Sony PS6, Microsoft neXt Series - 10th gen console speculation [2020]

[chris1515]

Why do devs want more CPU cores? What work is there that's better on 16 cores than a GPU if it's that parallelisable?

As I see it, you have three workloads - massive float power, ML work, and that ever doesn't fit into that. Silicon budget should be distributed accordingly to what yields best returns.

Because some workload well on CPU: Some part Physics engine, AI, gameplay code are on the CPU.

Doom Eternal will run at up to 1,000fps if your PC is up to it

Id Tech 6 maxed out at 250fps, but id Tech 7 engine can quadruple this with the right PC.

hexus.net

Like I said Id Tech 7 scale between 4 to 16 cores. New Decima physics jolt engine scale to 16 cores. Same for any game engine well optimize with ECS or job/fibers based like Bungie, Naughty Dog, Bluepoint or GG. All this engine are ready to scale to 16 cores and I am sure they aren't the only one...

Multithreading for game engines

Practical guide to vulkan graphics programming

vkguide.dev

Mostly since the ps4 generation of consoles, which have 8 very weak cores, architectures have evolved into trying to make sure all the cores are working on something and doing something useful. A lot of game engines have moved to a Task based system for that purpose. In there, you don’t dedicate 1 thread to do one thing, but instead split your work into small sections, and then have multiple threads work on those sections on their own, merging the results after the tasks finish. Unlike the fork-join approach of having one dedicated thread and having it ship off work to helpers, you do everything on the helpers for the most part. Your main timeline of operations is created as a graph of tasks to do, and then those are distributed across cores. A task cant start until all of its predecessor tasks are finished. If a task system is used well, it grants really good scalability as everything automatically distributes to however many cores are available. A great example of this is Doom Eternal, where you can see it smoothly scaling from PCs with 4 cores to PCs with 16 cores. Some great talks from GDC about it are Naughty Dog “Parallelizing the Naughty Dog Engine Using Fibers” 3 and the 2 Destiny Engine talks 4 5

Here code of Physics Engine scaling to 16 cores

GitHub - NVIDIA-Omniverse/PhysX: NVIDIA PhysX SDK

NVIDIA PhysX SDK. Contribute to NVIDIA-Omniverse/PhysX development by creating an account on GitHub.

github.com

GitHub - jrouwe/JoltPhysics: A multi core friendly rigid body physics and collision detection library. Written in C++. Suitable for games and VR applications. Used by Horizon Forbidden West.

A multi core friendly rigid body physics and collision detection library. Written in C++. Suitable for games and VR applications. Used by Horizon Forbidden West. - jrouwe/JoltPhysics

github.com

Doom Eternal scaling to a 16 cores CPU example with Digitalfoundry PC and this is maybe the only way to feed a Nvidia 4090. I would be curious to see framerate of Doom eternal with a 16 cores CPU and the 4090 probably approaching the 1000 fps limit. But they can do something heavier than Doom Eternal in the future.

Doom Eternal analysis: how id Tech 7 pushes current-gen consoles to the limit

Doom 2016 revitalised the fortunes of both id software and the classic Doom franchise, delivering a phenomenal, exhilar…

www.eurogamer.net

Of course, the PC version is the way to go if you have the hardware as it supports much higher frame-rates all around. I'm running a rather powerful rig equipped with a 16-core Intel i9 7960X and an RTX 2080 Tiso you'd expect greater performance and it's delivered without issue. In the marketing push up to launch, id promised astonishing performance and the team has delivered. In-game performance varied between 300 to 500 frames per second and Doom Eternal becomes one of the very few triple-A games that can actually deliver sustained, consistent performance for high refresh rate screens - up to and including the latest 360Hz displays.

I think it's way too early for 16 cores CPU, engines are not ready, don't need it, and it would be too expensive (APU size) for a console.

What they need to do is to improve current efficiency like for starters they could use a 8 cores CCX and more L3 cache on top of the usual tech improvements. For consoles and the specific requirement of always smaller APU a stacked L3 cache would be ideal!

