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
- Thread starter Lunatic_Gamer
- Start date
- News Rumor Hardware
PaintTinJr
Member
Surely in that scenario the code name would be: Smith or Agent Smith - as farmerboy said previously - no? As Sydney isn't part of the Matrix's codename lexicon options AFAIK.
Last edited:
It's not console related.
jm89
Member
Infinity Gems
Neo Member
The One Down Under? Is it Chip related?
Physiognomonics
Member
Wait, what did I miss? 32GB not possible using GDDR7? But the 5090 has that so there are 4GB chips out there.
Stacked 3d Cache wouldn't make sense in a console. Much to expensive. Only if they need more Die area for the cooling, something like big caches would make sense.
Hmm is a pipedream because it is even more expensive. By the time the ps6 is out, gddr8 should be the norm.
Also gddr7 allows 3gb chips on 32 bit interface. But there are still the conventional chips.
Hmm is a pipedream because it is even more expensive. By the time the ps6 is out, gddr8 should be the norm.
Also gddr7 allows 3gb chips on 32 bit interface. But there are still the conventional chips.
Smiggs
Member
The 5090 uses 2GB modules. If I remember, they've been doing a 'stacked' design on the backside since the 3090. Someone correct me if I'm wrong, though.
It probably depends on performance levels versus the next tier up. Looking at AMD chips, AM4 is beginning to show its age apart from the X3D chips which basically added another gen in performance levels. They could potentially use a cheaper node with 3D cache to make-up performance and end up with a good cost.
With a console it's better as games can be designed to have major gameplay loops small enough to fit in the cache rather than it being hit and miss like it is on PC.
TheRaidenPT
Member
They can put it on top of GPU and CPU.. Also AMD's general Halo APUs seem to be going that route.
There's much we don't know about UDNA and the overall Project Amethyst
Radical_3d
Member
These unjustified thread bumps with no new info…
Negotiator
Banned
RTX 5090 has 512-bit bus, while consoles have never exceeded 384-bit (Xbox One X, which is an outlier).
Sony has stuck to 256-bit for 2 generations in a row. At best you can expect 24GB with 3GB chips.
Source?
Modern game executables are compiled in x86-64, which has the same code footprint on both PCs and consoles.
We're talking about 50-60MB *.exe files (yeah, open world code is quite complex) and the PS5 CPU only has 8MB L3 cache (vs 32MB on desktop Zen 2)...
There's no way PCs have more cache misses vs consoles (that includes the iGPU, which only has 4MB vs double-digit amounts in Ada/Blackwell and RDNA2/3/4).
PS6 will happen, but:
1) It will have 2 SKUs (XBOX Series style). $450 vs $900 (without a Blu-Ray drive).
2) It won't be raster-heavy (Cerny has already tried to tame expectations by using the term "flopflation"). No more Teraflop system wars (XBOX is semi-dead either way).
FP32-wise don't expect a huge bump over PS5 Pro.
PS6 lite will likely have the same raster power and they will tout AI tech everywhere (enhanced PSSR -closer to FSR4 perhaps-, neural texture compression, AI frame generation + latency reduction etc.)
It will be more of a refinement (PS5 Pro Ultra) than a new generation and the cross-gen period will be even longer than ever before (2030+), especially with Switch 2 on the horizon.
Don't say you haven't been warned...
I was saying what would happen if consoles started using 3d cache. They would start optimising for the new cache limit, whereas at the moment on PC it's a bit hit and miss from game to game as they're not optimised specifically for the extra cache. That's why some games run way quicker, and some are pretty much unaffected by the extra cache.
If consoles started using 3d cache, all games would be optimised for it which would also benefit PC.
Negotiator
Banned
Quite the opposite: with a bigger cache, less optimization is needed.
Optimization means making the code smaller...
AMD never said 3D V-Cache requires optimization. It's just a way to reduce DRAM transfers.
Same thing with big caches in GPUs. The driver does all the job.
