Next Generation Hardware Speculation with a Technical Spin [2018]

[temesgen]

First steps in hardware tend to be dead-ends. XB360 had hardware tessellation - silicon that went to waste. PSP had hardware bezier support - used for squat. These RT cards will be used in offline raytracing and professional imaging. Applications will be optimised for performance, and nVidia will invest in making them faster and better. Pixar will release papers on best-case rendering techniques. Etc. So raytracing doesn't need gaming's involvement this early on to progress it. Putting it out there for game developers to dabble with while greatly limiting long-term performance is a very questionable choice. A console with a limited die budget has to allocate it to where it'll get the best returns. If raytracing at a useful rate requires 100 mm² of die out of 350 total budget for GPU and CPU, it's too expensive and we can live without. If RT can be achieved with only 20 mm² of that budget, it's worth doing.

Ultimately, it's about looking at the economy and not just blindly chasing a future. The future will happen, but it cannot be made to happen any early than technologically possible. Console's can't look to implement RT hardware until that can be done meaningfully in budget.

PlayStation one could draw between 90 and 360k polygons per second with PS4 we don't even really care to track it's drawing performance because the geometry isn't the issue. At this stage drawing triangles is far in diminished returns territory.

We're going to see significant improvement in RT from this early hardware to second and really third generation consumer hardware.

 

[Xbat]

PlayStation one could draw between 90 and 360k polygons per second with PS4 we don't even really care to track it's drawing performance because the geometry isn't the issue. At this stage drawing triangles is far in diminished returns territory.

We're going to see significant improvement in RT from this early hardware to second and really third generation consumer hardware.

Problem is back then the node shrinks were coming thick and fast and that's not the case anymore. We coming to a point where improvements are going to stagnate unless they make be break throughs.

 

[AlNom]

PlayStation one could draw between 90 and 360k polygons per second with PS4 we don't even really care to track it's drawing performance because the geometry isn't the issue. At this stage drawing triangles is far in diminished returns territory.

I think Shifty was getting at the inclusion of special HW that went mostly unused or practically unusable, be it due to flexibility with respect to integrating into the current-day rendering techniques or further performance caveats/trade-offs.

 

[OCASM]

The problem I’m talking about is that even with RT we would get games that do look pretty darn similar to what we would have without RT anyway. Only they’d run much worse.

Current RT game implementations are just a gimmick, a rushed afterthought. Hardly accurate benchmarks of the potential of this technology.

Some idea of the roadmap here:
http://on-demand.gputechconf.com/si...man-morgan-mcguire-real-time-ray-tracing.html

 

[phoenix_chipset]

PlayStation one could draw between 90 and 360k polygons per second with PS4 we don't even really care to track it's drawing performance because the geometry isn't the issue. At this stage drawing triangles is far in diminished returns territory.

360k with 1 light and one non textured color maybe.

Anyways, even now more geometry just looks better than using things like normal maps and parallax occlusion maps. They're just used because they're cheaper. Which is why on Xbox I think such techniques were a really bad substitute for geometry in most cases.

I love some of these old brute force style games coming out of japan like Nier and Nioh. And definitely when you look at NPCs in particular you can still spot those pointy edges on the modeling from a lack of polys.

Not to mention if we were at diminishing returns we wouldn't have an issue with LOD pop in either.

 

[OCASM]

360k with 1 light and one non textured color maybe.

Anyways, even now more geometry just looks better than using things like normal maps and parallax occlusion maps. They're just used because they're cheaper. Which is why on Xbox I think such techniques were a really bad substitute for geometry in most cases.

I love some of these old brute force style games coming out of japan like Nier and Nioh. And definitely when you look at NPCs in particular you can still spot those pointy edges on the modeling from a lack of polys.

Not to mention if we were at diminishing returns we wouldn't have an issue with LOD pop in either.

In terms of resolution I'd say we are at diminishing returns. 4K is a very expensive gimmick for the most part.

 

[phoenix_chipset]

In terms of resolution I'd say we are at diminishing returns. 4K is a very expensive gimmick for the most part.

