Next Generation Hardware Speculation with a Technical Spin [2018]

[dten]

If the diffence is not big enough to be worth it, it will keep mostly recieving cross-gen titles anyway, and the first company to put itself in that position leaves the door opened for their competitors to wait it out a couple more years and release a more significantly superior kit, leaving your rushed product on a half-gen linbo.
It happened to dreamcast.

We are likely getting a lot of cross gen titles even if it were a Zen 2 to Jaguar difference. Publishers don't want to lose money on a small user base. RAM is often the biggest difference and 8 GB is still a lot today and in a couple of years. Modern developers know how to scale CPU/GPU well.

People expecting Zen 2 to save us will probably be disappointed.

 

[Shortbread]

I don't understand why the overal tech forum userbase is always disapointed with the specs of new gen systems, as they were with ps4/bone and yet the moment there is a rumor of a next gen they want their companies to rush them out with whatever barely distinctive tech they can put out...
What do you guys want?

Exactly. This is what pisses me off about some console gamers. They want godly levels of PC compute, but for $399. And when the systems are finally introduced, they love them for about year, then complain about the lack of IQ and performance. Maybe introducing a two system launch (entry and pro-level system) could resolve some of the disappointment. I'm pretty sure there are some hardcore console gamers willing to spend $699 or more on a top of the line gaming console (I would, and have).

 

[Tkumpathenurpahl]

Agreed. Especially with the two tier launch:

Base model - built to be cheap and set an adequate baseline for the generation.
Pro model - built for nerds like us, who give a shit about the likes of SVGI.

-- PSP4table --

Why are people saying it's all but impossible?

The 14nm laptop APU's we've seen so far top out at 11CU's, with both CPU and GPU capable of higher clockspeeds than the PS4.

If 7nm brings a 40% power usage reduction, clockspeeds are reduced to PS4 levels, GDDR5 is replaced with HBM, and the CU's are upped to 18, wouldn't we be looking at a system with laptop levels of power draw?

A 10" PS4 tablet doesn't seem so out of the question. If they can stick some hardware in there that allows it to stream to a wireless PSVR headset, they'd have a very compelling unit IMO.

 

[ultragpu]

OK let's say Sony does a Pro version at launch which costs $599 if not higher in late 2019, what would be its maximum specs for that price? Is it even possible to get close to twice the power of a 9TF base model?
I mean let's make it slightly bigger than the PS4Pro with a 200w TDP, using 7nm die shrink should in theory get you 20 TF of performance which is twice of a 14nm Vega 56 consuming 210w. But I guess real world doesn't performance yield is not as much is it?

 

[McHuj]

Nowadays I’m wondering if we could the first MCM design utilizing dual APUs (anyone remember the dual APU rumors prior to X1 announcement).

In theory an 8 core Zen plus an underclocked/derivative Vega 64 could fit into a sufficiently small APU for 8TFs of performance. “Duct tape” two of them in a package for the pro version.

Otherwise, I don’t really see a console SOC being feasible with 15 teraflops and a big CPU in the next two years.

I would like to see 8TF of performance for 1080p rendering.

 

[Tkumpathenurpahl]

Nowadays I’m wondering if we could the first MCM design utilizing dual APUs (anyone remember the dual APU rumors prior to X1 announcement).

In theory an 8 core Zen plus an underclocked/derivative Vega 64 could fit into a sufficiently small APU for 8TFs of performance. “Duct tape” two of them in a package for the pro version.

Otherwise, I don’t really see a console SOC being feasible with 15 teraflops and a big CPU in the next two years.

I would like to see 8TF of performance for 1080p rendering.

That's exactly the kind of split I'd like to see. With the talk behind Infinity Fabric, it seems more feasible now than ever before.

I think it also makes the base and Pro split a better sell: the Pro model can play two different instances of PS5 games, or one instance with über graphics.

A pricing split of $350 and $600 would make the Pro cheaper than two separate base models, and be an attractive prospect to multi gamer households (couples, parents, home shares.)

 

[ultragpu]

Ah the good old dual gpu rumor back in the days , yes I recall such a dream. Maybe it will work after all.

 

[Shortbread]

Dual APUs or GPUs will never happen in the modern console space, even if power constraints and TDP weren't a factor. Developers have pretty much made it clear, they want clean designs (architectures) that they can easily access and spend more time/effort towards game designs, rather than hardware eccentricities that a dual APU/GPU would introduce.

 

[Tkumpathenurpahl]

True, but what if it was effectively invisible to developers and they can just treat the 2 GPU's as one?