That can happen but If they do not show the difference that "a console for AI" can do people simple wont buy, they can't expect to make a console with the objective of make the devs lives easy and forgot about the people who pay for all this, AKA the consumers
reinking
Gold Member
Kerotan
Member
Hopefully end of 2028 at the earliest. If Xbox goes kaput I hope they do 2029.
ZiggyPop1991
Member
Why are people even asking or excited for a PS6? Haven't we figured out already it isn't the actually hardware (Consoles and PCs) it is the developers not actually doing their jobs. High budgets and still sending out games that aren't optimized. Chasing graphics mean nothing when we have all of these "modes" to choose from anyways. Focus on making games actually run at a consistent stability with a timely release.
I don't even want to hear about the next system. When you games that aren't even that demanding being enjoyed by millions anyways. These teams need to seriously do better, lol.
I don't even want to hear about the next system. When you games that aren't even that demanding being enjoyed by millions anyways. These teams need to seriously do better, lol.
The issue happens all the time at the moment, hence some games / engines not getting the performance increases out of 3D cache that other games do. It's about using the extra cache well and for the right code segments. If they go over the increased available cache there's no benefit and you can see that in the benchmarks. By optimisation I'm not talking about making the code smaller by the way, it's sending the right code segments into the cache rather than pools of large one-use code chunks which have no impact on performance.
Digital Foundry did a decent video on it, and how ideally developers can optimise what goes into the cache to get the benefit. Ideally the continuously used code would be permanently in cache and the rest sent to DRAM.
Negotiator
Banned
This discussion reminds me of Cell SPEs local memory + DMA transfers. Back then, developers had 100% control of the hardware (but few made use of it).
Caches are for the most part automatic, with minimal dev intervention (if any)...
Told y'all it was an unpopular take
fart town usa
Banned
It's gonna take alot to get me to buy a PS6.
3rd parties will release a PS5 version and Sony hardly supports the PS5 now, PS6 will just be more remasters.
3rd parties will release a PS5 version and Sony hardly supports the PS5 now, PS6 will just be more remasters.
reinking
Gold Member
I do believe there will be a PS6 but the post is all for fun.Told y'all it was an unpopular take
Imtjnotu
Member
Physiognomonics
Member
No. Why repeat failure? According to rumors even MS aren't going to repeat their mistake. If they (MS and Sony) do a lite machine it will be a handheld like the SteamDeck.
I agree with your 2/. We can already see AMD not even talking about Flops with RDNA4 (and neither the gamers). But on the other hand, they are talking TOPs right now... I think the flopwar is going to be replaced with topwar.
This is basically what PS5 is right now. A handy PC box with time exclusivity for some games.
fart town usa
Banned
Exactly. It doesn't bother me any because I mainly game on a 1080p plasma and I'm perfectly content with the games I end up playing.
Funny thing is that my wife was going to surprise me with a Pro for my 40th, I actually asked her not to because it's a waste of money, lol. For my purposes at least. I got a Quest 3s instead. Less than half price of a Pro and I'll get more use from it.
Audiophile
Member
My latest bit of "SPECulation" (..more hopes than reality , click each for full-size):
Edit (2): Adjusted ROPs, RAM, Nodes, CPU & Style.
, click each for full-size):Edit (2): Adjusted ROPs, RAM, Nodes, CPU & Style.
Last edited:
Shaki12345
Member
PS5 will fail massively if Sony can't get the price under 699.Sort of thing I'd live to see:
'28
'32
I know, I'm a dreamer..
bitbydeath
Member
Launching two different SKU's is a recipe for disaster.My latest bit of "SPECulation", that's more hopes than reality:
'28
'32
I know, I'm a dreamer..
64gigabyteram
Reverse groomer.
under 499 consoles? in THIS economy?
Shaki12345
Member
Read my post again bro.
Kerotan
Member
700 year one and it sells out first 12 month's no problem. Sony won't be gifting scalpers free money this time.
Loxus
Member
My hopes and dreams.
