Middle Generation Console Upgrade Discussion [Scorpio, 4Pro]

[temesgen]

The low hanging fruit many be to simply use Scorpio's additional power to push 4k and have a VR offering, if that is the case though that would be disappointing as I'd like to see that power used for more than simply 4k rendering. On the bright side however late adopters like myself may opt for Scorpio and buy legacy XB1 titles that presumably will run upscaled at 4k and or better frame rates and I can say for myself I'd less reluctant to invest in a MS platform with their approach to BC.

All of that said my general preference is for new hardware to come out when there is new technology at a price point that can present a tangible improvement over the previous generation. I can't see myself upgrading an existing platform I own simply for slightly better visuals, for me that isn't enough. So again not being an XB1 owner yet makes scorpio a unique opportunity that I'll strongly consider.

 

[Deleted member 13524]

A 384 bit bus will make it pretty hard to cost reduce, I suppose.

Actually, with GDDR5 384bit they only need 12 chips, whereas with DDR3 in the original Xbone they needed 16 chips.
320GB/s at 384bit would also point to GDDR5 rated at 6600MT/s, which is lower-end-ish nowadays. There are ~$200 RX470 graphics cards being sold with 8GB of that memory.

That said, if the Xbone could come down to ~$200 and needs 16 chips of some of the fastest / most expensive 2133MT/s DDR3, there's little reason to believe Scorpio couldn't eventually do the same with 12 chips of cheapest / slowest GDDR5.

 

[Jay]

Also worth mentioning that it doesn't only have to be a shrink to make a Scorpio S, they could make changes to the architecture as long as it has same or slightly higher performance profile, much like the x1s.

I think 2 tiers is a reasonable approach, as @temesgen has highlighted people will buy into either one for different reasons and time.

If the cpu is back facing and not forward facing it makes the approach of making the premium mainstream after couple years harder to pull off though, and may be better served by just having clean gen breaks.

 

[Jubei]

Its a tricky situation. Releasing high end specs will cost more but the generation will last longer while a cheaper version will cost less and probably be more popular but will need an upgrade within a few years. I mean going Zen is a massive upgrade from Jaguar because of uarch but will Zen 2 really be that much more powerful? Intel has been doing 5% improvements for years now

Thats why PS5 will take as long as Sony needs because most likely it will be a very long generation, node improvements will slow down etc

 

[Jay]

Yea, that's the catch 22 problem, and depends on where they actually see generations/less moving forward.

The main benefit for Ms with beefier cpu is that when it drops to replace x1 that gen of machines won't be held back and they will have an already established user base. generationless model.

upclocked cpu benefit is cost, but then the scorpio's life is tied to x1 and ends at same time, and does not drop to replace x1 when new gen starts.

I seriously can see it going either way and wouldn't be surprised with either choice.

 

[Shifty Geezer]

I seriously can see it going either way and wouldn't be surprised with either choice.

Indeed, which is where we probably need to look at other reasons for determining their choice. Speculating based on market positioning of Scorpio can't get us any further than this dichotomy.

 

[iroboto]

Okay that makes sense. I guess we'll have to wait and see the feature set with Scorpio before we can make that call? But since it's just 1 year after PS4 Pro, I doubt there will be enough of a differentiation to call it a new gen.

Virtually No CPU (within the same architecture) could act as a generation shift. The only features that could trigger a generation shift would be on the GPU side of things.

Looking at the requirements of PC CPU for games or the number of "look at my 2500K PC still play modern games" posts are just a solid indication of how little CPU requirements have moved over
The years in gaming, GPU on the other hands has been moving quite well.

 

[eastmen]

Virtually No CPU could act as a generation shift. The only features that could trigger a generation shift would be on the GPU side of things.

Looking at the requirements of PC CPU for games or the number of "look at my 2500K PC still play modern games" posts are just a solid indication of how little CPU requirements have moved over
The years in gaming, GPU on the other hands has been moving quite well.

Or is it that waternoose , cell and jaguar were just terrible cpu's and games were all designed around what those chips can do ?

 

[iroboto]

Or is it that waternoose , cell and jaguar were just terrible cpu's and games were all designed around what those chips can do ?

My mistake. I'll edit. Relative to x86. A change of architecture would imply a different feature set.

edit: the issue isn't the CPU. It's cheaper to continually push for more compute on the GPU side of things and process there. Games are extremely parallel problems, we don't have random accesses all over the place (the result of multiple processes fighting for resources) that would need the finesse of a CPU.

 

[eastmen]

My mistake. I'll edit. Relative to x86. A change of architecture would imply a different feature set.

edit: the issue isn't the CPU. It's cheaper to continually push for more compute on the GPU side of things and process there. Games are extremely parallel problems, we don't have random accesses all over the place (the result of multiple processes fighting for resources) that would need the finesse of a CPU.

I'm still not following. The I5 2500k is a titan compared to the jaguar in current consoles. If current consoles had a more powerful cpu then games would take advantage of it and an i5 2500k would no longer perform well. Like if ps4 and xbox one had a 2.5ghz or 3ghz 8 core jaguar even with the ipc difrences that the i5 2500 enjoyed it would still not hold up as well.

So if scorpio had a 8 core zen in it and ms made a game that used all 8 cores then an i5 2500k would not play that game well

 

[Shifty Geezer]

Not all jobs are computable. Branchy AI remains CPU intensive. World simulation (GTA city) is I'm sure CPU intensive, unless there has been some major headway recently. And even if all these jobs can be turned into parallel compute tasks, it's going to be a number of years of software evolution before such practices are commonplace - devs don't have the luxury of taking a few years out to rethink their paradigms before their next game needs to be made. I'm certain a game targeting a beefy i7 wouldn't scale down at all to a PS4-class CPU. That can't readily be said of GPU tasks unless there's lots of non-graphical workloads.