Certainly, AMD have been talking about manufacturing smaller chiplets and glueing them together with Infinity Fabric. So it's not unthinkable, but it depends on the limits of that tech, I suppose.

 

[liolio]

Dual APUs or GPUs will never happen in the modern console space, even if power constraints and TDP weren't a factor. Developers have pretty much made it clear, they want clean designs (architectures) that they can easily access and spend more time/effort towards game designs, rather than hardware eccentricities that a dual APU/GPU would introduce.

PC are clean. Lots of complains have more to do with historic events than technical matters.
Throughout the ps360 era devs and studio have been presented with massive challenges from multi-core programming on system to cores providing really poor performances on their own.
There is not only hardware the scale of the projects and iterations speed in many cases have exploded, so yes indeed no random shit, many sucky cores, exotic approach that have you to rethink your engines, etc. That being said the level of performances required from hardware in turn has created a convergence on solution. Pretty much we are 100% sure that we will deal with PC IP: nothing exotic rigth here.

Clearly an old school CPU+GPU could make sense especially if one want later on interate on the GPU part, it means a NUMA system and I don't see why that would be an issue. SOC works best for relatively low performance parts.
Imo it was not a good choice this gen for either Sony or MSFT but I would say for most for the later as as I see thing may the PS4 had stuck to 4GB of RAM it would have made ultimately not much difference on screen, still an extremely straight forward system. That would have gave them extra margins.
What I really dislike with MSFT choices is that it did cost them lots of costly silicon, failed to deliver a competitive edge while not saving them the wide bus and relatively fast and expensive memory.

I do indeed believe that it would be non sensical to have an double APUs set-up. As for an APU+GPU why not if you go with a custom APU with few silicon invested on the GPU (for the UI and the low power mode /kill the GPU all together). At this point I'm not sure it is sensical to designing a (really) custom GPU
It is diffferent for the CPU, AMD APU are big yet only offer 4 cores, on the other end the Threadrippers are big to while providing too many cores while exceeding a silicon budget that should go toward GPU an fast V-RAM.

EDIT
I believe I read your post too fast.

 

[McHuj]

Dual APUs or GPUs will never happen in the modern console space, even if power constraints and TDP weren't a factor. Developers have pretty much made it clear, they want clean designs (architectures) that they can easily access and spend more time/effort towards game designs, rather than hardware eccentricities that a dual APU/GPU would introduce.

The complexity of the hardware can be abstracted away. It would really be no different than the PS4 and the Pro. You'd really be using the other APU just for the GPU in the MCM. My whole thinking was based on silicon reuse and not having to design two SoC's, one for the pro and one of the base console. I think large dies are going to be difficult (or too expensive) to achieve in the next couple of years on 7nm.

So if the MCM approach has any feasibility, you have the choice of going to: an APU plus an additional GPU for the pro, two identical APU's (maybe even CPU is disabled in the second APU), discrete CPU and two different GPU's (one for the pro and one for the base), or finally CPU plus two identical GPUs in the MCM. Of those choices, the reuse of two identical APU's would seem to be most economical. But that's assuming an MCM approach will even be considered or realistic.

 

[iroboto]

Wondering what type of platform features are needed for next gen. We had a lot of discussion about what the hardware would do for the games, but not the platform.

My xbox is currently being occupied by my kids watching netflix, but with HT and enough CPU, it could be playing a game on another screen or streaming it elsewhere.

Just something I'm thinking about right now, trying to figure out if this is an important aspect for the platform developers. I know the common rebuttal is 'my smart tv has netflix, don't need it on xbox etc'. But smart tvs are (a) slow in general, and (b) only support the most popular media services. Take longer to update etc. We are still discounting blu ray playing while playing a game.

There is a cross over point for MS, if they have tons of CPU and memory such that one person could be using the device for games and the remaining power could be used for another app by another user. That would be an almost weird mainframe style model, where all the processing could occur on on a single device and it would stream out to other devices if needed. I don't see that as being prominent, but I can't help but see a future where there are multiple users on 1 device. Whether it be M/KB or controller. We're talking supporting multiple cursors and keyboards etc, so that everyone can take advantage of this one device.

There's too much computing power out there for all of it to be wasted on a single workstation. Usage graphs will show that PCs are unlikely to be really pushed to it's maximum capability unless doing heavy professional work and even then, only a handful of workstations are going to be pushing the box to load 24/7. We are good at pushing GPUs to their limits. But in a non niche task, we are not good at pushing CPUs to their limits. And we continue to see more innovation in moving over more CPU work to the GPU such that it continually frees up more CPU to do other things.

