Baseless Next Generation Rumors with no Technical Merits [post E3 2019, pre GDC 2020] [XBSX, PS5]

[AlNom]

Really?

1 - Dynamic scalability isn't a holy grail that solves all. You most probably need to set the boundaries for resolution and framerate.
Imagine a game that pushes the Anaconda with insane amounts of per-pixel processing and uses a dynamic resolution that ranges between e.g. 1800p30 + upscaling for indoors and 1200p30 + upscaling for outdoors.
Put that game into a Lockhart that has a 3x smaller GPU and the 1200p30 parts (2133*1200 = 2.5MPixels) will go down to a resolution that is 3x smaller, so 2.5 MPixel / 3 = 833 KPixels, which at 16:9 is something like 1215*685.
Such a rendering resolution, on the typical >= 50" TV sets from nowadays, would probably look like ass. So what the devs must do is make the game for the smaller common denominator to avoid going below 1080p on it, meaning all the IQ juice they wanted to put there for Anaconda now needs to be scaled down for dynamic scalability to work (while the PS5 gets all the juice regardless). That or they needed to treat Lockhart as yet another entirely new device to develop for, which defeats the purpose.

Setting boundaries is now extremely difficult? Clarity is now important where AF is the first thing tossed out for consoles

2 - Geometry performance is now more dependent on compute performance (primitive culling compute shader on some engines, primitive shader on Navi). That means if geometry setup takes up a bunch of GPU core time to make sure you can use very large meshes and many of those, then the percentage amount of GPU compute needed for geometry is going to be different between the larger GPU and the smaller one. Since the performance hit from geometry will be asymetrically larger for the smaller GPU, the devs need to cut down geometry for the lowest common denominator or make two sets of geometry for each console that they to put in the disk.

You seem to be making different assumptions with respect to the make-up of the APU - or rather, in which way the GPU portion would have to be cut-down.

3 - More work for devs. Doesn't matter how close the architecture is, if it's a console you still need to do all quality for that hardware. Having the game consistently do CtDs on a PC when running on GPU XyZ with ZyX software / driver version / whatever won't be newsworthy. But on a console? That's opening a can of worms. Release year is going to be crazy for QA devs doing because multiplatform crossgen titles will release for XBone, PS4, XBoneX, PS4 Pro, PS5 and Scarlett. Four of those are fairly known quantities, but the new two are not. With Lockhart we'd be looking ay three instead.

A single example that can be a bit dubious by the small updates made to the architecture between Liverpool & Neo. If there are widespread issues for other titles, I'm not aware of them at this time.

Exactly what is "more work" defined as? Would it be more or less work to bust the resolution down to 720 or 900p for Durango titles with the higher end GPU on PS4 serving as the target console? Would it be a lot more work than Switch

----------

It may sound like a sarcastic set of questions, but they're not meant to be.

----------

What's odd to me is that we have a fair level of abstraction, and a lot of devs have been involved in multiplatform development for a long time now, so it seems a bit dubious of a complaint between Xbox & Windows development.

edit9357gflops

I'll certainly acknowledge extant factors on the manufacturing side that would be far more detrimental to mult-SKU with non-phone-costs.

4 - Virtual Reality. VR is extremely demanding on performance where devs must target absolute minimum 60FPS at the headset's maximum resolution which definitely won't be lower than WMR's current mininum of 1440*1440 + SMAA/TXAA (though preferrably SSAA) per eye. It's not like a VR game can dip down to 25 FPS and make people vomit, so performance must be very finely tuned. How would devs handle VR games for the slower console? Have it significantly reduce the immersive experience by reducing rendering resolution and/or AA and hurt the baseline? Say the VR headset isn't compatible with the new console?

er.... what

VR hasn't crossed my mind or a significant amount of the population, but it seems like a rather premium experience. How do devs cope with 4Base?

The CPU & the way engines are written can be a much bigger limiting factor for sustaining high framerates (along with bandwidth). In terms of the need for a GPU to feed native lol-resolution, well, perhaps a $499 console isn't the solution for a premium experience requiring high-end PC GPUs (whatever price is commanded on there).

That said, I don't know enough about the pro/cons for VR development budgets with respect to pushing the requirements (or art budget) so high.

 

[Tkumpathenurple]

My suggestion, which I considered absurd to begin with, was for MS to substitute Xbone S with a cutdown scarlet, not to substitute the Xbone X. So sony's equivalent would be to substitute base ps4, not the pro.

I'm not sure it's really absurd. The PS4 and XBoxOne are both likely to see 7nm iterations, and there are a variety of options open:

1. A straight forward shrink of their current SoC's. This has the benefit of 100% compatibility, but the downside of paying for porting old architectures to a newer node.
2. Use Zen and Navi. Only in the capacity of emulating the PS4/XBoxOne, and releasing them just as super slim versions of said consoles.
3. Use Zen and Navi to create cheap set top boxes - I'll call them PSTV and XBoxTV for ease of reference - which are like a waaaaay stripped down PS5/XBoxNext. Capable of locally playing PS4/XBoxOne games. Able to be targeted by devs if they feel the platform's worthwhile, a risk which is ameliorated by native compatibility with the home consoles.