I don't disagree with this as much, but even at 1080p with the kind of blurring AA techqniques devs tend to use it still leaves me wanting. Somewhere between 1440p and 1600p with non blurring post AA like smaa 1x would be my next gen sweet spot. (Insomniac quality TAA is fine though.) According to rich and the DF guys they seem to think anything over 1800p is just gravy, so..

You have to admit though that forza horizon 4 looks really damn good at 4k and 4x MSAA.

 

[davew]

…You have to admit though that forza horizon 4 looks really damn good at 4k and 4x MSAA.

Agreed, on this for certain.

Personally, I hope that console hardware developers put more emphasis on hardware that can deliver smooth motion, detailed gameplay and high, 4K resolutions on upcoming hardware, and achieve this at 60fps. Raytracing may indeed have it's place in the sun down the road, but for now, within the next 2 to 5 years, I think there are more pressing, issues that need to be sorted out. As others have said, at this point in time it certainly doesn't seem viable from a cost perspective for at the minimum, the next generation hardware. I doubt we are ready for $600 dollar consoles again.

 

[phoenix_chipset]

I doubt we are ready for $600 dollar consoles again.

I am lol.

We've been stuck at $400 consoles (xbox 360) for over a decade, so we're getting less and less for that $400 with inflation. (even though technically 7th gen was sold at a loss, not PS4 though)

Give me 200 watt monsters again with a full fat ssd, then down the road we can get models that maybe suck 70watt and are more built to last But yes we can agree we probably won't be seeing ray tracing in games next gen. Even if the consoles could get bespoke hardware for it from AMD, it's not worth the performance cost right now imo. And Navi probably has been done for a while so if that's what ps4 gets, it's safe to say it won't have RT abilities anyway.

Maaaybe MS if they launch a year after sony? It's a focus for them on PC at least, who knows.

 

[davew]

I am lol.

We've been stuck at $400 consoles (xbox 360) for over a decade, so we're getting less and less for that $400 with inflation. (even though technically 7th gen was sold at a loss, not PS4 though)

Give me 200 watt monsters again with a full fat ssd, then down the road we can get models that maybe suck 70watt and are more built to last But yes we can agree we probably won't be seeing ray tracing in games next gen. Even if the consoles could get bespoke hardware for it from AMD, it's not worth the performance cost right now imo. And Navi probably has been done for a while so if that's what ps4 gets, it's safe to say it won't have RT abilities anyway.

Maaaybe MS if they launch a year after sony? It's a focus for them on PC at least, who knows.

Very interesting perspective sir, but consoles priced too high, and a PC starts to become more sensible, not factoring in the console exclusive of course. Perhaps in the future PC will be the console of choice for all, and these super talented developers, Naughty Dog, Insomniac, etc, will be building their content for just one architecture... the PC.

In all seriousness, I'm all for the cutting edge hardware in consoles, if it could be affordable or if there was every a way to make a true modular console, that actually made sense, and did not fracture your userbase. Again, though what would differentiate that from the PC....

 

[Jay]

This is my thinking as well. I think it makes sense to spend the NRE to have contingency plans for the SoC depending on what your competition does. I imagine it would be quite a big deal if one launches in 2019 with a Navi based SoC while the other launches in 2020 with an Arcturus based SoC.

It sounds as if Arcturus could be their “Super-SIMD” architecture.

i find it hard to believe that they would have two totally different architectures so late into development.
With all the overhead of both having customisations being designed and developed, just in case the competition does something....
They know what base architecture will be available to the competition, price point and times they want to hit.
they may possibly have two versions/options of the same design later into development example take a hit on yields by enabling cu's that was for redundancy, simply more cu's, clocking higher, other 'minor' design differences, but two different architectures so far into development, just don't see it being viable

 

[anexanhume]

i find it hard to believe that they would have two totally different architectures so late into development.
With all the overhead of both having customisations being designed and developed, just in case the competition does something....
They know what base architecture will be available to the competition, price point and times they want to hit.
they may possibly have two versions/options of the same design later into development example take a hit on yields by enabling cu's that was for redundancy, simply more cu's, clocking higher, other 'minor' design differences, but two different architectures so far into development, just don't see it being viable

This assumes they need deep customization for both scenarios. I don’t think that’s necessarily the case. There may be very little customization for one path, with most of the effort simply being in different floorplanning, with the respective designs already in progress as part of AMD’s normal design cadence. It’s entirely possible the CPU is identical in both cases, too.

