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

[rntongo]

It is possible that their own consoles will come later, that is what the news says, but OEM hardware is much sooner than they are releasing specific information about this.

I don't agree that we should wait years, many people want a new Xbox, even if it is a PC hybrid. I have read more serious forums and the general feedback is that many people want to upgrade their Xbox, but since MS doesn't have a mid-gen console, the idea of XboxPC is becoming more and more popular. That is why MS deals with third-party OEM hardware. And this has now been proven.

If they release the third-party Handheld this year, they can also release the desktop versions, especially if customer feedback is good. I would bet that they will do this quite soon.

Okay that makes more sense, some sort of Xbox licensed devices for those that want to play current gen titles at home and with higher graphical settings. On the other hand, Jez was referring to Xbox's next gen offering. He purports its been approved by Satya Nadela since such systems take at least 3 years or more to develop I strongly believe Jez here. He's in the know and such news should be coming out this year about early development for next gen devices.

 

[rntongo]

What would those memory changes be? HBM is out of the question, and GDDR7 just makes sense to me.

Almost certainly GDDR7 but there will be extra optimizations to increase memory bandwidth for Machine Learning and RT workloads. Thats going to be a major focus.

 

[Charlietus]

For me the timelines were separate and then crossed paths.

- AMD was developing an AI solution for its FSR upscaling. RDNA 3 incorporated some matrix accelerator instructions that are important for achieving machine learning upscaling. This to me indicates that there was already an intention to do this probably in FSR3.
- Sony was developing an AI solution for the PS5 Pro, likely based on RDNA 3.
- Their paths crossed when Sony noticed that AMD's solution was better, despite being computationally heavier. So they joined forces to share models to improve both companies' solutions.

AMD started to work on AI FSR at the start of 2024 or so, as said by a AMD rapresentative. Apparently FSR 4 is the result of a year of collaboration with Sony, which is really fast for such good results.

 

[Kaotik]

- Sony was developing an AI solution for the PS5 Pro, likely based on RDNA 3.

The PS5 Pros "technology that's not in any AMDs own products yet" is(was) quite surely RDNA4s AI accelerators in the CUs (minus sparsity support) so why would they develop it based on RDNA 3?

 

[Quadbitnomial]

For me the timelines were separate and then crossed paths.

- AMD was developing an AI solution for its FSR upscaling. RDNA 3 incorporated some matrix accelerator instructions that are important for achieving machine learning upscaling. This to me indicates that there was already an intention to do this probably in FSR3.
- Sony was developing an AI solution for the PS5 Pro, likely based on RDNA 3.
- Their paths crossed when Sony noticed that AMD's solution was better, despite being computationally heavier. So they joined forces to share models to improve both companies' solutions.

The timelines are not separate.

- At no point was FSR3 ever machine learning based. RDNA 3 added matrix acceleration to their consumer GPUs because Nvidia had it and AI is big in prosumer market.
- It was the work Sony and AMD did with PS5 Pro and PSSR that made AMD pivot to machine learning for FSR4, which is how Project Amethyst was formed in 2023.

PS5 Pro project started in 2020 and in 2021 PSSR fully kicked off.

Mark Cerny Dec 18, 2024: The hardware definitely came first, because you can't know that your neural network is implementable unless you understand a bit about the hardware that it's going to be running on top of. So, simply put, in 2020 we did early hardware work. In 2021, we did more detailed hardware work and fully kicked off the PSSR project.

In 2023 the PS5 Pro development was concluding and that was when Project Amethyst was formed.

Mark Cerny Dec 18, 2024: Well, we've already started talking. Pretty much when PlayStation 5 Pro was wrapping up, we began discussions and development of those various aspects of Amethyst. To get more specific, I think we were getting pretty good momentum by the end of 2023. But as you know, it takes a phenomenally long amount of time to go from discussions and concepting to actually having a piece of hardware out in the marketplace.

Which also lines up with when Jack Huynh VP/GM of AMD Computing and Graphics Business Group pivoted the FSR team towards developing AI based FSR4

Jack Huynh September 13, 2024: Because FSR2 and FSR3 were analytical based generation. It was filter based. Now, we did that because we wanted something with a very fast time to market. What I told the team was, "Guys, that's not where the future is going." So we completely pivoted the team about 9-12 months ago to go AI based.

In the end of 2024 PS5 Pro and PSSR is released, and now FSR4 has launched which is the result of the collaboration with AMD and Sony. Current PSSR is a entirely Sony design AI upscaler, PSSR2 which will be similar to FSR4 is a jointly development from Project Amethyst.

 

[Sega_Model_4]

The PS5 Pros "technology that's not in any AMDs own products yet" is(was) quite surely RDNA4s AI accelerators in the CUs (minus sparsity support) so why would they develop it based on RDNA 3?

To accelerate PSSR, the PS5 Pro uses WMMA instructions derived from RDNA 3. Cerny calls the PS5 Pro GPU RDNA 2.X, as it contains RDNA 2 for raster, WMMA instructions from RDNA 3, and RT implementations from RDNA 4.