When I look at the stack for MS, I think we'll see Windows move in this direction. Users can build beefy machines that multiple users in a house hold can draw from, or a company etc. They could collaborate on a single screen locally or work on the device separately etc. But they could also have their own hardware to support it, and a powerful enough xbox may be able to do the trick to act as a central machine. It would be next gen for sure, and something I wouldn't expect until next gen. The question is whether they would go for that. Perhaps a pro model. But this time the differentiating factor is more GPU but instead it comes with more CPU and memory to support other users leveraging the hardware, the remaining game hardware is still 'locked' to perform the same as the base unit.

Perhaps what I'm getting at is, we talk about next gen too much in the context of games. But some of the real changes between generations have been platform improvements.

I don't suspect next gen will simply be just this gen with better games. There needs to be more to it to get people like myself to upgrade. And it can't just be 'becomes more PC like', they need to develop a whole solution that is easy and sensible and makes sense to use as an offering.

I think all platforms will need to think critically about this for next gen.

This generation has been great but still greatly paralyzed by install times; slow application loading etc. it's certainly something that can be improved upon. How to solve these issues I don't know, but if games could load and install faster, I would already consider that a win.

Consider the future of multiplayer in the same house hold.
1 console, but instead of 1 screen of 4K you have 4 instances of the same game running at the same time but at 1080p, streaming to 3 other users in your house and 1 on the main screen so that all four of you can play. That fixes a lot of the problem of needing to have multiple xboxes in a single household just to play some games together.

You just need enough CPU to support it and some hardware customization, but that would be a cool and neat platform specific setup.

It will still lend to why MS is continuing to push forward on the idea of GPU side draw call execution.

 

[manux]

Sometimes time travelling can give perspective. Old but yet kind of relevant and interesting discussion with john carmack. Even the best of us have trouble predicting future of console/technology landscape.

 

[BRiT]

At least it's not a Tim Sweeney series...

 

[liolio]

Wondering what type of platform features are needed for next gen. We had a lot of discussion about what the hardware would do for the games, but not the platform.

My xbox is currently being occupied by my kids watching netflix, but with HT and enough CPU, it could be playing a game on another screen or streaming it elsewhere.

I get the idea but it is better from a manufacturer POV to sell you another system, especially if through a tiered market approach manufacturers pass on the (failed) selling at loss (or close) business model.
I don't know for the platform but clearly people want to play on "movable" system.

Just something I'm thinking about right now, trying to figure out if this is an important aspect for the platform developers. I know the common rebuttal is 'my smart tv has netflix, don't need it on xbox etc'. But smart tvs are (a) slow in general, and (b) only support the most popular media services. Take longer to update etc. We are still discounting blu ray playing while playing a game.

Smart is a bad idea, TV last far too long compared to whatever hard and software platform is integrated, then there are "the things" provided by the ISP which are equally bad for the same reason. That's something that need standardization like we saw in the PC realm say there is a standard slot in the tv you plug whatever "smart thing you want (from Apple to Android or windows powered devices).

There is a cross over point for MS, if they have tons of CPU and memory such that one person could be using the device for games and the remaining power could be used for another app by another user. That would be an almost weird mainframe style model, where all the processing could occur on on a single device and it would stream out to other devices if needed. I don't see that as being prominent, but I can't help but see a future where there are multiple users on 1 device. Whether it be M/KB or controller. We're talking supporting multiple cursors and keyboards etc, so that everyone can take advantage of this one device.

I don't feel the same power is still more the leading concern to me, you have powerful system you want it to consume its "fair amount of power compared to the task it executes. This gen of console lacks proper power management as seen on mobile devices, laptop even desktop.
The server/client approach is sensible for manufacturers want to target specific markets or locations (with high optic fiber overage) and reach phone, tablets but also TV.

There's too much computing power out there for all of it to be wasted on a single workstation. Usage graphs will show that PCs are unlikely to be really pushed to it's maximum capability unless doing heavy professional work and even then, only a handful of workstations are going to be pushing the box to load 24/7. We are good at pushing GPUs to their limits. But in a non niche task, we are not good at pushing CPUs to their limits. And we continue to see more innovation in moving over more CPU work to the GPU such that it continually frees up more CPU to do other things.

Again I see what you describe as an opportunity to save power, it is not like performance comes without heat, noise, etc.