They’re also not committing to taping out and the development iteration phase. That’s tens of millions in mask, wafer and test costs they’re not committing to yet.

 

[London Geezer]

Are next gen designs really so ‘close to being locked down’?

I don’t understand why they’d commit to something using 2019 tech for something coming out some time in 2020, possibly even late 2020.

 

[Shifty Geezer]

Are next gen designs really so ‘close to being locked down’?
I don’t understand why they’d commit to something using 2019 tech for something coming out some time in 2020, possibly even late 2020.

Because they have to have the tech ready to test and manufacture without issue. If you wait for a new GPU arch due in mid 2020 for a Q4 2020 console, you just won't have any lead time to manufacture units to sell, or have HW for your devs to play with. It'll be maybe 18 months after a new hardware is released to the wild for devs to understand it. If the hardware is suitably abstracted, you could treat it like a PC, have devs work on older HW and then adapt to the new features of the new design when ready, but you risk driver issues screwing with your games - not good for a console! Plus the new arch is a complete redesign rather than an evolution of GCN, so you risk significant delays in it being ready and produced affordably.

 

[iroboto]

Plus the new arch is a complete redesign rather than an evolution of GCN, so you risk significant delays in it being ready and produced affordably.

I wonder how much time is gone into testing new architectures before deploying them.

 

[Jay]

The design has to be validated.
Have you ever wondered why something like a new ms Surface doesn't usually go with the newest chips?
And the chips are a lot closer in design than semi custom parts.

sony/ms has to do the following:
go over available base architectures
consider price and time frames, tools, compatibly (if bc)
price doesn't just include production of chips but cost of the IP, customization, working with amd. So to move forward with two different base architectures would mean paying amd for them both past the initial concept phase.
amd is not allowed to share or let it leak to other partners, forget about being sued, just loss of confidence would be bad enough. So whatever the other side hears will be pretty late in the day.

it takes years to bring a console to market.
parts may be 'semi' custom, but they are custom and not of the shelf. The blocks mostly are of the shelf though.

 

[McHuj]

Are next gen designs really so ‘close to being locked down’?

I don’t understand why they’d commit to something using 2019 tech for something coming out some time in 2020, possibly even late 2020.

I'd say from experience on the architecture side 2-3 years before commercial launch the design is locked down. Sure there are changes later in the cycle, but they can be very disruptive and can cause a design to miss the targets. After the architecture is defined, the long pole in the design cycle will be the verification (pre-silicon) and validation (post-silicon).

If there's a console launching next year, I bet its close to tapeout of the final silicon (if it hasn't already).

 

[eloic]

I would like to pose again the question of backwards compatibility.

With the latest rumours/speculation, in your opinion, what are the chances for next-gen consoles to be fully backwards compatible?

 

[anexanhume]

I would like to pose again the question of backwards compatibility.

With the latest rumours/speculation, in your opinion, what are the chances for next-gen consoles to be fully backwards compatible?

Extremely high. MS has committed to it as a platform feature and Sony has filed multiple patent applications regarding it since PS4 launched.

 

[Jay]

I would like to pose again the question of backwards compatibility.

With the latest rumours/speculation, in your opinion, what are the chances for next-gen consoles to be fully backwards compatible?

MS 100% will be bc. They've already pretty much confirmed it, confusion comes from if the X1 will be forward compatible.

Sony, decent chance but wouldn't be overly shocked if it wasn't, but i expect it to be. But people like to say sony is 100% because it's x86, but it's not that simple.