I really like the idea of the third option for both Sony and Microsoft. Especially as portables.

 

[cheapchips]

VR hasn't crossed my mind or a significant amount of the population, but it seems like a rather premium experience. How do devs cope with 4Base then?

Base PS4 copes with barely passable resolution or sub PS3 level graphics. It's good enough that millions of people are having fun with it. PSVR has made Sony a healthy amount of cash. MS missed out this gen.

 

[Deleted member 13524]

Setting boundaries is now extremely difficult? Clarity is now important where AF is the first thing tossed out for consoles

Setting boundaries is difficult if you're squeezing the maximum possible performance out of a console with a given minimum framerate. In one scene the director will say "cut down on the particles", in the next he'll say "cut down on shadows". Thresholds will vary between scenes. I gave the example of resolution, but of course other settings may apply.

You seem to be making different assumptions with respect to the make-up of the APU - or rather, in which way the GPU portion would have to be cut-down.

Primitive culling is being done on the shader units on the most current (and I'd say most successful) implementations, and with Navi GPUs it will most probably be done through primitive shaders now that they fixed the performance issues it had on GFX9 / Vega.
I'm not mistaking front-end width with number of compute units. I'm saying a lower amount of compute units (which is invariably where a smaller Lockhart SoC would cut down, besides back-end) will affect geometry performance.

Exactly what is "more work" defined as? Would it be more or less work to bust the resolution down to 720 or 900p for Durango titles with the higher end GPU on PS4 serving as the target console?

https://en.wikipedia.org/wiki/Quality_control
(sorry should have written Quality Control instead of Assurance.. in my native language it's the same word)

If you're suggesting the Lockhart <-> Anaconda work would be similar to XBone <-> PS4 or PS4 <-> PS4 Pro, then that's exactly the problem that developers wanted to avoid at least during the new consoles' release window.

VR hasn't crossed my mind or a significant amount of the population, but it seems like a rather premium experience. How do devs cope with 4Base?

4Base offers a pretty mediocre VR experience despite the headset's excrutiatingly low 960*1080 resolution per eye, and it's mostly decent for games that would otherwise run on smartphones.
It's a solution for pioneers that were fine with the experiment back in 2016, but wouldn't be as well accepted 4-5 years later.

PSVR2 will probably be the benchmark for next-gen, and Microsoft ought to at least try to keep up.

 

[iroboto]

I haven’t seen a particularly good technical explanation other than “it’s stupid”

no good technical explanation.

This is more around product marketing and cost efficiency. Each SKU needs varying parts, price points, supply lines, stock, maintenance, OS support etc.
It's just a lot more costs to endure.
If we think about technical scaling with say Lockhart - why not just scale it down to X1X levels. It's not that far off the mark, and i don't expect most titles to push the zen 2 to the point that it's become the bottleneck at 30fps. So why not have X1X run at some resolution at 30fps against lockhart for instance.

 

[AlNom]

If we think about technical scaling with say Lockhart - why not just scale it down to X1X levels. It's not that far off the mark, and i don't expect most titles to push the zen 2 to the point that it's become the bottleneck at 30fps. So why not have X1X run at some resolution at 30fps against lockhart for instance.

Scorpio has its own long-term cost issues (die size, power consumption, 12GB of a memory type that is being phased out) - so a tech revision that made sense economically would probably not be that far off from what Lockhart would have been in terms of specs (IMO).

e.g. is there a point to scaling down jaguar to 7nm, can they just switch to GDDR6 with half bus, improved GPU/bandwidth efficiency etc.

 

[Globalisateur]

Yes, the probability of taking a picture on any second is very small.
However, I am comparing the probability of a picture taken at exactly zero second on a 10 minute interval vs a picture NOT taken on exactly zero seconds.

If the picture was taken exactly at 12:00:00 AM wouldn't you find that suspicious? If so, why isn't a picture taking exact at 9:20:00 AM suspicious (less of course).

If the picture was taken at exactly 11:37:51 I would find it very suspicious. The odds are the same as taking it at 9:20:00

 

[Shifty Geezer]

However, I am comparing the probability of a picture taken at exactly zero second on a 10 minute interval vs a picture NOT taken on exactly zero seconds.

1 out of every 60 is going to be taken at 00 seconds into a minute. One out of every 600 (off the top of my head) is going to be taken on an exact 10 minute interval with 00 seconds. Considering there are hundreds of millions of photos taken every day, there are hundreds of millions of photos time-stamped xx0:00.

It's not a statistical anomaly. It's also irrelevant, isn't it, on account of the image already being debunked?

