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

[rntongo]

Methinks the next consoles will use a 192bit bus with 4GB GDDR7 modules, reaching an overall 24GB RAM.

A 256-bit Bus/32GB VRAM combo I think would explode prices.

I dont think developers would be happy with 24GB of RAM. 4k and 8k textures, real time raytracing, ML based upscaling will require additional memory and memory bandwidth for a true generational leap. So I dont see this 24GB happening, its too little. There are no signaling issues with a 256 bit or 320 bit bus as well so it makes no sense to go with 192 bit bus. Me thinks 256 Bit bus width with 8 modules at 4GB each would be the absolute minimum. So around 32GB of memory with ~1TB/s memory bandwidth. But its possible even with 40GB of RAM on a 320 bit bus at 36Gbps per pin data rate would be around ~1.4TB/s memory bandwidth. And if by 2028 GDDR7 has a higher per pin data rate of 40Gbps(assuming not yet 48Gbps max spec) then it could even be anywhere between ~1.2TB/s to ~1.6TB/s memory bandwidth if they choose to go with the 320 bit bus. So minimum 32GB of RAM I think max 40GB of RAM at ~1.6TB/s memory bandwidth. Over the lifetime of 7 years for next gen hw this will deliver a true generational leap at a low cost. GDDR7 is already available today 3 years before 2028 at around 36Gbps data rate.

 

[QPlayer]

I dont think developers would be happy with 24GB of RAM. 4k and 8k textures, real time raytracing, ML based upscaling will require additional memory and memory bandwidth for a true generational leap. So I dont see this 24GB happening, its too little. There are no signaling issues with a 256 bit or 320 bit bus as well so it makes no sense to go with 192 bit bus. Me thinks 256 Bit bus width with 8 modules at 4GB each would be the absolute minimum. So around 32GB of memory with ~1TB/s memory bandwidth. But its possible even with 40GB of RAM on a 320 bit bus at 36Gbps per pin data rate would be around ~1.4TB/s memory bandwidth. And if by 2028 GDDR7 has a higher per pin data rate of 40Gbps(assuming not yet 48Gbps max spec) then it could even be anywhere between ~1.2TB/s to ~1.6TB/s memory bandwidth if they choose to go with the 320 bit bus. So minimum 32GB of RAM I think max 40GB of RAM at ~1.6TB/s memory bandwidth. Over the lifetime of 7 years for next gen hw this will deliver a true generational leap at a low cost. GDDR7 is already available today 3 years before 2028 at around 36Gbps data rate.

AMD has just released a new generation of GPUs with such efficient memory management that they found 16GB of GDDR6 to be enough for their cards. According to benchmarks, they are right, there is no significant advantage to more RAM for now.

A PC equipped with a 9070XT VGA and a modern CPU can do 4-500% more than today's consoles. This is a generational leap. Since we know well that such a PC configuration costs at least 1200-1500 dollars, it makes no sense to expect a more powerful console from any manufacturer within 1-2 years. Component prices are not going to go down, and no one is going to buy a 1000+ dollar console that you can only play games on.

 

[Pinstripe]

With neural compression you can mitigate that.

I think they'll go with a smaller bus, otherwise it's gonna be a $1000 console.

 

[thicc_gaf]

They want to see the 100+ million gamers on their Game Pass service, that's what I meant. And these gamers will probably be reached on traditional PCs with a modernized Game Pass integrated into the Xbox OS. XboxPCs in console form will be located within the PC market, their role will primarily be to retain users who currently have Xbox consoles.

For all this, they clearly need to come up with a very good software/business model that is better than the current one.

Doesn't that rely on PC Game Pass growth? So again it comes back to, what is going to drive that growth?

MS are flirting about a hybrid PC type of model (both for the hardware and business/sales aspects), but those devices are going to be considerably more expensive than the current traditional consoles, since OEMs will need good margins on hardware to make the initiative worth it, and MS want to provide buffer on pricing with their own hardware variants if in fact they manufacture some systems of their own (so they themselves can have profit margins on the hardware vs. selling at-cost or at a loss).

There is no way MS gets to 100 million Game Pass subscribers next gen simply with the hybrid devices, be they like consoles or portables or whatever. The volume of production simply won't be there, because the prices will limit demand which will limit supply. I'm also sure they're not expecting every Series S & X owner to upgrade to the new systems as some have likely already switched over to general PC, or Steam Deck, or PlayStation or Nintendo, or a combination of those.