2 chiplets, a compute chiplet and I/O chiplet.
Compute Chiplet: N2P
• CPU: 12-core Zen 6c / 24MB L3 / 4.5 GHz
• GPU: 60CU RDNA 5 / 3.2 GHz
I/O Chiplet: N3P
• CPU: 2-core Zen 6LP
• NPU: XDNA 4 / 24 AI Tiles (Used for OS AI Assistant)
• MALL: 16MB
• Integrated SSD Controller: PCIe Gen 5
• 12× Flash PHY
• 8× 32-bit GDDR7 PHY
• 1× DDR6 PHY
• Display/Media Engine
• IO-Misc: Blu-ray/WiFi/Bluetooth/LAN
• PCIe/USB/HDMI PHY
Memory: GDDR7
• GDDR7: 24GB / 256-bit bus / 1280 GB/s
• DDR6: 4GB / 135 GB/s (Used as cache)
Base on AMD's Ryzen AI Max+ PRO 395
(Strix Halo)
RDNA5 is still reference internally,
UDNA seems like the marketing name.
So either or name may be used.
2 chiplets, a compute chiplet and I/O chiplet.
Compute Chiplet: N2P
• CPU: 12-core Zen 6c / 24MB L3 / 4.5 GHz
• GPU: 60CU RDNA 5 / 3.2 GHz
I/O Chiplet: N3P
• CPU: 2-core Zen 6LP
• NPU: XDNA 4 / 24 AI Tiles (Used for OS AI Assistant)
• MALL: 16MB
• Integrated SSD Controller: PCIe Gen 5
• 12× Flash PHY
• 8× 32-bit GDDR7 PHY
• 1× DDR6 PHY
• Display/Media Engine
• IO-Misc: Blu-ray/WiFi/Bluetooth/LAN
• PCIe/USB/HDMI PHY
Memory: GDDR7
• GDDR7: 24GB / 256-bit bus / 1280 GB/s
• DDR6: 4GB / 135 GB/s (Used as cache)
Base on AMD's Ryzen AI Max+ PRO 395
(Strix Halo)
RDNA5 is still reference internally,
UDNA seems like the marketing name.
So either or name may be used.
Last edited:
ExtendEmbraceExtinguish
Member
This thing will be 700 usd with Sony taking a hit. If they have the software it won't matter.
This console will be Sony's moment to inform us who exactly their customers are; casuals or core.
Shaki12345
Member
I wonder what's even the point of a PS6 since PS5 is still hungering for games.
Loxus
Member
By the time the PS6 releases, the PS5 hardware would be 7-8 years old. It would be time for new hardware regardless of what anyone thinks.
Games would still release on the PS5 like it did with previous generations anyway.
This will also partly depend on who/if they believe their competitors will be. My guess is on the continuation of relatively powerful console hardware for core fans, since they have dominated that segment for years. Current Sony management also seem to favour profitability over growing the user base (casuals). As evidenced by their hesitation to price PS5 more aggressively in its 2nd half of live.
awwr999999
Member
ExtendEmbraceExtinguish
Member
Threat from competitors. Switch 2 and potential oem nextbox. Remember it's about the ecosystem.
It still boils down to games.
awwr999999
Member
No, but RISC is an obvious fit. Both the PS2 and PSP already use MIPS and it's open source and free of any patent fees.
In the long run PlayStation consoles can't afford not to be running proprietary Sony/Sony HW.
At this point the move to AMD/AMD x86 has bought PlayStation enough time to R&D their way off of x86 and AMD.
Sony could improve upon the existing Emotion Engine as opposed to starting fresh.
A modern MIPS CPU with EE backwards compatibility and a proprietary Sony GPU with GS backwards compatibility would easily be the most exciting thing PlayStation has ever done.
awwr999999
Member
PlayStation being driven by competition brought us the PS3.
PlayStation competing with itself (PS1) gave us the PS2.
PlayStation can't afford another PS3.