The timelines are not separate.

- At no point was FSR3 ever machine learning based. RDNA 3 added matrix acceleration to their consumer GPUs because Nvidia had it and AI is big in prosumer market.
- It was the work Sony and AMD did with PS5 Pro and PSSR that made AMD pivot to machine learning for FSR4, which is how Project Amethyst was formed in 2023.

PS5 Pro project started in 2020 and in 2021 PSSR fully kicked off.


In 2023 the PS5 Pro development was concluding and that was when Project Amethyst was formed.


Which also lines up with when Jack Huynh VP/GM of AMD Computing and Graphics Business Group pivoted the FSR team towards developing AI based FSR4


In the end of 2024 PS5 Pro and PSSR is released, and now FSR4 has launched which is the result of the collaboration with AMD and Sony. Current PSSR is a entirely Sony design AI upscaler, PSSR2 which will be similar to FSR4 is a jointly development from Project Amethyst.

It doesn't make sense to me that AMD didn't think of a machine learning solution for its upscaling when including hardware for this in its GPUs.

For me, AMD has always intended to do something with machine learning. GPU Open has had a lot of articles for a long time about different ways to use Radeon for AI, there is even an SDK, Radeon ML. Sony may have planned PSSR in 2021, but it obviously already had in mind the hardware on which it would base the AI training, otherwise, where would it do it? It would be inefficient to do it on different hardware, it goes beyond the purpose of a console.

In my opinion, the paths were somewhat separate until AMD's research yielded better results, which made Sony adapt what they had already been developing to unify efforts. In the interview with Cerny, this is clear to me, but it seems to me people are trying to build a story where Sony is responsible for everything that AMD brings.

 

[Quadbitnomial]

It doesn't make sense to me that AMD didn't think of a machine learning solution for its upscaling when including hardware for this in its GPUs.

Nobody said they weren't thinking about ML for upscaling. They just were not actively working on it as they were focused on analytical based FSR solutions that works across their whole portfolio of gaming GPUs from RX 6000 down to RX 460.

For me, AMD has always intended to do something with machine learning. GPU Open has had a lot of articles for a long time about different ways to use Radeon for AI, there is even an SDK, Radeon ML.

AMD makes GPUs for ML training and inference in the professional sector so of course they have all sorts of ML libraries and SDKs but it does not mean they were working on AI upscaling for games.

Sony may have planned PSSR in 2021, but it obviously already had in mind the hardware on which it would base the AI training, otherwise, where would it do it? It would be inefficient to do it on different hardware, it goes beyond the purpose of a console.

That is why in 2020 they designed the specification for the hardware in house using AMD RDNA as the building block, it is not based on RDNA3 CU even if they might seem similar. Its not the first time we've seen Sony do that.

In my opinion, the paths were somewhat separate until AMD's research yielded better results, which made Sony adapt what they had already been developing to unify efforts.

Your opinion does not line up with the facts. I have laid down the timeline of events that has led to FSR4. At no point was AMD developing an ML based upscaler prior to PSSR.

AMD FSR 1.0 (spatial upscaler) released on 22nd June 2021

PSSR (ML upscaler) starts development in late 2021

AMD FSR 2.0 (temporal upscaler) released on 12th May 2022

AMD FSR 3.0 (temporal upscaler and frame generation) released on ‎29th September 2023

Project Amethyst starts in 2023

PSSR launches in 7th November 2024

AMD FSR 4 released March 6, 2025

Again AMD VP/GM of Computing and Graphics Business Group, Jack Huynh September 13, 2024:

Because FSR2 and FSR3 were analytical based generation. It was filter based. Now, we did that because we wanted something with a very fast time to market. What I told the team was, "Guys, that's not where the future is going." So we completely pivoted the team about 9-12 months ago to go AI based.

This was when project Amethyst was formed. FSR4 is the creation of AMD and Sony according to AMD themselves so we do not need to conject about it.

In the interview with Cerny, this is clear to me, but it seems to me people are trying to build a story where Sony is responsible for everything that AMD brings.

I don't see that but it is clear working with Sony is what pushed them to pivot the FSR team towards going AI based for FSR4 and the results will also be implemented in PSSR 2

 

[Kaotik]

To accelerate PSSR, the PS5 Pro uses WMMA instructions derived from RDNA 3. Cerny calls the PS5 Pro GPU RDNA 2.X, as it contains RDNA 2 for raster, WMMA instructions from RDNA 3, and RT implementations from RDNA 4.

Do you have some source for this? If it was just RDNA3 WMMA it would be nowhere near 300 TOPS (which is pretty much spot on TOPS per CU per MHz compared to RDNA 4 strongly suggesting same capabilities sparsity aside)

Project Amethyst starts in 2023

Your link is broken. Officially it was announced late 2024, not 2023.

 

[Quadbitnomial]

Your link is broken. Officially it was announced late 2024, not 2023.