Some engine are currently ready for 16 cores like Tiger engine of Bungie, Id tech 7, Naughty Dog engine, Decima Engine, Bluepoint game engine and so on. Id tech 7 showed it in 2020 with Doom Eternal.

EDIT:

ND Engine task based engine is ready since 2013/2014. What is working for 8 cores Jaguar, work too for 8 cores 16 threads of PS5/Xbox Series and can work for 16 cores 32 threads of next generation consoles.

Parallelizing the Naughty Dog Engine Using Fibers

This talk is a detailed walkthrough of the game engine modifications needed to make The Last of Us Remastered run at 60 fps on PlayStation 4. Topics covered will include the fiber-based job system Naughty Dog adopted for the game, the overall...

www.gdcvault.com

https://www.createursdemondes.fr/wp-content/uploads/2015/03/parallelizing_the_naughty_dog_engine_using_fibers.pdf

Bungie engine is ready since Destiny

And the engine not ready have now 6 years to do the job if next generation consoles release in 2028 and it will help the engine to run better on current gen console and it would have help engine to run better even on PS4/Xbox One.

 

[Globalisateur]

Because some workload well on CPU: Some part Physics engine, AI, gameplay code are on the CPU.

Doom Eternal will run at up to 1,000fps if your PC is up to it

Id Tech 6 maxed out at 250fps, but id Tech 7 engine can quadruple this with the right PC.

hexus.net

Like I said Id Tech 7 scale between 4 to 16 cores. New Decima physics jolt engine scale to 16 cores. Same for any game engine well optimize with ECS or job/fibers based like Bungie, Naughty Dog, Bluepoint or GG. All this engine are ready to scale to 16 cores and I am sure they aren't the only one...

Multithreading for game engines

Practical guide to vulkan graphics programming

vkguide.dev

Here code of Physics Engine scaling to 16 cores

GitHub - NVIDIA-Omniverse/PhysX: NVIDIA PhysX SDK

NVIDIA PhysX SDK. Contribute to NVIDIA-Omniverse/PhysX development by creating an account on GitHub.

github.com

GitHub - jrouwe/JoltPhysics: A multi core friendly rigid body physics and collision detection library. Written in C++. Suitable for games and VR applications. Used by Horizon Forbidden West.

A multi core friendly rigid body physics and collision detection library. Written in C++. Suitable for games and VR applications. Used by Horizon Forbidden West. - jrouwe/JoltPhysics

github.com

Doom Eternal scaling to a 16 cores CPU example with Digitalfoundry PC and this is maybe the only way to feed a Nvidia 4090. I would be curious to see framerate of Doom eternal with a 16 cores CPU and the 4090 probably approaching the 1000 fps limit. But they can do something heavier than Doom Eternal in the future.

Doom Eternal analysis: how id Tech 7 pushes current-gen consoles to the limit

Doom 2016 revitalised the fortunes of both id software and the classic Doom franchise, delivering a phenomenal, exhilar…

www.eurogamer.net

Some engine are currently ready for 16 cores like Tiger engine of Bungie, Id tech 7, Naughty Dog engine, Decima Engine, Bluepoint game engine and so on. Id tech 7 showed it in 2020 with Doom Eternal.

EDIT:

ND Engine task based engine is ready since 2013/2014. What is working for 8 cores Jaguar, work too for 8 cores 16 threads of PS5/Xbox Series and can work for 16 cores 32 threads of next generation consoles.

Parallelizing the Naughty Dog Engine Using Fibers

This talk is a detailed walkthrough of the game engine modifications needed to make The Last of Us Remastered run at 60 fps on PlayStation 4. Topics covered will include the fiber-based job system Naughty Dog adopted for the game, the overall...

www.gdcvault.com

Bungie engine is ready since Destiny

And the engine not ready have now 6 years to do the job if next generation consoles release in 2028 and it will help the engine to run better on current gen console and it would have help engine to run better even on PS4/Xbox One.

