Next Generation Hardware Speculation with a Technical Spin [2018]

[Shifty Geezer]

I see the sense in that. We're seeing amazing games still and PS4 is still selling. There'll be significant tech advances - new architecture from AMD, raytracing/data-traversal processors, SSDs - that'll make for a machine very different to what's possible in 2019/2020.

This does conflict with the Navi reports though. Navi is supposed to be superseded by a ground-up new architecture in 2020/2021, which you'd expect a 2021 console to target. Why would Sony be investing in Navi to release later this year if its not going to drive their next console? So either Sony aren't investing in Navi and that rumour is bogus, or that relationship is for something else, or we're still looking at a nearer release.

 

[Deleted member 11852]

I think this is more reasonable since the timeline is revealed by SIE.

Interesting timeline if accurate.

This was discussed to death in back in May when the statement was made.

It's not a reference to when PS5 will launch but when profits will dip which will be after PS5 launches given Sony will continue selling PS4 and making megabucks through licensing of PS4 games. The profitability dip point will be when PS4 software sales start to naturally decline but before PS5 has market penetration to compensate for this - even allowing for increased revenue derived from subscriptions and services.

And because financials are reported in arrears, it'll be three-to-twelve months after the fact that this will be known.

 

[DieH@rd]

This does conflict with the Navi reports though. Navi is supposed to be superseded by a ground-up new architecture in 2020/2021, which you'd expect a 2021 console to target.

Maybe Sony wants a stable and non-disruptive architecture so that BC would be easier to achieve. I would not be surprised if PS5 GPU has some multiple of 18CUs so that they could more easily handle running of PS4 titles.

 

[Deleted member 11852]

Maybe Sony wants a stable and non-disruptive architecture so that BC would be easier to achieve. I would not be surprised if PS5 GPU has some multiple of 18CUs so that they could more easily handle running of PS4 titles.

They did this for Pro but I'd hope that for PS5 Sony/AMD will have developed a better technique to redistribute PS4/Pro CU workloads, whether they are traditional graphics or compute, across whatever new configuration of CU cores PS5 comes with. In many ways Pro was an easy/lazy approach - just double up on everything while refining it. Certainly easier for backwards compatibility for sure. But some things (like CPU/GPU instructions) that exist in AMD's PS4/Pro architecture have been deprecated in Navi (or post-Jaguar, pre-Navi architectures) and it's those differences Sony/AMD need to focus on if compatibly is going to be a thing.

 

[KeanuReeves]

https://www.gamespot.com/articles/ps5-still-three-years-off-says-playstation-boss/1100-6459155/

I think this is more reasonable since the timeline is revealed by SIE.

Look at the rapid price drop of SSD: cheaper 500 GB SSD is below 80 dollars now. By the time of 2021, SSD is cheap enough to become standard storage device of consoles. PS5 targets true next-gen gaming then it needs 3 years for everying: large, fast SSD, 7nm+ process and ray-tracing hardware.

Is it not adequate to just have a 128gb high-speed SSD cache? And then a standard 2tb HDD akin to what we have now and then the game data just transfers to the cache over time after the game is launched? Kind of how the games install in the background and you don't need to download the entire thing to start the games.

 

[DieH@rd]

If they feel incredibly lazy, they could retain entire double Jaguar on PS5 APU for OS and old games, which would leave all Zen cores available for gaming. Or just one quadcore module for OS.

In any case, I hope they will focus more on that side-DDR3 OS RAM pool from Pro, and significantly enlarge it on PS5. That would free more fast RAM for gaming.

 

[Tkumpathenurpahl]

In any case, I hope they will focus more on that side-DDR3 OS RAM pool from Pro, and significantly enlarge it on PS5. That would free more fast RAM for gaming.

Agreed. Especially if combined with a more powerful side processor. It could mean instant transitions between the game, dashboard, and apps. It could allow for PIP, allow them to significantly expand upon Remote Play, and capture 4K60 video.

Is there anything that the side processor could do to alleviate some load from the main hardware? Any sort of upscaling or AA, for example?

