Next Generation Hardware Speculation with a Technical Spin [2018]

[milk]

How small would Cell be at 7nm? I'd like to see it return just for the sake of Custom Hardware!

Using the power of The Internet:
Cell was 240M transistors and 220 mm^2 at 90nm. 7nm us 7 process nodes later, at half the size each node, would be 1/2^7 = 1/128th the original size, give or take. So 2mm.

C'mon, I'll happily donate 2mm of die space to an integrated CBE! Audio processing, ray tracing, physics, all in an efficient and totally unusable, dev-hated package.

Of cell's 8 SPEs, only 6 were actually used by the games, so there goes more size reduction there. Also, you probably can emulate the PowerPC part with a regular CPU. I mean, Xbone emulates a 3 core variant of it on the jaguars... So yeah, it may be the case you could add (parcial) hardware emulation of Cell in there for not so much die space. Gotta put some EE sauce in there too, since we are at it.

 

[dten]

Jason Schreier from Kotaku has released an article that sheds more light on current status of PS5. In short, it ain't coming anytime soon and their sources are generally uncertain about the shape and timing of next generation of hardware.
https://kotaku.com/sources-the-playstation-5-is-still-a-ways-off-1825152206

Basically just chatter at this point. Early dev kits might or might not be out. That makes sense. It's like a bunch of forum posters are trying to will the PS5 into existence, not here but elsewhere. It's been pretty sane here all things considered.

A Fall 2019 console release would mean target specs are more or less locked down and game engines and asset pipelines are actively being developed with these targets right now. We'd most likely hear real rumors with code names and specs leak out.

 

[MrFox]

If the goal is emulation, maybe less expensive to add hardware helpers for opcode translation, instead of reimplementing an entire obsolete architecture.

 

[bitsandbytes]

Basically just chatter at this point. Early dev kits might or might not be out. That makes sense. It's like a bunch of forum posters are trying to will the PS5 into existence, not here but elsewhere. It's been pretty sane here all things considered.

A Fall 2019 console release would mean target specs are more or less locked down and game engines and asset pipelines are actively being developed with these targets right now. We'd most likely hear real rumors with code names and specs leak out.

You could argue PS5 is much closer to PS4 Pro and therefore it might be wise to look at how the development of that went? I think it was less than 9 months before release before we got solid rumours (that nobody believed to start with).

I don't think "new" engines will be needed this time round as it isn't going to be a completely different "language" change like going from PS3 to PS4.

 

[Deleted member 13524]

How small would Cell be at 7nm? I'd like to see it return just for the sake of Custom Hardware!

It would be complicated to make it available for TSMC, perhaps?

AFAIK the Cell was never built outside IBM fabs, which in the meantime were sold to Globalfoundries.. who in turn may not have enough capacity and/or performance for a PS5 SoC..

 

[mpg1]

Jason Schreier from Kotaku has released an article that sheds more light on current status of PS5. In short, it ain't coming anytime soon and their sources are generally uncertain about the shape and timing of next generation of hardware.
https://kotaku.com/sources-the-playstation-5-is-still-a-ways-off-1825152206

Those two people both told me that the next PlayStation is unlikely to release in 2019, let alone 2018, although they were careful to be clear that these plans are always shifting. “On a multi-year project, a lot can happen to shift schedules both forward and backward,” one person said. “At some point, Sony’s probably looked at every possible date. It’s all about what they think is the best sweet spot in terms of hardware.”

what are the biggest variables impacting this sweet spot?
1. 7nm availability
2. Memory tech/costs(GDDR6, HBM3?)

 

[dten]

You could argue PS5 is much closer to PS4 Pro and therefore it might be wise to look at how the development of that went? I think it was less than 9 months before release before we got solid rumours (that nobody believed to start with).

I don't think "new" engines will be needed this time round as it isn't going to be a completely different "language" change like going from PS3 to PS4.

There was never a PS4 Pro only game. Just getting a good pipeline for a 10 TF+, Zen+ target machine is not trivial. There won't be a PS5 only game coming out in less than 18 months that isn't in development now. Not one I'd want to play anyway.

 

[chris1515]

Of cell's 8 SPEs, only 6 were actually used by the games, so there goes more size reduction there. Also, you probably can emulate the PowerPC part with a regular CPU. I mean, Xbone emulates a 3 core variant of it on the jaguars... So yeah, it may be the case you could add (parcial) hardware emulation of Cell in there for not so much die space. Gotta put some EE sauce in there too, since we are at it.

I think it is the memory

 

[MrFox]

I think a lot of buzz came from very unreliable sources making us think 2018-2019 were possible.

2018 is almost impossible by now, all previous consoles had plenty of corroborated info more than a year in advance, and a november launch would already have had an official event in february or something. And they launched the Pro 18 months ago, so why would they have spent all this R&D on the Pro?

Even 2019 is questionable.

I thought they wouldn't make a Pro, let alone this early, so my track record of prediction isn't much of a reference

 

[borntosoul]

Does Sony beleive they need to release earlier then ms in the next gen? Or is it the other way around?
This is a factor that could be important in what gets released and when.

I believe the current consoles are strong enough for people to still keep for a year or two, $599 asking price is not extreme in this day and age- taking inflation into account that is, I’m not sure they can hold on to a set price point of $399 anymore. new phones are being bought at unheard of prices , GPUs the same.
If the consol shows a big improvement in performance/features and has a great launch line up people will buy it.
The early apdators will always pay if the machine is worth it to them. When the consol gets cheaper to produce over time so will everyone else.