 

[iroboto]

Scorpio has its own long-term cost issues (die size, power consumption, 12GB of a memory type that is being phased out) - so a tech revision that made sense economically would probably not be that far off from what Lockhart would have been in terms of specs (IMO).

e.g. is there a point to scaling down jaguar to 7nm, can they just switch to GDDR6 with half bus, improved GPU/bandwidth efficiency etc.

I'm not sure where licensing costs fit into the picture. But i see where you are going with this.
Harder to market still. To have a next generation console be more or less sitting at X1X levels with some newer parts. X1X was a straightforward 4K Xbox one Machine. To be told the a better configuration which is newer is being locked to 1080p would be confusing for some people who don't know anything.

 

[AlNom]

pffffauxK

(yeah, I know marketing is hard, and MS has enough problems there)

The general "4K" marketing has been pretty blurry though since the One S (for media but not games but also ok for upscaling), and then there's Sony's dynamic fourk while Scorpio's "true 4k" isn't really true.

 

[Proelite]

If the picture was taken at exactly 11:37:51 I would find it very suspicious. The odds are the same as taking it at 9:20:00

Yes, the odds of picture exactly at 11:37:51 is the same as exactly at 12:00:00AM, both are very small.
The odds of a timestamp not ending in 00 seconds is 59/60.
The odds of a timestamp not ending in 10:00 seconds is 599/600.
The odds of a timestamp not ending in 0:00 seconds is 3599/3600.

 

[Proelite]

1 out of every 60 is going to be taken at 00 seconds into a minute. One out of every 600 (off the top of my head) is going to be taken on an exact 10 minute interval with 00 seconds. Considering there are hundreds of millions of photos taken every day, there are hundreds of millions of photos time-stamped xx0:00.

Yes, but what are the odds that a randomly chosen picture out of those would be stamped 0:00.

It's irrelevant, but I want to settle an argument about probability right now.

Math is serious business.

 

[iroboto]

pffffauxK

(yeah, I know marketing is hard, and MS has enough problems there)

The general "4K" marketing has been pretty blurry though since the One S (for media but not games but also ok for upscaling), and then there's Sony's dynamic fourk while Scorpio's "true 4k" isn't really true.

If the new design goals is extend on what Scorpio worked towards (4K), then 4K and 60fps is a natural logical step. I don't necessarily believe something around Scorpio's footprint will be able to pull that off consistently with most titles.

 

[function]

I think PS5 is possibly closer to Anaconda in performance than MS were expecting (perhaps even ahead in some areas?). Playstation is so important to Sony that I think they're prepared to push the boat out in a way they didn't with PS4 and PS4Pro, following on from the difficult PS3.

If your halo product is only trading blows with the competition and not leading it, perhaps you don't need the distraction (with customers, retailers, developers and internally) of a weaker value product that will initially be competing with the previous gen's refresh models anyway.

 

[BRiT]

This an honest question here, does EXIF not include higher precision of time than just seconds? It seems rather imprecise if it doesn't.

 

[Shifty Geezer]

However small they are, across the entire internet, millions of people are sharing photos. It is inevitable that some of those will have 0:00 timestamps. The fact that one of those images shows up here is nothing special because it would happen somewhere.

 

[Proelite]

Out of curiosity, if the developers were just going to target the lowest spec machine and scale up, why not make two skus that take advantages of this?
The problem with Lockhart was that it was too weak, and Anaconda would be not used to its full potential.

What if we raise the baseline to an essentially a cutdown version of Anaconda, i.e. Vega 56 vs Vega 64. Devs target the baseline, and make no development effort for the higher end sku.
We already see this in action with the Xbox 1 and Xbox 1 s.

The lower end version will run at a lower framerate / dynamic resolution / with less effects. The higher end version will automatically have better framerates and resolution ala boost mode.

Lower end console will have slower GDDR6 ram, and slower GPU with disabled CUs. It'll be cooled with simple heatsink and fan. $399
Higher end console will have faster GDDR6 ram and binned GPU with more enabled CUs at higher clocks. Possibly a bigger SSD. It'll be cooled with vapor chambor. $549.

I think it'll be very helpful for the industry for one next generation console to come in at $399.

 

[AlNom]

What if I waited until 00:00:00 specifically because I have an OCD-like issue.

----------

#384-bit'th post :V Nothing to see here.

 

[Proelite]

However small they are, across the entire internet, millions of people are sharing photos. It is inevitable that some of those will have 0:00 timestamps. The fact that one of those images shows up here is nothing special because it would happen somewhere.

It's nothing special, but the probably of any image showing up in our relevant discussion with 0:00 timestamp, with the condition that it's not chosen for the 0:00 timestamp, is a very small probability.

 

[BRiT]

What if I waited until 00:00:00 specifically because I have an OCD-like issue.

That wouldnt have matched with the purpose of the meeting being for business.