 

[Shifty Geezer]

If you're going to have a side processor also used in processing the game, why not just use 20% more normal processing and allocate that extra 20% to the system?

 

[Deleted member 13524]

If you're going to have a side processor also used in processing the game, why not just use 20% more normal processing and allocate that extra 20% to the system?

Because 20% more normal processing may cost a lot more transistors and die area than a small side processor. For example an extra Zen core Vs. a couple of Cortex A55 cores.

 

[BRiT]

Because 20% more normal processing may cost a lot more transistors and die area than a small side processor. For example an extra Zen core Vs. a couple of Cortex A55 cores.

But wouldnt the additional complexities on the memory bus, storage buses, and PCBs wipe out any benefits the second processor would get you?

 

[Deleted member 11852]

If they feel incredibly lazy, they could retain entire double Jaguar on PS5 APU for OS and old games, which would leave all Zen cores available for gaming.

Basically what they did to provide PlayStation compatibility with PS2, but that was a different time with great advances in semiconductor complexity and manufacturing occurring between the two generations.

This would add cost and complexity to PS5 so Sony would have to be super desperate if they resort to this.

 

[DieH@rd]

Since they already have secondary processor that stays up an running in standby mode, and have added side DDR3 pool for main OS, they could work more on offloading work from beefy CPU cores.

On the other hand, they have to be frugal and save every cent [when the aim is to sell 100M consoles, even 50 cents of unnecessary HW is a big deal, heck they did not even wanted to pay for UHD disc/videoplayer licensing]. The cheapest approach is to create the smallest APU that fits their needs, one memory pool, and then shove-in small ARM CPU for rest mode network connectivity. That was their approach with base PS4.

 

[Shifty Geezer]

Because 20% more normal processing may cost a lot more transistors and die area than a small side processor. For example an extra Zen core Vs. a couple of Cortex A55 cores.

How about just throwing lots of smaller, independent processors together then? Would be a lot cheaper than one/two big ones.

I think an independent subsystem should be just that. If you're going to tie it in with games, that's conceptually no different to time-sharing some existing resources. An eight-core processor with two reserved for OS, or a six-core processor and two-core OS processor, amounts to the same thing as far as the games are concerned, but having a sub processor requires a more system complexity as BRiT says. Put an OS SOC in there, enclosed on a single die maybe and walled off from the rest of the system other than memory access for streaming and output, and you have minimal impact. The moment games have access to this second CPU, you are generating complexities that are probably more than devs will ever want. If this processor isn't exposed to games but handles stuff like a theoretical post effects, these effects are then an OS feature. I wouldn't try to create OS level game effects though.

These days you could probably fit an OS processor into the southbridge or some other topology. Mobile shows what can be done with a minuscule die, and you could handle all input remapping etc. at the OS level. I'd be inclined to put a mobile chip in there for the OS. It'd probably be feature overkill, but should be very cost effective versus a custom IC. This suggestion coming with zero knowledge of how much a custom chip would actually cost!

 

[BRiT]

Maybe I'm remembering the earlier reports of PS4 idle modes, but I thought the secondary processor in PS4 really isnt used at all and the larger Jaguar CPU is used when doing updates and patches anyways. If that is the situation, then why continue on with a failed idea with little to no benefits?

 

[Shifty Geezer]

The explanation there was that the little processor couldn't cope with the de/encryption. There must be workarounds to that. If necessary, create a custom chip with a functional download block. We know mobile chips are very capable of running an OS with apps. A console company that can't make a system with a mobile SOC as functional as my years-old smartphone is plain incompetent.

 

[Deleted member 13524]

But wouldnt the additional complexities on the memory bus, storage buses, and PCBs wipe out any benefits the second processor would get you?

Maybe, but that's something only the IC designers would know.
Whatever AMD is doing, it'll use some iteration of infinity fabric, and I don't know if that makes the glue easier and/or smaller.

How about just throwing lots of smaller, independent processors together then? Would be a lot cheaper than one/two big ones.

Because you need single-threaded performance too.

