Next Generation Hardware Speculation with a Technical Spin [2018]

[HBRU]

with Switch selling so much with weak HW and millions PS4 still capable around ?!? I think during 2018 will be announced a ps4-next doubling hw of ps4pro (with 16 giga of GDDR6 instead of 8 of GDDR5) but same CPU type maybe 12 cores, same GPU but doubled to reach 8.4 teraflops... Dont believe in Sony not trying to extend even further ps4 commercial life

 

[bitsandbytes]

with Switch selling so much with weak HW and millions PS4 still capable around ?!? I think during 2018 will be announced a ps4-next doubling hw of ps4pro (with 16 giga of GDDR6 instead of 8 of GDDR5) but same CPU type maybe 12 cores, same GPU but doubled to reach 8.4 teraflops... Dont believe in Sony not trying to extend even further ps4 commercial life

I highly doubt this. The PS4 family is now well above 70 million units sold and so there just isn't enough consumers left to sell to to justify the cost IMO. The costs of designing 3 (PS4 super slim, PS4 super Pro, PS5) 7nm APUs would be north of a billion $!

 

[HBRU]

I see too many dreamers here... How long lasted last gen ? Why should this gen last shorter ? I see both Ms and Sony going conservative in keeping existing hardware compatibile with future sw production. They make money on sw not on hw.... Please remember.

 

[eloic]

I see too many dreamers here... How long lasted last gen ? Why should this gen last shorter ? I see both Ms and Sony going conservative in keeping existing hardware compatibile with future sw production. They make money on sw not on hw.... Please remember.

Your tone is a tad too... patronizing.

I don't think anybody is dreaming, here. I've read a couple of reasonable, good points in this thread (including some of the things you said).

 

[Theeoo]

I see too many dreamers here... How long lasted last gen ? Why should this gen last shorter ? I see both Ms and Sony going conservative in keeping existing hardware compatibile with future sw production. They make money on sw not on hw.... Please remember.

I see your point. We haven't even had a single GTA yet, and since they haven't announced it yet, it's at least a good 1.5-2 years away.

 

[McHuj]

Techreport reported the specs for the new Ryzen APU's.

https://www.techpowerup.com/240834/amd-reveals-specs-of-ryzen-2000g-raven-ridge-apus#g240834

Basically in 65W, you get 4 core/8 thread Zen at 3.6 GHz with a 1.76 Tflop GPU. I think a conservative estimate for next gen of 2X the CPU and 6X the GPU of this APU is really feasible. I do think the GPU architecture has to be Navi based since in the console space you need to maximize performance/watt. At least I thought AMD was mentioning that Navi will increase performance/watt and not just through process scaling, but architecture as well.

I'm still hoping for something close to 15 Tflops for next gen.

 

[function]

7nm looks it'll offer a little more than 2x density and a little less than 2x perf/watt than 16nm. Or something thereabouts.

I'd probably cap my potential "Xmas 2019" PS5 expectations at around double the GPU fp32 (with some efficiency gains) and around double the bandwidth. And that's if Sony want to lead with a more expensive machine than PS4 was at launch.

If Sony want to stay under under 150W again, and avoid a vast 384-bit bus, you may well be looking at something like 10 TF, 16GB (+ancillary memory) and 448 GB/s.

I think a late 2019 PS5 would be rather less than double the X1X, except in CPU where it could easily be 2.5 ~ 3 times as fast.

 

[function]

Basically in 65W, you get 4 core/8 thread Zen at 3.6 GHz with a 1.76 Tflop GPU.

... though not necessarily at the same time. It will be interesting to see if things throttle when you hit it with a heavy CPU and a heavy GPU load at the same time. Remember, consoles don't throttle, and are clocked accordingly.

 

[MrFox]

The mobile 4x ryzen with 10x vega cores is 25W max tdp (limited to 15w with thermal throttling) being clocked at 2.2Ghz cpu, 1.3ghz gpu. So ryzen cores are really not power hungry at lower clocks even on 14nm.

If we follow the fabs statements, at the same clock and same transistor count, a shrink from 14 to 7 would provide 1.5x density and 65% lower tdp.

So a Vega 64 would scale down below a console's 150W limit, wild guessing the die size a bit above 400mm2 with 8 ryzen cores at something like 2.5Ghz. (and assuming a move to gddr6). So still something like 12TF in 2019.

From that, moving to 7nm+, the improved density and clock could provide 15TF with less than 400mm2 die size, in 2020. That would be pretty good.

Wait another year for process stability/improvement/cost and... 18TF in 2021?

