Could next gen consoles focus mainly on CPU?

Theeoo

Regular
Reading this got me thinking.

http://www.eurogamer.net/articles/digitalfoundry-2018-core-i7-g-with-radeon-rx-vega-spec-analysis

If Intel can pack a GPU with the power of PS4 pro inside a tiny case, but with a much better CPU inside, could console manufacturers take the same approach? Rather than waiting for 7nm which might not be mature enough for the big-arse APU's that will need to go in consoles beofre 2020, they could use existing nodes to release a console sooner, with roughly the same power as Xbox One X but with four Ryzen cores which would be a massive step up from Jaguar.

In fact Digital Foundry covered this question already.

http://www.eurogamer.net/articles/digitalfoundry-2017-could-sony-release-ps5-in-2018-in-theory
 
Last edited by a moderator:
I would say no. CPU improvement is absolutely needed, but visuals sell because they easy to market to you're average consumer. The next set of consoles shouldn't be released until a noticeable graphics improvement can be seen.
 
Well that GPU is still under what a PS4 Pro can do, never mind the Xbox One X. It would be silly to release a console with a huge jump in CPU power and decrease / slight increase in GPU power. Only if there was some groundbreaking use for CPUs in gaming.
 
Well that GPU is still under what a PS4 Pro can do, never mind the Xbox One X. It would be silly to release a console with a huge jump in CPU power and decrease / slight increase in GPU power. Only if there was some groundbreaking use for CPUs in gaming.

The cutbacks that some titles make to animation update rate is irritating enough on current gen. :p
 
mmhm.......

Plus whomever is buying the stock of the Intel/Radeon thing is probably shoving it into a pre-built system with much higher margins to make it worthwhile. Volume is probably also a lot lower than the demands of consoles (say, average 10 million a year per console company). I'm not sure that HBM is anywhere close to that sort of production.
 
The cutbacks that some titles make to animation update rate is irritating enough on current gen. :p
There will always be need to cutback somewhere. Stronger CPUs will be nice but it's not like gaming couldn't do all those things before with CPUs 1/8 the speed last gen.
 
I would say no.
Well, it seems unlikely that we would see another console gen with jaguar CPU cores; those things are at their road's end where performance is concerned and they're not going to see more development (AMD dissolved the devteam responsible for them did they not?)

Another console based on jaguars would not see a significant leap in capability. After all, these things struggle enough as it is.
 
There will always be need to cutback somewhere. Stronger CPUs will be nice but it's not like gaming couldn't do all those things before with CPUs 1/8 the speed last gen.
Indeed.

Just saying that CPUs shouldn't be ignored if they're trying to do current level of gameplay experience at proper update rates; personally, I don't find the current gameplay particularly mindblowing as it is, but it's clear they need a bigger jump next time if they are struggling in places as it is.

Anyways, as others have mentioned, there's no point in another current gen update just to replace the CPUs. The mid-gen twins practically just came out, and it'd still take time to validate a new ASIC & have adequate QA on the software side.

Devs that just launched games are probably looking to do launch titles/cross-gen, and that'll take time as well.
 
CPU will probably see a bigger jump from ps4 to ps5 than it did from ps3 to ps4. The feedback from many devs was that the cpu was disapointjngly slow for a next gen. But that shouldnt come at the cost of gpu performance. GPU is always the most important piece of a gaming machine. Give it a weak gpu and strong cpu, and devs will be offloading graphics tasks to the cpu to compensate, just like it happened on ps3/360...
 
CPU will probably see a bigger jump from ps4 to ps5 than it did from ps3 to ps4. The feedback from many devs was that the cpu was disapointjngly slow for a next gen. But that shouldnt come at the cost of gpu performance. GPU is always the most important piece of a gaming machine. Give it a weak gpu and strong cpu, and devs will be offloading graphics tasks to the cpu to compensate, just like it happened on ps3/360...
true, but that's only because cell was an insanely powerful piece of hardware relatively compared to it's GPU.

There's no x86 processor that can match the TF performance of modern high powered GPUs, and thus that reality is unlikely to ever occur again unless a technology switch occurs.
 
