Predict: Next gen console tech (9th iteration and 10th iteration edition) [2014 - 2017]

[AlNom]

I know this is bordering a little on off topic, but will there be any potential die shrinks available for a next-gen console launching at the 7nm process node?

Will this likely impact the design, i.e. perhaps a motivation to future proof by going "bigger" or even going "smaller" depending on whether you think future-proofing performance is more important than future-proofing price?

It may be a longer road, but there ought to be a shrink possible.

Perhaps they'll shift towards a design with more emphasis on clock speed to mitigate the inherent cost of a larger chip although I'm not sure what the crossover point is between yields affected by wafer defects vs target clocks TDP.

Scorpio's "Hovis method" sounded like they were speed binning chips and then adjusting voltages to clock the chip appropriately i.e. overclock/underclock, so they are potentially throwing TDP variability into the mix in order to get higher # of usable chips @ target performance. The question there is how much does it cost them for a suitable cooling system and what variance is tolerable, but ultimately their internal projections must have shown that it was worthwhile...

16nmFF certainly isn't cheap at the moment, and before the reveal there was speculation as to how wide they would go in order to hit 6TF in the first place given an expectation of low clock speed (relative to desktop counterparts). With the new info, it seems there are things they have to do in order to balance the chip size & costs there.

¯\_(ツ)_/¯

 

[Rikimaru]

All chips should sustain target clock. They are tuning voltages to be as low as possible for CPU and GPU to target TDP.
1 chip maybe runs at higher CPU voltage and lower GPU voltage than average and vice versa.

 

[Prophecy2k]

It may be a longer road, but there ought to be a shrink possible.

Perhaps they'll shift towards a design with more emphasis on clock speed to mitigate the inherent cost of a larger chip although I'm not sure what the crossover point is between yields affected by wafer defects vs target clocks TDP.

Scorpio's "Hovis method" sounded like they were speed binning chips and then adjusting voltages to clock the chip appropriately i.e. overclock/underclock, so they are potentially throwing TDP variability into the mix in order to get higher # of usable chips @ target performance. The question there is how much does it cost them for a suitable cooling system and what variance is tolerable, but ultimately their internal projections must have shown that it was worthwhile...

16nmFF certainly isn't cheap at the moment, and before the reveal there was speculation as to how wide they would go in order to hit 6TF in the first place given an expectation of low clock speed (relative to desktop counterparts). With the new info, it seems there are things they have to do in order to balance the chip size & costs there.

¯\_(ツ)_/¯

All chips should sustain target clock. They are tuning voltages to be as low as possible for CPU and GPU to target TDP.
1 chip maybe runs at higher CPU voltage and lower GPU voltage than average and vice versa.

How high could they feasibly go with clock speed though, if they have to resort to manually tuning voltages for each chip and tailoring board components to each?

Unlike Scorpio, PS5/Next-box will be designed as mass market consumer devices that they'll want to be able to manufacture, assemble and ship in the tens of millions each year. Adding additional manufacturing complexity when you can simply avoid that by clocking lower and having a bigger die seems like it might present a safer option (assuming you can ensure good enough production yields).

 

[AlNom]

Sure, clocking lower is easier and cheaper. Clearly, you're not going to have a higher performing part as a result.

 

[bunge]

There's also the profit factor to bear in mind. Sony is doing so well with the PS4 at the current entry price, that there's little justification for them dropping the price down to $99 (even if they could without taking a loss on each unit). Why sell at $99 at cost when you can sell at $150 or $175 and make a healthy profit on each box. Current Sony is all about making money, given the success of their Playstation division over their other arms.

Long term mind share. There's value in getting a box into homes that might not normally buy a box. Those kids grow up and buy more boxes in the future.

No doubt the mass market success of the PS2 still drives sales today. The PS4 might be able to make some headway into the casual market if it hits a low enough price.

 

[MrFox]

If we are entering an era where shrinks don't lower cost per transistor as much as before, maybe it makes sense to spend more on cooling than silicon (higher clock, higher voltage). Seems to be the difference between MS and Sony for this mid-gen thing.

I would still prefer clever secret sauces... Like checkerboard, reprojection, and reusing previous frames buffers instead of trying to brute force stuff. But why not both?

 

[MrFox]

Obviously it depends on which price point MS ends up with.

 

[Megadrive1988]

8 core zen based APU + Vega, if we are lucky HBM 16GB, even luckier standard PCIe based SSD, with an optical drive.

Given that next generation consoles are at least 3 years away would mean they'll be built around Navi GPU architecture, not Vega.

 

[Rodéric]

Given that next generation consoles are at least 3 years away would mean they'll be built around Navi GPU architecture, not Vega.

Could be, we don't know that though.
I expect new console in a couple years, but maybe they'll only be announced then and released one year later, in which case their hardware should be ready about a year before launch, so Navi could be an option.

 

[Deleted member 11852]

Given that next generation consoles are at least 3 years away would mean they'll be built around Navi GPU architecture, not Vega.

Whilst not everything is in their control, I genuinely believe Sony released Pro at the mid-gen point and that the plan is for PS5 to be released in Holidays 2019. Time will tell.