It's probably safe to say at this point, as a decent guess, some 20 - 25% of Series S & X owners are not going to invest in a new Xbox device next gen, and getting PC Game Pass growth off them would depend on both those players going primarily Windows PC and seriously committing to Game Pass in addition to/in lieu of just using platforms like Steam. A very tough hill for MS to climb, no matter what, and the clock is ticking.

I think the probability of using HBM in PS6 is low, but not zero. HBM is expensive, but GDDR7 is not cheap and should be more power hungry than HBM. It would be too cost prohibitive to utilize both technologies. It will be one or the other.

PS5 pro has 16GB of GDDR6 with 576 BW and as a console it's pushing what I would consider a reasonable power consumption level. I would argue that next gen will need at least 2x bandwidth and 1.5-2.0x memory capacity so I think that forces at least GDDR7.

GDDR7 on a 256-bit bus with 3GB chips would give you a max of capacity of 24 GB and a max bandwidth of 1.5 TB/sec. But this bandwidth is likely past the optimal power efficiency point.
A HMB3e single stack of HMB3e could provide 24-36 GB of capacity with up to 1.2 TB/sec, but I bet at a drastically lower power consumption than GDDR7. ( GB200 in comparison has 8 HBM stacks )

By 2028, NVIDIA should be fully transitioned to HBM4 so maybe that will leave free up supply of HMB3e for other users...

Do you think it may be possible they could get away with 1 TB/s or even sub 1 TB/s (but close to it) bandwidth if they fitted the system with some form of HBM-PIM?

Because I've been thinking, with HBM-PIM they could cut down both in bandwidth and capacity requirement, although current PIM memory is mainly geared towards AI & HPC. Maybe it could make, say, a hypothetical PS6 w/ 24 GB HBM-PIM @ 1 TB/s possible without feeling constraining. Or even clocking the memory lower for say 960 GB/s (assuming 24 GB with 2.5 Gbps pins).

I'm not exactly sure what current HBM-PIM is based on tho; it could be HBM3, but it may also be HBM2e. I think even 2e would be enough for a PS6 depending on how the PIM component's implemented, which would require close working with AMD to optimize the APU design around it.

GDDR7 will be "cheap" (relatively speaking) once the industry transitions and economy of scale kicks in.



The thing to keep in mind is HBM brings higher packaging complexity and costs, and yield loss costs. It's not just a comparison of the memory chips itself compared to GDDR.

A long while ago there was talk about a cost-reduced version of HBM meant to cut out the interposer to make it easier for package integration and whatnot. Has there been any movement on that?

That's nonsense, why would you expect future compatibility for existing platforms? There's never been any expectation for future titles to work on past platforms. At the same time if APIs are that different (they aren't) then it's a case of straight up porting between platforms, something engines handle today for things with way more differences than what's considered here (e.g. PS5 vs XSX|S).

Well MS are the ones out here talking about forward compatibility and getting rid of generations, not me :/

But what I was mentioning isn't actually so much about Series systems playing the next Xbox's games, it's about the next Xbox having a way to run Xbox Series (and older) games, especially 360 & OG Xbox titles. Because AFAIK, you can't simply run a 360 or OG Xbox binary on a Win32 PC. There aren't any compatibility modes with that type of support and Xbox emulation isn't great (especially for OG Xbox).

Yet a lot of the Xbox fans who would be interested in that hybrid type of system have lots of old OG Xbox & 360 titles they want to bring with them. It's up to MS to sort out how that'll work.

 

[QPlayer]

Doesn't that rely on PC Game Pass growth? So again it comes back to, what is going to drive that growth?

MS are flirting about a hybrid PC type of model (both for the hardware and business/sales aspects), but those devices are going to be considerably more expensive than the current traditional consoles, since OEMs will need good margins on hardware to make the initiative worth it, and MS want to provide buffer on pricing with their own hardware variants if in fact they manufacture some systems of their own (so they themselves can have profit margins on the hardware vs. selling at-cost or at a loss).