When I look at the stack for MS, I think we'll see Windows move in this direction. Users can build beefy machines that multiple users in a house hold can draw from, or a company etc. They could collaborate on a single screen locally or work on the device separately etc. But they could also have their own hardware to support it, and a powerful enough xbox may be able to do the trick to act as a central machine. It would be next gen for sure, and something I wouldn't expect until next gen. The question is whether they would go for that. Perhaps a pro model. But this time the differentiating factor is more GPU but instead it comes with more CPU and memory to support other users leveraging the hardware, the remaining game hardware is still 'locked' to perform the same as the base unit.

If computing power turn inot such a commodity it is more economically sensible to sell multiple devices making a profit on hardware.
I would add that through the Switch we can the extend to which modern games are scalable, the truth is a lot of laptops can do as good a job as the Switch. PC editors are pushing the bar with system requirement and minimal config also to drive hardware adoption (I don't know if they get money or help/man hours in return). The thing is the lowest common denominator which is pretty much the suckiest windows set-up is mostly up to the task (running Switch games) or really soon will. Then it is commercial will to push gaming. MSFT should work on "pocketable/practical" pad (so something pretty flat) that looks a little like their surface smart cover, something you can conveniently put into you pocket or slide into you bag and use it to play on phone, tablte and laptop.

I don't suspect next gen will simply be just this gen with better games. There needs to be more to it to get people like myself to upgrade. And it can't just be 'becomes more PC like', they need to develop a whole solution that is easy and sensible and makes sense to use as an offering.

I don't expect a jump, I would make sense for manufacturers to introduce hardware that are pretty much XBX slim+ and a PS4PRO slim+ for Sony.
The main improvement would(should) be on power consumption and CPU power. They could also target a relatively low launch price and a sexy form factor.
Through that approach you keep customers that same way Applle does for example. Customers are pretty much stuck to your environment, you offer BC even forward compatibility with old games taking advantage of new hardware.

This generation has been great but still greatly paralyzed by install times; slow application loading etc. it's certainly something that can be improved upon. How to solve these issues I don't know, but if games could load and install faster, I would already consider that a win.

I can't see that changing, storage is still slow, faster storage is too expensive, games are huge so are patches. Faster CPU will help but only that much.
I would say that editors should consider lighter version of games (simpler asset from graphics, to sounds) they could deploy on the windows store targeting laptop use for example (OT I know).

Consider the future of multiplayer in the same house hold.
1 console, but instead of 1 screen of 4K you have 4 instances of the same game running at the same time but at 1080p, streaming to 3 other users in your house and 1 on the main screen so that all four of you can play. That fixes a lot of the problem of needing to have multiple xboxes in a single household just to play some games together.

Every body wants crossplay and somehow affordable handheld that would play modern games. There are clearly way to do that, through software that can run on laptop (targeting a proper lower common denominator or a tiered approach to software) or through other "movable" affordable gaming devices.

On the matter of affordable "movable" gaming device I think it is every as doable as 199€ laptops. Though the form factor must evolve.


Long story short I think that manufacturers as well as editors should look at the Switch success as well as the public reception of their mid-gen upgrade and go with a more global approach of gaming. People wants to play the same-ish (the ish is important) games without being bound to TV or their living room.
As you said smart-tv suck => there is role for console here but they have to be affordable (at least a SKU).

 

[liolio]

It will be tough for manufacturers to differenciate next-gen, so pretty much it will be about keeping your costumers and pleasing them. The war could be on what the subscription provides, how old games run, etc. as I said in the previous post it could also be about the "limit" of the said environment, can you play on the go (streaming through the cloud or on laptop/dedicated devices).

For hardware the XBX is already pretty much where I see the future system starting.
x4 "Zen 2" cores, GPU able to push 4K(ish) resolution, 8GB of RAM 4GB of VRAM. That could two chips or one, though I favor the later, 12GB of fast memory sounds like a non economical approach to the problem. I think the difference between the system will comes to tweaks and the environment/look of the system, they should end even closer to one another than the PS4 pro / Xbox One X duo.

In the context of extending the "limit" or reach of the environment it could good to design an power efficient APU used as a CPU in the home console and an APU in the portable system (whatever its shape). AMD is clearly lagging Nvidia, Intel or ARM and Co but it should be doable to pull-out something decent throwing some silicon than necessary clocked slower.

 

[MrFox]

I can't see a split ram being more economical. From teh interwebs...

8GB GDDR5 $68 (up from $52, according to digitimes)
8GB HBM2 $160 (also hynix said 2.5x more expensive than gddr5, so it sort of fit with estimates)
8GB DDR4 2133 around $64 (dramexchange)

No price on GDDR6 yet, but samsung said the gddr6 parts will be 10nm-class which supposedly increases their working chips per wafer by 30%. The 16gbits parts shouldn't be much bigger than the current 20nm-class gddr5 8gbits considering they can fit in that small package.

