Soldato
There’s RT reflections everywhere.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Ratchet and Clank: Rift Apart RDNA 2 Ray Tracing
- Thread starter shankly1985
- Start date
There's no way this gen lasts less than 10 years. Maaaaybe we'll see PS6 et al in '28 but I reckon we get a Pro version from both first parties regardless, with a focus on RT+ML. The CPU & SSD are too good for anything less than that to make sense.
Cyberpunk 2077 still looks better overall. Refls. in eyes is standard for RT.
Associate
Regardless of quarter resolution or whatever, the additional detail that RT effects put into the footage that is being posted from ratchet and clank is to me very impressive. Especially at the start of the console life.
Of course it’s not perfect, but by heck it’s very very good all round. The amount of objects, the dynamics of load times and shifting worlds is all worth give credit to.
Of course it’s not perfect, but by heck it’s very very good all round. The amount of objects, the dynamics of load times and shifting worlds is all worth give credit to.
It's all relative to the pace of technology.
Just because the new consoles are in a very good position relative to PCs at the start doesn't mean they won't be trash in 8+ years.
Because the EUV laser is finally up and running, node progress will get back on track. And we're also entering an era of simultaneous memory innovation, packaging innovation (chiplets, 3D stacking), and specific hardware acceleration (RT cores, also new techniques like Mesh Shaders which need new hardware).
10 years, even 8 years, will turn out to be a crazy estimation for how long the new consoles can last. In 8 years, the top hardware will be along the lines of:
- PC hardware on 0.5nm (whatever is two nodes after 2nm)
- Mobile SoCs on the node after 0.5nm
- The best mobile SoCs definitely capable of having PS5/Xbox SX games ported to them with no sacrifices
- Top PC GPUs being something like 8x the performance of an RTX 3080, and much more than that in specific metrics which have new hardware acceleration associated with them. Also likely 40+GB of VRAM (and/or new techniques to significantly enhance VRAM efficiency)
- PCIe 7.0, allowing 56 GB/s NVMe SSDs, and 224 GB/s on a x16 slot
- VR market will also be massive by then, 200+ million user-base. This market size will warrant, and spawn, its own specialised hardware, since the money/demand will be there to warrant it, and therefore will leave behind hardware without this.
- The VR market exploding will also cause a massive increase in R&D spending for low-power GPUs (for standalone devices like Quest, and AR glasses too), so it's plausible we see a period of 50+% perf/W improvements for low-power SoCs
Last edited:
Don't forget that in order to play any game first you need to have a complete PC, not just the video card. So the console had good value back then too.
I'm just hoping that this generation of consoles brings some genuine innovation in gaming rather than just looking prettier - I was really hoping Cyberpunk would live up to all of CDPR's promises but - fun as it is - the entire design of that game is rooted in PS3-era gaming.
It has nothing to do with PCs, not sure why you took it that way. I'm talking about business, and as you can see from even this elongated cross-gen period studios will push to keep previous audiences as potential markets for as long as they can. I see nothing in what you said that contradicts that and in fact only many reasons as to why that will be the case: all the tech you outline is great & good advances but they're also something else which is a big problem - very costly. On the other hand games have already reached a certain level where all those future advances will mostly bring in diminishing returns in terms of visuals so the proposition of transitioning players early from PS5 to PS6 is even harder & keeping games cross-gen is even easier. For expenses see the Microsoft presentation @ Hotchips for just how quickly these costs have started going up and how much of a problem they are.
As for mobile phones playing AAA like a console - lol, never. Even the rumoured Switch Pro with its custom gaming-focused SoC will at best be equivalent to a PS4 and that's only with DLSS. Something beefy & specific to only one company, the M1, in something like the native Tomb Raider port is still only somewhere around an RX 560 in power (that's a best case scenario for them too), and still far above what they actually ship with phones. So no, you're vastly overestimating the power of mobile SoCs (never mind that they're not going to target just the top end chip). You simply cannot escape the need for more power, hence we see newer top end GPUs pushing the traditional power envelope higher & higher, and why all the MCM next-gen GPUs will also push that even further and guzzle more juice. You're not gonna take a 5-15w chip from here to 6700 XT level in 5-6 years, it's just not gonna happen by any metric's history & projection you want to look at.