PlayStaytion
Member
Audiophile
Member
My thoughts with the mockups I did above is how to make something that scales across a range without hamstringing it too much. I too am sceptical about a multi-SKU approach, but I think it's because the XSS vs XSX gulf was just too much. Having a GPU with 33% of the power, different/lower clocks on the CPU & GPU, 60% of the RAM and worst of all: 40% of the bandwidth!! ..was just nuts, I don't know what they were thinking..
In my eg. above, it'd be modular/scalable with chiplets, where each console is fundamentally identical in featuresets and the main board etc., they'd share the same CPU with exception to some extra V-cache added on the main system, then some more for the Pro. The "lite" would use 1 Graphics Core Die while the main system would use 2 of the same GCD. Only to make up a little of the gulf, the higher yield units will be fully utilised in the lite and be fully activated to offer 57.5% of the power rather than just 50% (plus with lower thermals, they could just peg the continuous boost scheme higher, or even go the opposite way and use it to provision with a more conservative thermal design), meanwhile the 2 GCDs in the main system will have a 10% redundancy, using lower yield versions of the same chips. Then the MCD's containing the memory bus and infinity cache will scale with the chips. Then on the latter Pro, it'd be the same approach albeit with upgraded GPU architecture, add one GCD, increasing the power by ~50%, add an MCD to increase L3 cache and the memory bus by 50%. Keep the CPU CCD same from the beginning to the end on the same process to save money, keep the MCDs on ~N6 for the first console lineup, only upgrading when it's viable to do so. By saving in all of these areas, you can focus on using a cutting edge process on the GPU GCDs where it's most valuable, getting backside power delivery + full gaafet will allow it to be pushed harder and/or use less power, which has knock on performance gains and/or cost savings. Not to mention if both are fully designed for, there may be an advantage in terms of topology and area.
Given that it's chiplets, if they have an excess of high yield PS6 Lite level chips, low yield PS6 base level chips or any even lower yield chips that can't be used, they can sell them back and AMD can repurpose them for any number of non standard products, same goes for the CPU dies and the MCD dies. When it's all modular, little has to go to waste.
Unlike the XSX/XSS gulf, the divide here for the base system and lite would be 58% GPU power, 72% Usable RAM, 70% Memory Bandwidth, everything else the same. You'd save money on 1GCD vs 2, you'd save money on no 3D V-Cache, you'd save money on half the storage, you'd save money on cooling and PSU, you'll get a smaller, slightly lighter chassis and you'd save on weight/size for shipping. You could use an identical board labeith with less/more VRMs, RAM chips and different density NVMe chips. But all the core features and architectural advantages are there, and the trade-off could -- as an eg. -- be something along the lines of a PS6 base game running 1296-1536p with full fledged upscaling and lots of rays cast, vs say 1080-1152p with slightly less comprehensive scaling and less rays cast. It lowers the barrier of entry significantly but the downgrade isn't excessive.
Also, XSS/XSX are two distinct APUs with ~half the total volume of PS5's APUs (and obviously much less by themselves). Here they'd have the economies of scale on their side not just by being the likely market leader again, but the dies will be shared between everything, with exception to the eventual Pro using an updated GCD die.
The advantage I see here is that with the lower barrier of entry it gets more people over to the new platform, incentivising a somewhat quicker switch over to next gen only development. As opposed to PS5 remaining the choice for the budget conscious and then remaining a necessary platform for too long. I'd rather a fundamentally next gen console with some visual features/quality dialed back be the limiting factor to new games rather than one that's significantly less powerful with 8yr old architectural features; and which may fundamentally alter the makeup of any ambitious games... All that said, if we do see another ~4yr predominantly cross-gen period, it's not gonna be quite as bad as the last one, given that what was holding us back before on PS4/X1 was a godawful underclocked 2012 era netbook CPU and a mechanical HDD with high latency and a meagre guaranteed throughput of ~20MB/s. Whereas with PS5 has an 'adequate' slightly underclocked desktop CPU and a great, low-latency 5.5GB/s RAW SSD with a solid I/O stack and "oodles" (lol) of decompression capability.