 

[Shifty Geezer]

with Switch selling so much with weak HW and millions PS4 still capable around ?!? I think during 2018 will be announced a ps4-next doubling hw of ps4pro (with 16 giga of GDDR6 instead of 8 of GDDR5) but same CPU type maybe 12 cores, same GPU but doubled to reach 8.4 teraflops... Dont believe in Sony not trying to extend even further ps4 commercial life

Who would buy such a machine? Generations are typically far more powerful than previous consoles, producing visibly superior results. If you don't offer your customers that, you aren't offering what they're after, so are unlikely to sell them the machine. Switch sells because it's portable and connects to a TV. Same reason Wii sold because of its unique features. A general console with no USP/gimmick is going to need to offer a suitable improvement on what people are currently playing, and a ~4x PS4 won't manage that. There's definitely not going to be a second PS4 mid-gen machine - that's just a confusing mess of hardware. PS5 will be notably more powerful and a proper new console. It may or may not have BC with PS4 games.

 

[MrFox]

Sony didn't say much about next gen, but they did say in an interview that there won't be any other ps4 upgrades before ps5.

 

[function]

The mobile 4x ryzen with 10x vega cores is 25W max tdp (limited to 15w with thermal throttling) being clocked at 2.2Ghz cpu, 1.3ghz gpu. So ryzen cores are really not power hungry at lower clocks even on 14nm.

I'm not convinced Raven Ridge will maintain anything like 1.3 gHz under load on the CPU and GPU. I don't think these are console-applicable frequencies, especially on the GPU end.

If we follow the fabs statements, at the same clock and same transistor count, a shrink from 14 to 7 would provide 1.5x density and 65% lower tdp.

So a Vega 64 would scale down below a console's 150W limit, wild guessing the die size a bit above 400mm2 with 8 ryzen cores at something like 2.5Ghz. (and assuming a move to gddr6). So still something like 12TF in 2019.

From that, moving to 7nm+, the improved density and clock could provide 15TF with less than 400mm2 die size, in 2020. That would be pretty good.

Wait another year for process stability/improvement/cost and... 18TF in 2021?

Not sure I share your optimism there, at least based on Vega 64!

Vega 64 exceeds 300W while throttling below 12TF clocks, and that's without console level optimisations and heavy use of async to inevitably cause even more power consumption / pressure to throttle. And it's also likely saving a few tens of watts from using HBM2 as opposed to GDDR6 (I don't expect even GDDR6 will be as friendly on the juice a HBM2). Plus, Vega 64 is a fully active chip without the four redundant CUs a similar console system would have (with the associated die penalty).

I think Vega 56 is probably a better guesstimation target as it i) doesn't push clocks so aggressively past what the architecture and the silicon can handle and ii) does have the requisit number of disabled CUs.

Unless AMD have pulled out some clock speed magic for 7nm console chips, I don't see how Sony or MS could hit Vega 64's supposed 12TF in 2019. I'm not even convinced AMD will have a GPU out in 2019 that can hit 12TF with low enough power consumption (~110W?) to be suitable candidate for a > 150W-at-the-wall console.

 

[MrFox]

If vega 56 is a better clock target, all AMD needs is for Navi to give a 10-15% improvement beyond what the next process provides. And a few more CU than vega 64 for yield. Hopefully Navi will provide something more than just a shrink.

Once gddr6 card come out we might be able to compare wattage against equivalent hbm models. That will be intetesting because it's unclear right now how much it saves.

 

[Recop]

I see too many dreamers here... How long lasted last gen ? Why should this gen last shorter ? I see both Ms and Sony going conservative in keeping existing hardware compatibile with future sw production. They make money on sw not on hw.... Please remember.

So they have no reason to launch another mid-gen refresh... if everything is right at the moment, then they have nothing to change.

I agree with you, the PS4 sells too well for the moment. The best strategy for Sony is to wait the most they can to get the best available technology at the lowest price.

Unless they want to launch their console one year before MS...

 

[bitsandbytes]

So they have no reason to launch another mid-gen refresh... if everything is right at the moment, then they have nothing to change.

I agree with you, the PS4 sells too well for the moment. The best strategy for Sony is to wait the most they can to get the best available technology at the lowest price.

Unless they want to launch their console one year before MS...

I don't believe sales of PS4 has much of any bearing on when PS5 releases. Barring something out of Sony's control (like TSMC not being able to make 7nm chips for example) then PS5 will come out after being cooked for 4 years* in late 2019 or 2020.

*Mark Cerny says it takes 4 years to make a console.

 

[Tkumpathenurpahl]

I think there's an argument to be made in favour of launching later - 2020 or so.

But a 2019 launch would probably give the PS5 a year on the market to itself, whilst MS can wait it out and launch at a higher spec.

The sales thread has shown us that the XBox division is profitable. So, if the X1X keeps selling well, and keeps people buying games in the MS ecosystem, there's evidence that a year long wait is fine, as long as it provides better hardware.