I think once supply gets back to normal, 24GB gddr6 shouldn't cost any more than the ps4's memory in 2013. But the main SoC won't be anywhere near the launch ps4 $100, so they will probably have less memory and cut cost a few other places.

 

[itsmydamnation]

I can't see a split ram being more economical. From teh interwebs...

8GB GDDR5 $68 (up from $52, according to digitimes)
8GB HBM2 $160 (also hynix said 2.5x more expensive than gddr5, so it sort of fit with estimates)
8GB DDR4 2133 around $64 (dramexchange)

No price on GDDR6 yet, but samsung said the gddr6 parts will be 10nm-class which supposedly increases their working chips per wafer by 30%. The 16gbits parts shouldn't be much bigger than the current 20nm-class gddr5 8gbits considering they can fit in that small package.

I think once supply gets back to normal, 24GB gddr6 shouldn't cost any more than the ps4's memory in 2013. But the main SoC won't be anywhere near the launch ps4 $100, so they will probably have less memory and cut cost a few other places.

Remember the AMD HBCC demo, 2gb of HBM memory using HBCC paging to and from main memory, so its not about split symmetrical pools, its about split asymmetrical pools, singe 2hi stack of HBM2 is going to give 2GB at upto 300GB/s of bandwidth which can then be backed by DDR/GDDR. Also consider effective bandwidth issue of this generation, given that most of the GPU read/writes will go to the HBM this means the contention for access on the "main dram" should be much lower then currently experienced by the consoles.

Depending on cost/yields they could to 2x 1hi or 2x 2hi stacks for upto 600GB/s of bandwidth and both of those should come in at most 1/2 of your $160 HBM2 cost. Also i have big doubts on your cost of 24GB of GDDR6 going forward, manufacturing costs are increasing more then the density is improving, 20nm product is dual patterned, 10nm parts will be quad patterned etc. If some how 10nm memory prices move how you expect HBM can move to 10nm as well.

The longer this generation the more likely i think we are to see split memory, the key here is the packaging technologies ( not needing TSV's SLIM/SWIFT etc) , the sooner a PS5 is release i think the less likely we will see split memory.

 

[phoenix_chipset]

I think this would be a great ps5, obviously a touch weak in the gpu department (for next gen), but that's the easiest thing to fix with a Pro or X release down the line.

Ryzen 8 core cpu with hyperthreading, 3ghz or greater
8TF AMD gpu in the vein of Pro > Ps4 for backwards compatibility
24gb gddr6 on a 384 bit bus
1TB sata SSD

With things like temporal injection (much better than checkerboard) and dynamic res, the gpu limits won't matter as much. And 8tf as a baseline is a great leap from the base xbox one if devs aren't so focused on resolution.

A pro release 3-4 years later could be more than 12TF with 32gbs ram. Fast storage and a full fat ryzen cpu have become my most wanted components for next gen. And actually ssd prices are coming down unlike ram. Hopefully ram prices can stabilize in time for this

Hypothetically if that is the ps5, and the xbox 2 were to have 10 tf or something with no ssd, I think the ssd would be the much greater advantage for games honestly.

 

[ultragpu]

I think this would be a great ps5, obviously a touch weak in the gpu department (for next gen), but that's the easiest thing to fix with a Pro or X release down the line.

Ryzen 8 core cpu with hyperthreading, 3ghz or greater
8TF AMD gpu in the vein of Pro > Ps4 for backwards compatibility
24gb gddr6 on a 384 bit bus
1TB sata SSD

With things like temporal injection (much better than checkerboard) and dynamic res, the gpu limits won't matter as much. And 8tf as a baseline is a great leap from the base xbox one if devs aren't so focused on resolution.

A pro release 3-4 years later could be more than 12TF with 32gbs ram. Fast storage and a full fat ryzen cpu have become my most wanted components for next gen. And actually ssd prices are coming down unlike ram. Hopefully ram prices can stabilize in time for this

Hypothetically if that is the ps5, and the xbox 2 were to have 10 tf or something with no ssd, I think the ssd would be the much greater advantage for games honestly.

I don't think most people would rush out and buy a 8TF PS5 no matter how you can justify its graphics leap. Customers like numbers and first thing they'll do is compare it to the 1X and 8 vs 6 is a terrible showing of power, it's borderline a carbon copy. 8TF won't even render some of the more demanding current gen games at 4k/30fps much less next gen. I really hope Sony don't release anything less than a 12TF base PS5 regardless of its graphics leap, it's marketing suicide otherwise.