In my eg. above, it'd be modular/scalable with chiplets, where each console is fundamentally identical in featuresets and the main board etc., they'd share the same CPU with exception to some extra V-cache added on the main system, then some more for the Pro. The "lite" would use 1 Graphics Core Die while the main system would use 2 of the same GCD. Only to make up a little of the gulf, the higher yield units will be fully utilised in the lite and be fully activated to offer 57.5% of the power rather than just 50% (plus with lower thermals, they could just peg the continuous boost scheme higher, or even go the opposite way and use it to provision with a more conservative thermal design), meanwhile the 2 GCDs in the main system will have a 10% redundancy, using lower yield versions of the same chips. Then the MCD's containing the memory bus and infinity cache will scale with the chips. Then on the latter Pro, it'd be the same approach albeit with upgraded GPU architecture, add one GCD, increasing the power by ~50%, add an MCD to increase L3 cache and the memory bus by 50%. Keep the CPU CCD same from the beginning to the end on the same process to save money, keep the MCDs on ~N6 for the first console lineup, only upgrading when it's viable to do so. By saving in all of these areas, you can focus on using a cutting edge process on the GPU GCDs where it's most valuable, getting backside power delivery + full gaafet will allow it to be pushed harder and/or use less power, which has knock on performance gains and/or cost savings. Not to mention if both are fully designed for, there may be an advantage in terms of topology and area.
Given that it's chiplets, if they have an excess of high yield PS6 Lite level chips, low yield PS6 base level chips or any even lower yield chips that can't be used, they can sell them back and AMD can repurpose them for any number of non standard products, same goes for the CPU dies and the MCD dies. When it's all modular, little has to go to waste.
Unlike the XSX/XSS gulf, the divide here for the base system and lite would be 58% GPU power, 72% Usable RAM, 70% Memory Bandwidth, everything else the same. You'd save money on 1GCD vs 2, you'd save money on no 3D V-Cache, you'd save money on half the storage, you'd save money on cooling and PSU, you'll get a smaller, slightly lighter chassis and you'd save on weight/size for shipping. You could use an identical board labeith with less/more VRMs, RAM chips and different density NVMe chips. But all the core features and architectural advantages are there, and the trade-off could -- as an eg. -- be something along the lines of a PS6 base game running 1296-1536p with full fledged upscaling and lots of rays cast, vs say 1080-1152p with slightly less comprehensive scaling and less rays cast. It lowers the barrier of entry significantly but the downgrade isn't excessive.
Also, XSS/XSX are two distinct APUs with ~half the total volume of PS5's APUs (and obviously much less by themselves). Here they'd have the economies of scale on their side not just by being the likely market leader again, but the dies will be shared between everything, with exception to the eventual Pro using an updated GCD die.
The advantage I see here is that with the lower barrier of entry it gets more people over to the new platform, incentivising a somewhat quicker switch over to next gen only development. As opposed to PS5 remaining the choice for the budget conscious and then remaining a necessary platform for too long. I'd rather a fundamentally next gen console with some visual features/quality dialed back be the limiting factor to new games rather than one that's significantly less powerful with 8yr old architectural features; and which may fundamentally alter the makeup of any ambitious games... All that said, if we do see another ~4yr predominantly cross-gen period, it's not gonna be quite as bad as the last one, given that what was holding us back before on PS4/X1 was a godawful underclocked 2012 era netbook CPU and a mechanical HDD with high latency and a meagre guaranteed throughput of ~20MB/s. Whereas with PS5 has an 'adequate' slightly underclocked desktop CPU and a great, low-latency 5.5GB/s RAW SSD with a solid I/O stack and "oodles" (lol) of decompression capability.
Last edited:
Solarstrike
Gold Member
Fukt yeah let's go
Audiophile
Member
Man, I need Stray 2!
Fukt yeah let's go
Similar threads
- Locked