Doom Eternal is the worst game to use as a base to design next-gen CPUs IMO. Future of gaming is big open world engines needing high sustained I/O and RT requirements like Spider-man (or UE5). In those games adding CPU cores is not worth it as there are many others bottlenecks to be taken care off by APU designers (to improve performance) before stupidly adding more CPU cores.

Besides you don't add CPU cores when you have a CPU with ridiculous amount of L3 cache (which was not the case in any high-end CPU using the fastest available GPU). On a console (which uses laptop technology) Just adding more L3 cache would be way more cheaper for engines as seen with 5800x3d. I also doubt next-gen consoles will have 4090 level GPU, so the example is even more less relevant.

Sure at some point we'll have 16 cores for CPU on consoles, it's inevitable even, but it's way too early yet when current consoles still have inefficient 2 CCXs and ridiculously low amounts of L3 cache.

 

[chris1515]

Doom Eternal is the worst game to use as a base to design next-gen CPUs IMO. Future of gaming is big open world engines needing high sustained I/O and RT requirements like Spider-man (or UE5). In those games adding CPU cores is not worth it as there are many others bottlenecks to be taken care off by APU designers (to improve performance) before stupidly adding more CPU cores.

Besides you don't add CPU cores when you have a CPU with ridiculous amount of L3 cache (which was not the case in any high-end CPU using the fastest available GPU). On a console (which uses laptop technology) Just adding more L3 cache would be way more cheaper for engines as seen with 5800x3d. I also doubt next-gen consoles will have 4090 level GPU, so the example is even more less relevant.

Sure at some point we'll have 16 cores for CPU on consoles, it's inevitable even, but it's way too early yet when current consoles still have inefficient 2 CCXs and ridiculously low amounts of L3 cache.

Again the next generation is the last one and I talk about 4090 but it doesn't mean 16 cores CPU needs a 4090. For Doom eternal Digitalfoundry was using a 2080 Ti with the 16 core Intel CPU.

Any genre of games can use 16 cores not only Doom Eternal. High requirement I/O has no relationship with number of core...

Why do you think we will have the same amount of L3 on next generation console CPU?

My prediction would be something like this for next generation in 2028. For Tflops I was thinking 40 Tflops. But knowing we can probably go to higher frequency,I think we will have a 60/70 Tflops GPU in next generation consoles.

CPU 16 cores 32 threads SMT 3.5 to 4 Ghz
GPU 60/70 Tflops above 3 Ghz and lower than 4 Ghz
RAM 16 to 20 GB GDDR7 around 1 TB/s 32 Gbps or 36 Gbps
SSD 1.6 to 2 TB and speed at least 22 GB/s PCIE 5 or 6

EDIT: For Tflops count I think it will be like 7900 XTX or a little bit above and I continue to think we won't have more memory on next generation consoles or maybe a little bit more

 

[Shifty Geezer]

Because some workload well on CPU: Some part Physics engine, AI, gameplay code are on the CPU.

At the moment, because gameplay ML is in its infancy. But the speed-improvements of ML physics over classical solvers is how many orders of magnitude? If we want meaningful improvements of a real generation next hardware, we need a whole new way to tackle things, same as moving workloads onto compute.

 

[chris1515]

At the moment, because gameplay ML is in its infancy. But the speed-improvements of ML physics over classical solvers is how many orders of magnitude? If we want meaningful improvements of a real generation next hardware, we need a whole new way to tackle things, same as moving workloads onto compute.

I don't expect ML physics to be used for everything but where CPU or GPU aren't powerful enough to do it like complex cloth physics or fluid simulation or muscle simulation or very complex scene. I expect physics engine to be a mix between classical solver and ML physics.

 

[Globalisateur]

Again the next generation is the last one and I talk about 4090 but it doesn't mean 16 cores CPU needs a 4090. For Doom eternal Digitalfoundry was using a 2080 Ti with the 16 core Intel CPU.