I don't think the Intel CPU + Radeon on the same package is really applicable to consoles, you have a clear disadvantages like not sharing ram and I doubt they wouldn't get a better deal from just AMD, which offers very close performance/power usage to Intel with Ryzen... now if it was using a Nvidia GPU, it could be interesting, because there is a clear performance/efficiency gain on the GPU side compared to AMD;

if they stick with AMD they should have very fast Zen2 cores available for the next consoles, so it would be an obvious route to have far better CPUs, but, GPU is always the priority, still, I think I would prefer a console with the PS4Pro GPU and faster CPU (like 4 zen cores with SMT at 3GHz) over the Xbox One X easily, actually to a point even a standard PS4 GPU with a fast CPU would deliver a better experience (running some current games at 1080P60 compared to 4K30 with enhanced visuals) I think, so I can see the argument.
 
What exactly is there in a game's code that is hopelessly single-threaded and needs such a gigantic departure from the 2.1/2.3GHz Jaguars we have in the mid-cycle refreshes right now?

It's not physics as those have been running pretty well as GPGPU code for several years.
It's definitely not A.I. because nowadays parallel FP16/FP32 is good for learning and parallel INT8 is great for inference, both of which gets insanely higher throughputs in modern GPUs.
At some point there was a lot of pressure to achieve very high framerates (90-120Hz for 2 viewpoints) because of VR. But even that has significantly lowered because of async space warp and other techniques. So devs can now get away with 60 FPS for VR.


To the best of my knowledge, only smartphones are ballooning in single-threaded performance because all websites and most web-apps run on a hoplessly single-threaded Javascript.
Game devs aren't really using Javascript, are they?


So what do game devs really need e.g. a 4.5 GHz Zen2 for?




Because the way I see it, we're likely to get something like a Zen+ / Zen2 derivative with 4 cores at 3GHz. Something that won't consume more than say 10W or less, at 7nm.
Or if they really need 8 cores for the sake of backwards-compatibility, it won't be clocked any higher than the 2.3GHz of the current Jaguars in the Xbone X.
 
If Sony wants VR to succeed, they don't have much choice. They need more games like Resident Evil 7, Doom, Gran Turismo Sport or DiRT Rally - big current-gen games which are running at 60fps, so adding VR support is easy. They can't do this with 30fps games, most devs are avoiding VR because of this, and when they don't, then their VR version aren't on consoles (see Fallout 4 VR). And to get more current-gen games to 60fps, they need a powerful CPU.

But that doesn't mean that they can get away with an underpowered GPU. They need at least 10+ TF. With a GPU like that they can render games like The Last of Us Part II at 4K CB / 60fps, and even have resources left for stuff like next-gen lighting.
 
Many types of gameplay are no longer CPU limited, as the rules of most games are extremely simple and for a good reason - humans need to understand and enjoy them. There are exceptions obviously; anything with large scale simulation, anything sufficiently sandboxy, anything hoping to have complex and unpredictable "emergent gameplay" (kind of glad that pretentious sounding phrase isn't in use much these days).

Scripting however is an important reality of game design, and gameplay / AI scripting doesn't necessarily lend itself to easily running well on GPUs. The biggest limitations of scripting are well thought out rules and refined, interesting behaviours, for which designer time is the biggest limiting factor. For some games there will still be a definite benefit to faster CPUs - that a anyone with a good idea can throw their scripts at even with little programming expertise. The fewer restrictions you have on what you can pull off and how you can get there the greater the opportunities are.

And it would be disappointing if at a bare minimum, "next gen" consoles couldn't run what struggles to run today at 30 fps (e.g. PS4/X1 Witcher 3 in towns) at a solid 60 fps. That would take nearly double where the X1X is.

8 Zen cores at around 3 gHz should be something like 3~4 x faster than what we have in PS4/X1, and would probably not be too bad in terms of layout unlike, a single quad which wouldn't nicely take up the narrow end of an APU and leave a large square-ish rectangle for the symmetrical CU arrangements.

I don't think a 2 x 4 Zen 2 arrangement @ ~ 3 ghz on 7nm would be outrageous in terms of either die or power. Even 2 x 6 might be somewhat feasible depending on chip dimensions, as you could lose one core for yields, reserve one for the dash, and have a nice symmetrical 2 x 5 arrangement for games.
 
Back
Top