best option, in this case there would be no mid gen refresh.

If they opt for 1 sku then 599 its not happening. I would easily bet my account.
Or Sony can just only make the PS6x to avoid the XBSS/XBSX problem, sell it at a $100 loss and make back that loss on subscription and software sales until hardware prices come down.
 
Or Sony can just only make the PS6x to avoid the XBSS/XBSX problem, sell it at a $100 loss and make back that loss on subscription and software sales until hardware prices come down.
With Xbox being less of a threat to Playstation now, it's safer to take the risk of subsidising the console, but with the current economy and an unstable and unpredictable US I think convincing shareholders to sell at a loss will be very hard.
 
Or Sony can just only make the PS6x to avoid the XBSS/XBSX problem, sell it at a $100 loss and make back that loss on subscription and software sales until hardware prices come down.
Yeah or maybe they have been keeping existing prices high to pre-subsidize knowing that the first 20m sales need sold at a good price and with a built-in disc drive to avoid any negativity at the start, A revised slim later on with an optional drive and price increases after a successful two year launch window now seem to be the safer approach to keep the door closed from enticing a new competitor to displace xbox as a platform.

£549 ($715) including taxes with a built-in optical and just 512GB storage on the mobo- intended to be expanded via NVME at day one like the old 20gb PS3 - but with a 4TB model on the mobo and with the PS6 Pro controller and 12months free PSN standard at £799($949) including taxes to get less price sensitive buyers to select the premium model that is still a good deal when charging £200 for a pro controller, but a lot of easy mark up for Sony with 100 more to the BoM, but with 250 more to the price and 300-400 more to value.
 
Or Sony can just only make the PS6x to avoid the XBSS/XBSX problem, sell it at a $100 loss and make back that loss on subscription and software sales until hardware prices come down.
The xbox s was not an equivalent machine it was basically a gen of difference from series x... if you had two exactly equal skus and the only difference being mainly the GPUs (just like ps5 to ps5pro) this would be avoided.
 
I'm not familiar with virtually any of the tech that people are discussing.

I would hope that, in a nutshell, Sony makes a system that is powerful enough that we don't need to go through another generation that has a "Pro" update.

In terms of how strong it'll be, I think it's going to lean far more heavily into AI doing a lot of the extra work.

I think you'll have a console whose specs may not be as powerful as people want, but it punches far beyond what it should due to that AI.

I'd still like higher specs for a more "Pro" experience than have a Pro later on.

I don't see any way the system launches below $700, same as the Pro is now.
 
3D stacking will also be applied to the price of the console.
People are still expecting chiplets for next consoles?
AMD has gone from chiplets to monolithic between RDNA3 and 4 and this alone resulted in ~33% price cuts.
Chiplets make sense only for high end systems where you can't build a chip bigger than what the process allows for. Anything price sensitive is poised to be monolithic since that is just cheaper.
Consoles are the opposite of high end systems and so expecting any sort of chiplets in them is completely misguided. The upcoming gen will most likely use monolithic APUs again just like the two previous ones. And if I had to guess I would expect them to be on N4 (realistically) or N3 (at best).
 
The NextBox and PS6 Need to achieve the following:

-Skip PCIE 5.0 and be PCIE 6.0
-NVME SSD speeds 15GB/Sec to 30 GB/Sec or moar
-Wifi7
-Latest Bluetooth
-GPU Clock Speed slightly past 3Ghz threshold
-CPU Clock Speed slightly past 5Ghz
-2nm+ process
-25GB-32GB GDDR7 RAM available for games with bandwidth slightly past 1 TB/sec
-Multiple USB 4.2 ports (80gbps)
-Raytracing/Path tracing (with recent patents and being realistic) up to 5080 performance
-8K 120fps with checkerboarding and FSR4.0+
-neural rendering and compression
-30-50 TitieFLOP range
-500-1000 TOPS+ with NPU in CPU + GPU
-Super effecient and has brute force at same time.
Even holdiays 2028 thats 2k usd at the least, both cpu and gpu clocks u mentioning are super unrealistic, ssd speed too- if ps6 has similar/bit faster ssd, as long as its 2tb, it will be fine.
PS6 will be barely able to do 4k60 after ai upscaling, forget 8k120.
Remember we will want many games of graphic fidelity like the matrix demo/gta6 trailers and those run barely over 1080p, below 1440p and that is already after ai upscaling on base ps5.
Realistic is around 30tflops gpu(so 3x base ps5) but that means games as goodlooking as gta6/enter the matrix will have to be 4k60 upscaled from maybe around 1440p or below that.

And yes games that look much worse will be able to hold stable 4k60, same way we had 4k60 on ps4pr0 (4,2tflop gpu and jaguar cpu).
 
With Xbox being less of a threat to Playstation now, it's safer to take the risk of subsidising the console, but with the current economy and an unstable and unpredictable US I think convincing shareholders to sell at a loss will be very hard.
Share prices would drop anyway if they launched a 700+ USD console and nobody buys it. Especially with crossgen support expected to be even longer, it would probably end up selling only a little better than the Pro. But they could end up in a scenario where the same people that already have the Pro upgrade to PS6, and base PS5 owners upgrade to used Pro versions, making adoption even slower.
 
you people are out of your minds, there wont be any ps6, forgett about it, the ps5 pro tanked, its impossible to reduce the price for moore law is dead
 
Top Bottom