Putting it another way, do supercomputers with massive GPU stacks make do with tiny little ARM CPUs or do they pair those big GPUs up with big CPUs?

 

[iroboto]

I'm still not following. The I5 2500k is a titan compared to the jaguar in current consoles. If current consoles had a more powerful cpu then games would take advantage of it and an i5 2500k would no longer perform well. Like if ps4 and xbox one had a 2.5ghz or 3ghz 8 core jaguar even with the ipc difrences that the i5 2500 enjoyed it would still not hold up as well.

So if scorpio had a 8 core zen in it and ms made a game that used all 8 cores then an i5 2500k would not play that game well

I don't disagree with that, we're looking at economies of scale. You can use CPUs to do say, Neural Net Deep Learning, or you could use GPUs. One does the task better, and the CPU is much more expensive.

When we look at CPU load for games, we have to ask what it is first and foremost, and that load, can it be offloaded to cheaper silicon.
CPU loads for games (that max out CPU) is still mainly the render block, which is mainly just drivers and draw calls.
Consoles remove as much overhead for API calls, but when the scene continually gets more complex, the CPU has to keep checking what needs to be drawn and what not.
But at the same time... it's a parallel problem, as, anytime we have 1000s of objects to process, the GPU is a better processor for that type of thing.
And so...
http://www.dsogaming.com/news/direc...proves-performance-greatly-reduces-cpu-usage/
We use things like ExecuteIndirect...

And it's usable in a lot of places... usable with Async Compute Calls!
Like culling!

And there will be continually more uses as time goes on and we continue to offload these render based codes back onto the GPU.
Freeing up the CPU to do other things, whatever they may be.

And we are _far_ from this point in time (probably another wave or 2 of games to get here) So while Zen may solve the problem through brute force muscle, its likely more than possible that developers can offload a _ton_ of work to the GPU and not require that beefy CPU to do it all at all saving costs to the consumer, while reducing the power footprint. Then you can have this cheaper CPU doing things like AI, which is what the CPU would be good at.

 

[Shifty Geezer]

And we are _far_ from this point in time.

Which is really the issue. Can you afford to put in a weak CPU in a 'next gen' system in the hopes that everything moves to compute when the rival comes out a couple years after? What if compute solutions evolve slower than that and you're dependent on CPU grunt for the next 6 years?

 

[iroboto]

Which is really the issue. Can you afford to put in a weak CPU in a 'next gen' system in the hopes that everything moves to compute when the rival comes out a couple years after? What if compute solutions evolve slower than that and you're dependent on CPU grunt for the next 6 years?

Dunno shrug =( I guess it's similar to asking why move to compute shaders at all if we can brute force everything. You just do because of constraints.
I guess I'm making some assumptions on the lifetime of our current generation consoles. If there are say 4 more years left there, then I think we're in a position to start seeing these technologies begin to form.

I'm unsure if I'm sold on AI as being this generational leap in games, most people just want beatable AI, or scripted AI. A smart AI (like in Star Control 2 hardmode) for instance was way too hard for most people to compete with.

That's why I feel, if developers are starting to hit that CPU wall now, and they really need more CPU, they will invest the time in GPU generated draw calls to get it. But if they aren't investing in these technologies, then I'm making some assumptions that they don't need it, or it's not worth their time.

 

[mrcorbo]

Which is really the issue. Can you afford to put in a weak CPU in a 'next gen' system in the hopes that everything moves to compute when the rival comes out a couple years after? What if compute solutions evolve slower than that and you're dependent on CPU grunt for the next 6 years?

There has been a very clear trend in console hardware design that, to me, very clearly points in this exact direction. Especially if you look at Xbox -> 360 -> XBOne in terms of the CPU/GPU balance. PS3 -> PS4 -> PS4 Pro follows the same pattern.

Is it not a safe bet to expect the trend to continue for Scorpio and consoles beyond? What's a possible impetus behind a reversal of this trend?

 

[Rodéric]

I hope they'll name it properly, e.g. XBox Two.

 

[RancidLunchmeat]

I hope they'll name it properly, e.g. XBox Two.

If MS names Scorpio "Xbox 2", regardless of the hardware configuration or the ability of games to be "forward" compatible, does it end this discussion as to whether it is a next gen console?

 

[iroboto]

If MS names Scorpio "Xbox 2", regardless of the hardware configuration or the ability of games to be "forward" compatible, does it end this discussion as to whether it is a next gen console?

we're going to kill this whole debate because of a marketing term??

 

[Jay]

There has been a very clear trend in console hardware design that, to me, very clearly points in this exact direction. Especially if you look at Xbox -> 360 -> XBOne in terms of the CPU/GPU balance. PS3 -> PS4 -> PS4 Pro follows the same pattern.

Is it not a safe bet to expect the trend to continue for Scorpio and consoles beyond? What's a possible impetus behind a reversal of this trend?

Do you think the next gen will be basically upclocked jaguars or some such?
The point is that the next gen will have better cpus than this one. If anything the fact that cpu power is growing slower than gpu in consoles is even more of a reason to put a better one in Scorpio that way the next machine definitely won't be held back and rolling gen would work.
If the cpu was going to be a huge leap then putting a beefier one in there wouldn't work.

 

[fehu]

I'm unsure if I'm sold on AI as being this generational leap in games, most people just want beatable AI, or scripted AI.

You are thinking about fps where ai runs around or does squatting behind a crate, but what about crowd ai and autonomous npc interactions?