I don't expect the PS5's release to be effected by PS4 sales. As others on this board have said: the PS2 sold tens of millions after the release of the PS3 and X360. A PS4 Super Slim and a PS4Pro Slim would still be attractive prospects and could sell for many years into the life of the PS5, especially if the latter is backwards compatible.

The PS4 will become the cheapest entry point into the PlayStation ecosystem and routinely be sub-HD, the Pro will become the standard 1080p-1440p console. Developers will target the ~100 million PS4/Pro platform, and offer enhancements for the PS5. Sony will only charge a single licensing fee for releasing a game on the PlayStation ecosystem - maybe offering some kind of incentive for implementing PS5 enhancements.

-- Onto hardware --

I expect that the APU of the PS5 - in particular, its size - will be designed around 7nm+ which is due in 2020. In order to release in 2019, they'll utilise 7nm, manufacture a larger chip, and absorb the costs, knowing that 7nm+ will render it break even (or the PS4 style of break even: a game and year of PS+) a year later.

As an example, and pulling the figures out of my arse:
- the PS5 launches in late 2019, on 7nm, with a 450mm2 APU
- when 7nm+ is ready, they transition the APU to that node, for a 350mm2 APU
- there's no redesign of the chassis, no slim at this point, just a modest internal revision

On the topic of revision: I'm going to revise my RAM estimate to 16GB of GDDR6 or HBM3, with barely any of it dedicated to the OS. I'm sticking with the NVME prediction though.

DDR4 is fast enough for a 4K console but it's expensive. But it's also cheaper than GDDR6 or HBM3, so using a healthy chunk of it for the OS and apps will come with a tradeoff, which I think will be the "game memory."

This has me thinking though: GDDR6 can only reach its crazy bandwidth when there's 24GB of the stuff on a 384 bit bus. Would 16GB offer enough bandwidth? Would 16GB be a small enough amount that HBM3 becomes cost effective?

 

[Recop]

I don't expect the PS5's release to be effected by PS4 sales. As others on this board have said: the PS2 sold tens of millions after the release of the PS3 and X360. A PS4 Super Slim and a PS4Pro Slim would still be attractive prospects and could sell for many years into the life of the PS5, especially if the latter is backwards compatible.

It's not completely comparable. Sony made several mistakes with the PS3... they won't repeat them.

If the PS5 is as appealing as the PS4 at launch, we should see another outcome.

 

[milk]

I don't expect the PS5's release to be effected by PS4 sales. As others on this board have said: the PS2 sold tens of millions after the release of the PS3 and X360. A PS4 Super Slim and a PS4Pro Slim would still be attractive prospects and could sell for many years into the life of the PS5, especially if the latter is backwards compatible.

The PS4 will become the cheapest entry point into the PlayStation ecosystem and routinely be sub-HD, the Pro will become the standard 1080p-1440p console. Developers will target the ~100 million PS4/Pro platform, and offer enhancements for the PS5. Sony will only charge a single licensing fee for releasing a game on the PlayStation ecosystem - maybe offering some kind of incentive for implementing PS5 enhancements.

In that scenario ps5 ends up becoming just a glorified PS4Pro+. The way sony comunicated until now, gives off the impression they are interested in a bigger tech leap for ps5 rather than an incremental improvement like they did with Pro. And understandably so. They already have a cheap and a premium model. What good does it make to introduce yet another one. Would it gather enough new sales to justify the R&D cost and the logistical burden of having yet another target spec for software? I don't think so.

 

[Theeoo]

I see your point. We haven't even had a single GTA yet, and since they haven't announced it yet, it's at least a good 1.5-2 years away.

Just realised this comment reads wierdly. I meant we haven't had a next-gen GTA with assets and code specifically built for the hardware of the PS4/Xb1, whereas GTA5 was built for the PS3/360 and then enhanced/remastered.

 

[Tkumpathenurpahl]

In that scenario ps5 ends up becoming just a glorified PS4Pro+. The way sony comunicated until now, gives off the impression they are interested in a bigger tech leap for ps5 rather than an incremental improvement like they did with Pro. And understandably so. They already have a cheap and a premium model. What good does it make to introduce yet another one. Would it gather enough new sales to justify the R&D cost and the logistical burden of having yet another target spec for software? I don't think so.

I agree that there'll be a bigger leap than we saw going from PS4 to Pro, but why would a backwards compatible PS5 be nothing more than a ProPro?

To be clear, I'm not saying that developers should have to target the PS4+Pro userbase, just that it's best to foster the devs who do. There will still be PS5 exclusives and exclusive features (like the nemesis system being present in the PS4/XB1 versions, but not the PS360,) but leave it up to developers/publishers.