 

[AlNom]

Could be, we don't know that though.
I expect new console in a couple years, but maybe they'll only be announced then and released one year later, in which case their hardware should be ready about a year before launch, so Navi could be an option.

#NaviGate

#DadJokes

 

[Scott_Arm]

Could be, we don't know that though.
I expect new console in a couple years, but maybe they'll only be announced then and released one year later, in which case their hardware should be ready about a year before launch, so Navi could be an option.

This is the part of the value proposition that I'm having a hard time with for PS4 Pro and Scorpio. If PS5 is a new-gen that clean slates the software, how many years would I get out of a PS4 Pro before it's obsolete. I think buying in right away (2016), you're getting 3 years minimum. I think 4 years max (2020).

Scorpio is trickier. Could it be 2 years until they launch Xbox One: Scorpio 2: Scorpio Harder (The sequel to Xbox One: Scorpio)? I don't think they're going to clean slate their software, and the new box will likely fully emulate the older ones. That seems to be how they're going, but do I want to buy scorpio if a new console comes out two years later? I could wait the two years. If it's three or four years, then I'd probably just get scorpio rather than wait it out. But four years puts them probably way too late against PS5. Really depends if they want to make sure they're head to head with PS5, or if they can launch a year later.

 

[Megadrive1988]

Phil Spencer Comitted To Console Gaming, Already Planning What Comes After Project Scorpio
The next Xbox console is already being planned.

http://gamingbolt.com/phil-spencer-...ady-planning-what-comes-after-project-scorpio

 

[bgroovy]

What about an APU with 2TB/s HBM backed by like 80GB of Optane RAM? I'm really curious what that kind of topology change in memory architecture would have on game engines outside of increasing Flops.

 

[iroboto]

If we are entering an era where shrinks don't lower cost per transistor as much as before, maybe it makes sense to spend more on cooling than silicon (higher clock, higher voltage). Seems to be the difference between MS and Sony for this mid-gen thing.

I would still prefer clever secret sauces... Like checkerboard, reprojection, and reusing previous frames buffers instead of trying to brute force stuff. But why not both?

Reminds me of this youtube video showing that Maxwell and Pascal are the same flop for flop

 

[Megadrive1988]

Could be, we don't know that though.
I expect new console in a couple years, but maybe they'll only be announced then and released one year later, in which case their hardware should be ready about a year before launch, so Navi could be an option.

Whilst not everything is in their control, I genuinely believe Sony released Pro at the mid-gen point and that the plan is for PS5 to be released in Holidays 2019. Time will tell.

 

[Michellstar]

Yes and Vega was to be released at the dawn of 2017, and this year i´m afraid it´s going out just for the top tiers, next week they release just polaris rebrands.

And HBM2 it´s still too expensive, and nexgen memory could be just a prospective wording that never materialises...

I think with the middle-gen refreshes the real next gen it´s going to take a while.

2021
x86 + 8Gb HBM2 (or the cheap variant if it´s avaliable) and 32 DDR4/5 + a GPU

 

[Prophecy2k]

Yes and Vega was to be released at the dawn of 2017, and this year i´m afraid it´s going out just for the top tiers, next week they release just polaris rebrands.

And HBM2 it´s still too expensive, and nexgen memory could be just a prospective wording that never materialises...

I think with the middle-gen refreshes the real next gen it´s going to take a while.

2021
x86 + 8Gb HBM2 (or the cheap variant if it´s avaliable) and 32 DDR4/5 + a GPU

Actually, this has been my thinking all along. When we first started hearing about PS4 Pro and Scorpio rumours, many believed them to be a foreshadowing of an industry-wide move away from discrete console generations.

Once we knew more, however, it became clear to me at least that the intention was to use them as a means to extend the generation beyond historical conventions for a typical generational length.

2020/21 sounds about right to me.

 

[msia2k75]

IMO the more i think about it, the more HBM seems to be increasingly unlikely to make its room inside a next-gen console. Nvidia reserved the HBM configuration only for P100 and AMD seems to be committed putting HBM only in the high-end GPUs. We're in 2017. Assuming a PS5 release in 2019, i doubt stacked memory will be cheap enough to be put in a mainstream console at $399 pricetag. My guess is GDDR6 will power PS5... maybe in conjunction with DDR5?

 

[Prophecy2k]

IMO the more i think about it, the more HBM seems to be increasingly unlikely to make its room inside a next-gen console. Nvidia reserved the HBM configuration only for P100 and AMD seems to be committed putting HBM only in the high-end GPUs. We're in 2017. Assuming a PS5 release in 2019, i doubt stacked memory will be cheap enough to be put in a mainstream console at $399 pricetag. My guess is GDDR6 will power PS5... maybe in conjunction with DDR5?

I believe the price issue is exactly what HBM+ is supposed to remedy since HBM2 ended up being too expensive for mainstream adoption. HBM+ is also supposed to be low power too (relative to HBM2) meaning it has a better shot at being a viable option for next-gen consoles.

It's still early days now either way, so until we get to maybe the end of 2018, we're still gonna be too far out to tell how things will land.