Any genre of games can use 16 cores not only Doom Eternal. High requirement I/O has no relationship with number of core...

Why do you think we will have the same amount of L3 on next generation console CPU?

My prediction would be something like this for next generation in 2028. For Tflops I was thinking 40 Tflops. But knowing we can probably go to higher frequency,I think we will have a 60/70 Tflops GPU in next generation consoles.

CPU 16 cores 32 threads SMT 3.5 to 4 Ghz
GPU 60/70 Tflops above 3 Ghz and lower than 4 Ghz
RAM 16 to 20 GB GDDR7 around 1 TB/s 32 Gbps or 36 Gbps
SSD 1.6 to 2 TB and speed at least 22 GB/s PCIE 5 or 6

EDIT: For Tflops count I think it will be like 7900 XTX or a little bit above and I continue to think we won't have more memory on next generation consoles or maybe a little bit more

3.5ghz CPUs in 2028? Same frequency than now when it's obvious you get way more perf from your money with higher clocks (PS5 design) because of the increasingly higher cost of wafers? We got from 1.6ghz to 3.5ghz in one generation!

Adding 8 more caches will be completely unnecessary when it's obvious more dedicated hardware is already sorely needed elsewhere like AVX, decompression for I/O (PS5 already has that), but most importantly more ML hardware for true DLSS like reconstruction on consoles and finally improving current CPU performance (current CPUs use lowly mobile techs). I predict in next-gen APU they'll reduce the size allocated for CPU by %.

 

[chris1515]

3.5ghz CPUs in 2028? Same frequency than now when it's obvious you get way more perf from your money with higher clocks (PS5 design) because of the increasingly higher cost of wafers? We got from 1.6ghz to 3.5ghz in one generation!

Adding 8 more caches will be completely unnecessary when it's obvious more dedicated hardware is already sorely needed elsewhere like AVX, decompression for I/O (PS5 already has that), but most importantly more ML hardware for true DLSS like reconstruction on consoles and finally improving current CPU performance (current CPUs use lowly mobile techs). I predict in next-gen APU they'll reduce the size allocated for CPU by %.

You don't have real dedicated ML hardware in AMD GPU, they have some little ML accelerator in RDNA 3 sharing resources inside CU and I don't expect more. And again I take an easy bet, we will have before the end of the current generation a DL based reconstruction at least on Sony side* and I would be surprise to not see it one on Microsoft side. What Nvidia is doing on PC side is irrelevant for what will be done on consoles side.

I said between 3.5 and 4Ghz for the frequency maybe it will be a little more but we will see, same maybe we will have a good surprises with more than 4Ghz on GPU side but I prefer stay realist until I see above 4Ghz GPU.

* called DLAS and Sony Interactive Entertainment have a patent without the acronym or maybe they will change the name but it will be there after cross gen period probably and 2nd wave of games.

 

[Remij]

A couple of questions.

What real technical advantages would consoles with Unified Memory Architectures still have over a split memory PC architecture that could directly access storage from VRAM AND RAM without any need to copy data around between them?

And also..

What kinds of fundamental architectural improvements do you guys think the next consoles could improve upon? What else could be offloaded from the CPU for example?

Dedicated CPU and memory for all OS and secondary functions seems obvious.. but what else can they improve?

 

[Johnny Awesome]

Why not just make a 500 watt console for more power. What's the big deal? PC GPUs are getting there.

 

[techuse]

Why not just make a 500 watt console for more power. What's the big deal? PC GPUs are getting there.

Noise, size, heat output and cost.

 

[cheapchips]

Noise, size, heat output and cost.

But apart from those?

 

[BRiT]

Noise, size, heat output and cost.

There is a lot of space they could use even if they kept it the size of the Sumo PS5. That sucker is massive. Lots of dead space not used.