Thanks here is the corrected link. Officially announced with PS5 Pro in 2024 but the project was started in 2023

Mark Cerny: Well, we've already started talking. Pretty much when PlayStation 5 Pro was wrapping up, we began discussions and development of those various aspects of Amethyst. To get more specific, I think we were getting pretty good momentum by the end of 2023. But as you know, it takes a phenomenally long amount of time to go from discussions and concepting to actually having a piece of hardware out in the marketplace.

 

[Panino Manino]

FSR4, the current version, is really possible on PS5Pro?
It seems they confirmed that PS5Pro has 300 TOPs, but I think I see somewhere AMD saying that FSR4 requires around 700 TOPs, which for me made sense why the PS5Pro came with PSSR instead of the new FSR4, because it's "lighter".
How would PS5Pro support FSR4? Why they need another year to offer support? Because they need to develop a FSR4 Light that uses less TOPs?

 

[Love_In_Rio]

FSR4, the current version, is really possible on PS5Pro?
It seems they confirmed that PS5Pro has 300 TOPs, but I think I see somewhere AMD saying that FSR4 requires around 700 TOPs, which for me made sense why the PS5Pro came with PSSR instead of the new FSR4, because it's "lighter".
How would PS5Pro support FSR4? Why they need another year to offer support? Because they need to develop a FSR4 Light that uses less TOPs?

700 TOPs will be a figure with sparsity. Cerny already said ML power of PS5 pro was in the ballpark of 9700`s without sparsity, that seems per Cerny of not much help with FSR4 algorithm. I supposse the adjustment comes migrating the transformer block from FP8 (that seems PS5 PRO hasnt) inference to INT8, using model quantization.

 

[Kaotik]

FSR4, the current version, is really possible on PS5Pro?
It seems they confirmed that PS5Pro has 300 TOPs, but I think I see somewhere AMD saying that FSR4 requires around 700 TOPs, which for me made sense why the PS5Pro came with PSSR instead of the new FSR4, because it's "lighter".
How would PS5Pro support FSR4? Why they need another year to offer support? Because they need to develop a FSR4 Light that uses less TOPs?

No, and they're not saying there will be FSR4 but something similar to it.
"our target is to have something very similar to FSR 4's upscaler available on PS5 Pro for 2026 titles as the next evolution of PSSR"

Mark Cerny: FSR 4 for PS5 Pro is the "next evolution of PSSR"

Last week, AMD released its new wave of graphics hardware, developed around the new RDNA 4 architecture. Machine learni…

www.eurogamer.net

 

[Sega_Model_4]

Nobody said they weren't thinking about ML for upscaling. They just were not actively working on it as they were focused on analytical based FSR solutions that works across their whole portfolio of gaming GPUs from RX 6000 down to RX 460.

AMD makes GPUs for ML training and inference in the professional sector so of course they have all sorts of ML libraries and SDKs but it does not mean they were working on AI upscaling for games.

That is why in 2020 they designed the specification for the hardware in house using AMD RDNA as the building block, it is not based on RDNA3 CU even if they might seem similar. Its not the first time we've seen Sony do that.


Your opinion does not line up with the facts. I have laid down the timeline of events that has led to FSR4. At no point was AMD developing an ML based upscaler prior to PSSR.

AMD FSR 1.0 (spatial upscaler) released on 22nd June 2021

PSSR (ML upscaler) starts development in late 2021

AMD FSR 2.0 (temporal upscaler) released on 12th May 2022

AMD FSR 3.0 (temporal upscaler and frame generation) released on ‎29th September 2023

Project Amethyst starts in 2023

PSSR launches in 7th November 2024

AMD FSR 4 released March 6, 2025

Again AMD VP/GM of Computing and Graphics Business Group, Jack Huynh September 13, 2024:

This was when project Amethyst was formed. FSR4 is the creation of AMD and Sony according to AMD themselves so we do not need to conject about it.
View attachment 13298

I don't see that but it is clear working with Sony is what pushed them to pivot the FSR team towards going AI based for FSR4 and the results will also be implemented in PSSR 2

They are using this tweet as an argument, when the post talks about the collaboration that will occur between Sony and AMD to share models. The text is in the future, it is something that will still happen.

Do you have some source for this? If it was just RDNA3 WMMA it would be nowhere near 300 TOPS (which is pretty much spot on TOPS per CU per MHz compared to RDNA 4 strongly suggesting same capabilities sparsity aside)


Your link is broken. Officially it was announced late 2024, not 2023.

RDNA 3 WMMA is just the technology, these are instructions for calculating matrices, how much performance you will get will depend on how much silicon is dedicated to this.
The source is the PS5 Pro Deep Dive, where Cerny said he merged features from RDNA 2 and 3, but does not use all the existing features in RDNA 3 to avoid incompatibilities.
It lacks RDNA 3 shader executables and double floating point calculations, so what's left are just the WMMA instructions. Other than that, RDNA 3 resembles RDNA 2.