There is no way MS gets to 100 million Game Pass subscribers next gen simply with the hybrid devices, be they like consoles or portables or whatever. The volume of production simply won't be there, because the prices will limit demand which will limit supply. I'm also sure they're not expecting every Series S & X owner to upgrade to the new systems as some have likely already switched over to general PC, or Steam Deck, or PlayStation or Nintendo, or a combination of those.

As I mentioned earlier, MS will probably not achieve a significant increase in Game Pass subscribers directly with hardware but with software. I also see what you mean, that Xbox PCs will only make up a slice of this pie. Yes, definitely at first. The focus could be on the Xbox OS, which will be released for almost all PCs. A console-style gaming interface that simplifies and makes it easier for many people to use games on PCs. Now you can say that this is just an assumption, but let's not forget that until yesterday the Handheld Xbox was just that, and the OEMs' strategy was also. You can say, okay, but this is already on Steam. Maybe, but what if it could be done even better, for example by integrating it into Windows. How many users can they address with that?

However, it is unquestionable that a better Game Pass and service package than the current one is needed for all this.

 

[rntongo]

AMD has just released a new generation of GPUs with such efficient memory management that they found 16GB of GDDR6 to be enough for their cards. According to benchmarks, they are right, there is no significant advantage to more RAM for now.

A PC equipped with a 9070XT VGA and a modern CPU can do 4-500% more than today's consoles. This is a generational leap. Since we know well that such a PC configuration costs at least 1200-1500 dollars, it makes no sense to expect a more powerful console from any manufacturer within 1-2 years. Component prices are not going to go down, and no one is going to buy a 1000+ dollar console that you can only play games on.

Developers are primarily the ones who determine what amount of RAM they need based off the feedback they give to Sony and MS, in addition to telemetry sent back from each devkit and console. And they have been clamoring for more RAM this gen. Add real time raytracing and the need for higher quality textures, larger more interactive worlds, doubling of RAM will be a minimum for a next gen device thats going to last 7-8 years. You're confusing PC and consoles. On PCs you can buy new RAM while consoles have fixed hw for 7-8 years. They'll likely have much more RAM and memory bandwidth than what most PC gamers have in their machines at the start of the gen. This was the case during the 8th gen as well as the 9th gen and will likely be the same during the start of the 10th gen. My estimate is between 32-40GB with memory bandwidth between 1.1-1.7TB/s if we have 40+Gbps GDDR7 by 2028.

 

[raytracingfan]

32GB memory is the minimum for a console to support running LLMs and other game logic AI models. Whether Sony and Microsoft actually want that ability remains to be seen. In regards to AI for graphics, upscaling decreases VRAM usage but frame generation increases it. Keeping highly detailed geometry in the BVH for ray tracing would also require a lot of memory, although we still don't know whether path tracing is a goal either.

 

[rntongo]

32GB memory is the minimum for a console to support running LLMs and other game logic AI models. Whether Sony and Microsoft actually want that ability remains to be seen. In regards to AI for graphics, upscaling decreases VRAM usage but frame generation increases it. Keeping highly detailed geometry in the BVH for ray tracing would also require a lot of memory, although we still don't know whether path tracing is a goal either.

Yes exactly well put. I dont see them doing anything less than this for a console thats going to be around for 7-8 years. I think this is going to be the biggest cost for the next gen systems after the APU. cost of SSDs will be quite low.

 

[fehu]

32GB memory is the minimum for a console to support running LLM

Can LLM run on a slower dedicated memory pool?

 

[raytracingfan]

Can LLM run on a slower dedicated memory pool?

Sure. Strix Halo mini-PCs and Macs with LPDDR5(X) can run LLMs just fine. Dedicating the slower pool specifically for AI doesn't really make sense though - if the pool is small it won't be enough, and if the pool is large developers that aren't using AI will want to use that memory for CPU tasks.

 

[Dominik D]

Because AFAIK, you can't simply run a 360 or OG Xbox binary on a Win32 PC.

I know that both have been hacked to oblivion but it would be extremely bizarre for MS to allow running X1/X360 on Windows which, unlike console, isn't a locked system. Beyond that there are technical reasons why 360 games can't natively run on Windows that have absolutely nothing to do with API (you can transpile binaries to win32 just fine). First: PC drives can't read Xbox 360 discs and this is a hardware limitation that's unlikely to change. Second: 360 has pretty unique design with EDRAM that enabled things that haven't been possible on PC until fairly recently.