I know there are billions of dollars at stake here and it’s not as easy as 1,2,3. But it’s fun being an armchair expert.

 

[DieH@rd]

what are the biggest variables impacting this sweet spot?
1. 7nm availability
2. Memory tech/costs(GDDR6, HBM3?)

Release date is too far away to have educated guess about hw price or manufacturing bottlenecks.

PS4 is selling great, and Sony should be in no rush to produce sucecssor [but they should be prepared to do something if MS launches an early attaack with their own gen9 machinne].

 

[Sonic]

What is the feasibility of having a dual APU solution? What if each APU had its own stack of HBM? This patent application from AMD seems like it would be a beast of a console if both APU's were decently powerful.

http://appft.uspto.gov/netacgi/nph-...&s1=20170185514&OS=20170185514&RS=20170185514

It would be probably be cost prohibitive and not make sense given the simpler single APU but hey maybe for a Pro version later down the line?

 

[borntosoul]

I don' beleive any launch of this magnitude can be reactive. Takes much planning and hundreds of things to bring a launch together. Will be planned much in advance.

 

[cheapchips]

Jason Schreier from Kotaku has released an article that sheds more light on current status of PS5. In short, it ain't coming anytime soon and their sources are generally uncertain about the shape and timing of next generation of hardware.
https://kotaku.com/sources-the-playstation-5-is-still-a-ways-off-1825152206

Excellent, I won't need to upgrade my PC until 2022. *

* Unless future HMDs enable a workable desktop or Star Citizen unexpectedly stops being garbage.

 

[McHuj]

I don' beleive any launch of this magnitude can be reactive. Takes much planning and hundreds of things to bring a launch together. Will be planned much in advance.

Exactly, even for a 2020 launch, everything is already locked down. New SOC design cycles from paper specs to shipping silicon are on the order of 3-4 years.

 

[borntosoul]

Theyre probably already running simulators to gauge performance.and set targets.

Can someone who knows chime in on this please.

 

[iroboto]

Theyre probably already running simulators to gauge performance.and set targets.

Can someone who knows chime in on this please.

We assume so. Most design firms will run simulations on their designs to get an idea of performance and churn out 1 design and see how it works with some real code. Make adjustments and such.

It’s expensive to churn out the silicon for 1 design though. So I’m guessing that’s not often.

 

[anexanhume]

What is the feasibility of having a dual APU solution? What if each APU had its own stack of HBM? This patent application from AMD seems like it would be a beast of a console if both APU's were decently powerful.

http://appft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=/netahtml/PTO/search-bool.html&r=1&f=G&l=50&co1=AND&d=PG01&s1=20170185514&OS=20170185514&RS=20170185514

It would be probably be cost prohibitive and not make sense given the simpler single APU but hey maybe for a Pro version later down the line?

This appears to be related to memory coherency in a multi-socketed system, not a multi-die package situation.

what are the biggest variables impacting this sweet spot?
1. 7nm availability
2. Memory tech/costs(GDDR6, HBM3?)

7nm should be good to go even for a 2019 Fall launch. The question would be if any 7nm+ comes into possibility. My guess would be no. Still, 7nm+ wouldn’t be a huge performance differentiation.

They’d also be on the cusp of the Zen 2 to Zen 3 transition, but again, maybe a 10-15% difference.

The biggest thing would be post Navi GPUs (and post GCN, for that matter). AMD’s roadmap has that at 2020, but it just doesn’t seem feasible. AMD recently moved some Zen engineers over to the Radeon group to help them out with speed/efficiency/power, but it may be too early to see the fruits of that.

Memory is more of a cost than an availability thing at this point. Samsung, and soon others, have 16Gb modules ready to go for GDDR6, HBM variants, etc. The problem is that RAM costs are still sky high, and TSVs needed for HBM are hella expensive. There are cheaper solutions out there such as LCHBM that tries to compromise bus width for speed (at cost of power), but that doesn’t have any track record to compete against right now. Intel’s EMIB would also be a candidate, but that’s Intel’s tech

It can start to get interesting when you talk about MCMs and chiplets, but I think those are Titan-level type solutions at this point.

 

[DieH@rd]

No tech info [except that discs will not be phased out in next gen], but Andrew House expects that PS4 gen will be long
https://www.polygon.com/2018/4/11/17225834/playstation-5-next-generation-andrew-house

 

[Grall]

Maybe a little over 5% of an SOC.

Hm. I wonder how fast you could clock the original Cell design on 7nm lithography... It ran at 3.2GHz on 90 friggin nanometers.

Also, quite the hotspot on the chip as well...

I think a lot of buzz came from very unreliable sources making us think 2018-2019 were possible.

Maybe that, and just general tech/console/gaming geeks geeking out over the possibility of a new toy in the near-ish timeframe.

Personally I never thought Sony would launch a next-gen until at least 2019, and likely some even later date. They're doing very well right now; there's no genuine need to launch anything new. Actually prematurely launching a next generation would just irritate their customers, and if they're too quick it would lead to tech disparity between PS and Xbox.

They have a new-ish device in PS4 Pro already, that'll tide people over for quite a while.