I'm talking keeping the console the same base 3D Size constrained by the outside bounding box but having smarter use of the internal dimensions for cooling and noise dampening.

 

[techuse]

There is a lot of space they could use even if they kept it the size of the Sumo PS5. That sucker is massive. Lots of dead space not used.

I'm talking keeping the console the same base 3D Size constrained by the outside bounding box but having smarter use of the internal dimensions for cooling and noise dampening.

Would that affect cost though?

 

[BRiT]

Would that affect cost though?

Most likely it would.

 

[Jofan]

Could be wrong, but I'm sure there was a discussion on angstroms going on in here, so thought I'd post this.
Intel Paves The Way To A Trillion Transistors In Next Gen Chips By 2030 - featuring innovative new materials just 3 atoms wide, pretty remarkable

Moore's Law Lives On: Intel Paves Way To A Trillion Transistors In Next-Gen Chips By 2030

At IEDM, Intel Research has showcased how Moore's Law is Alive & how Chipzilla plans its next-gen chips with Trillion transistors by 2030.

wccftech.com

 

[Tahir2]

By 2026/7 the CPU will be at 5GHz Peak Frequency in consoles and the GPU at 3.2GHz in my estimation. This is cheaper than going for a wider CPU and GPU or using a lot of cache as on desktop CPUs.

Hopefully AMD will improve its RT ability and FSR 2.0 seems to be working fine and should be much better in the future.

Hoping we don't start targeting 8K in game engines. The sweetspot in resolution to screen size to distance viewed is 4K and hopefully remains so by 2026/7.

The major issues for the next gen will be power consumption and the cost of the wafers.
..

 

[%zXÁïæeÑ9ùPÔFPZ"D;»Ä]

Hopefully technologies like DLSS3 frame generation (FG) matures and further research is made on decoupling viewport manipulation from game render (async reprojection). This would allow games to always output consistent vsynced 120fps or what ever locked fps for consistent image persistence and would enable the possibility to use display technologies like black frame insertion (BFI) or other strobing mechanisms reliably without crosstalk on any game. Even if underlying render loop only runs for example 45fps (without FG), async reprojection would make the input feel like 120fps.

Lots of unanswered questions and research to be done but my main point is next generation is more about innovation on various aspects in the full software stack than raw hardware breakthroughs.

 

[cheapchips]

(Figured I'd use this thread, or do we envisage having an annual thread for Gen 10 speculation?)

In the ABK case it's come light that MS don't see gen 10 starting until at least 2028. Sony have already said after 2027.

TSMC's roadmap would put them on 2nm or possibly 1.4nm in that timeframe.

 

[Seanspeed]

(Figured I'd use this thread, or do we envisage having an annual thread for Gen 10 speculation?)

In the ABK case it's come light that MS don't see gen 10 starting until at least 2028. Sony have already said after 2027.

TSMC's roadmap would put them on 2nm or possibly 1.4nm in that timeframe.

Hopefully Intel is up and running for customers by that point, and maybe even Samsung could get things together in order to provide decent alternatives to help keep costs down.

Though I wonder how it would work in terms of using AMD IP on an Intel fab. You'd think it wouldn't matter much, especially as it's not actual AMD products, but I dont know if there'd maybe be some branding sensitivity there or something.

 

[cheapchips]

Hopefully Intel is up and running for customers by that point, and maybe even Samsung could get things together in order to provide decent alternatives to help keep costs down.

Though I wonder how it would work in terms of using AMD IP on an Intel fab. You'd think it wouldn't matter much, especially as it's not actual AMD products, but I dont know if there'd maybe be some branding sensitivity there or something.

Wouldn't have thought it would matter much, if Intel are serious about being a fab for everyone.

More interesting to me is if Intel are in a position to shop around a console SoC in the next few years. Mostly this is due to XeSS and dedicated ML / ray tracing being interesting features vs knowing nothing about how RDNA will evolve.

Not really sure it's a worthwhile business venture for Intel, before you get into risks.