These days you could probably fit an OS processor into the southbridge or some other topology. Mobile shows what can be done with a minuscule die, and you could handle all input remapping etc. at the OS level.

I get the feeling the southbridge is going away eventually, especially on systems that are becoming increasingly integrated on a single SoC like consoles.
Southbridges aren't really adding functionality as of late, so their integration on the SoC is increasingly likely.

On the PC it still makes some sense to iterate motherboard generations and support new peripherals and new CPU microcodes, but on consoles that doesn't apply.

 

[Tkumpathenurpahl]

Maybe I'm remembering the earlier reports of PS4 idle modes, but I thought the secondary processor in PS4 really isnt used at all and the larger Jaguar CPU is used when doing updates and patches anyways. If that is the situation, then why continue on with a failed idea with little to no benefits?

You're remembering it right. And although you're right that the execution failed, I think the concept's still sound: use a smaller, low power processor to handle the OS - including UI, messaging, Web browser, store, and apps - and leave as much of the expensive, gaming focused hardware as unburdened as possible. The PS4's and XB1's have squandered about 3GB of their memory, and it'd be lovely to see that go away.

Put an OS SOC in there, enclosed on a single die maybe and walled off from the rest of the system other than memory access for streaming and output, and you have minimal impact. The moment games have access to this second CPU, you are generating complexities that are probably more than devs will ever want. If this processor isn't exposed to games but handles stuff like a theoretical post effects, these effects are then an OS feature. I wouldn't try to create OS level game effects though.

That's the sort of thing I meant, yes. Why wouldn't that be worth implementing, at least as a toggle for developers, and perhaps users too? I seem to recall a touted feature of the X360 being it's free AA because of its EDRAM, which seemed to work rather well IIRC.

 

[Shifty Geezer]

The EDRAM thing was hardware in the GPU for graphics. That's very different to having a second OS processor running additional game tasks. The game engine has all the data for image reconstruction so can do a far, far better job than an outside processor.

 

[manux]

I see the sense in that. We're seeing amazing games still and PS4 is still selling. There'll be significant tech advances - new architecture from AMD, raytracing/data-traversal processors, SSDs - that'll make for a machine very different to what's possible in 2019/2020.

This does conflict with the Navi reports though. Navi is supposed to be superseded by a ground-up new architecture in 2020/2021, which you'd expect a 2021 console to target. Why would Sony be investing in Navi to release later this year if its not going to drive their next console? So either Sony aren't investing in Navi and that rumour is bogus, or that relationship is for something else, or we're still looking at a nearer release.

Perhaps schedules and plans changed. There is a lot happening making it a not so safe bet to release next gen console in 2019. Neural networks are becoming a real thing, hybrid rendering(ray+raster) might be becoming a thing. Who knows what intel is promising and is coming out around 2020 which at worst could be used as leverage to gain better deal out of amd or at best who knows?

Is there even any consumer device sold with 7nm soc that is console sized this year? Qualcom just announced 7nm highend soc but doesn't ship it yet, apple came out with theirs but that soc is tiny compared to what console needs. On apple soc it's surprisingly big area of silicon that is dedicated for neural engine, it's not a separate tiny chip anymore.

I wouldn't feel confident releasing traditional next gen console on 2019 with brand new 7nm process when competition could do serious leapfrogging 1-2 year later and start the next next gen. Typically consoles have not been the arrowhead high margin product manufactured on bleeding edge process. I could see 2019 if ps4 struggling and sony needed a reset. However ps4 sales and market share seem to be healthy not forcing sony's hand on timing.

edit. I guess if sony really does ps5 2019 it would be 2 chip solution on 12/16nm for the sake of manufacturing process and all it entails. Shrink to 7nm or better later once the process matures and becomes cheaper. However I feel the better solution is to just wait a bit and see where world is going instead of making a mad guess.

 

[Shifty Geezer]

Generally speaking, you can't wait on technology. There's always something better 12-18 months away, and sometimes if you wait, you'll end up waiting 3 years for it to actually arrive.