So - what's the business case for Sony/MS to make a faster transition? And what's the draw for players (many of whom will only be able to purchase a PS5 2-3 years after launch even) to make that switch (and I'm talking about the multitude of users who buy 5-6 games over the console's life, who constitute the majority, not the tech focused people like us on this forum)? And don't forget - simply having better RT performance (or general graphics improvement) isn't enough, that's why Pro consoles can (and imo will) exist. For PS5 the need was due to the horrific CPU & HDD - but that's no longer an issue. So what precisely is going to push the need for a PS6 and in the timeframe you specify, as opposed to a longer cycle (with a Pro console) like I'd wager on?
In the first demo the reflections looked quite bad at 1/4 resolution. They are actually still native 1/4 resolution - but Insomniac managed to use Checkerboard rendering to upscale the reflections so they look way way better
Right, but the ultimate point is about business.
The technology enables new produces/experiences/markets.
The "edge" that consoles have is their hardware allows for awesome experiences that can't be had anywhere else for anywhere near the cost.
If they massively fall behind in technology they lose their only edge. And also the VR market will grow to the point they need to address it, and will require new hardware to do so most effectively (remember we're talking 8+ years, by your timescale)
This isn't a correct comparison, and also the Switch Pro should be significantly more powerful than the Xbone/PS4.
Render target is a very important metric, since the consoles are chasing 4K. If your render target is 1080p (which it can be easily on a small screen device) you obviously need less raw power.
Bear in mind the Switch Pro is also meant to be targeting 4K (with DLSS), so if it was less powerful than a PS4 it would barely be able to do 4K even with cartoonish graphics (and no chance in hell of having PS5/Xbox SX games ported to it, even with lots of sacrifices).
But anyway, you're underestimating the pace of technology over the next 8+ years, due to the node progress stall we had over the PS4 era and also looking at 5-6 years suddenly instead of the 8+ you were talking about before.
The M1 is only built on a node ~11.5x the density of the PS4's launch node, but the node after 0.5nm (three nodes after 2nm) should be ~25x the density of 7nm. So ~1.5 full node generations ahead of the M1, relatively speaking.
So, in other words, is it plausible to build a mobile SoC as powerful as the PS4 on a node better than 3nm and slightly worse than 2nm? Absolutely.
Another way of looking at it is perf/W. You only need to get a ~20x perf/W improvement from the PS5 to get it into the ~10W range. You can already get a ~4x using 5nm, and by the nature of low-power (voltage/clockspeed geometrically increases power draw, so being low power increases perf/W for "free", e.g. look at perf/W of laptop GPUs vs desktop), so you only need to get a 5x perf/W increase in 5 node and ~8 architectural generations, also spurred on by increased R&D money pouring in due to VR/AR over that period. The idea of not getting a 5x perf/W increase in that scenario is highly questionable.
Competition from VR/AR and the Mexican stand-off between the three major consoles (i.e. Sony, Microsoft, Nintendo).
A new console cycle will come when sales significantly drop and/or technology allows significant differences/improvements to what can be offered to consumers.
As mentioned, currently this seems like 2026 will be the year for the stars to align.
Yes RT reflections
Associate
- Joined
- 26 Dec 2020
- Posts
- 156
I don't buy this Crytek sob story. I don't remember the name of Crytek CEO but he had delusions of grandeur and an entitlement attitude towards PC gamers. Many people upgraded their hardware to play Crysis2 and they got shafted.
About the price increase, that was started by Nvidia when they knew there was no competition against them. Now even if AMD competes with them, new price points have become normalized.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 29,852
AMD aren't in a race to the bottom anymore, whilst Nvidia are leading they're pretty much just gonna price match them and make big margins on their chips.
Associate
Yeah I get the point, but end of the day game developers have always found new ways to spend the increased computational power. So general increase in the quality of games means that specialist cases like very large multiplayer games are still require some trade off by spending less technical resources in other areas. Crysis is a bit of a different beast because the load was more on the GPU and you can turn down settings, but with larger player counts stressing the CPU more that's less of an option.
I'm aware of RT not being new, but it's the innovations in hardware and software which has made it actually do-able in real time. It was true in the past that consoles had a lot more of their own innovation, before they were based heavily on PC architecture and before cross platform games. You need really aggressive research into R&D to keep making large leaps in hardware and the consoles don't have the kind of business model to keep up with that.
Bad is a relative term in this case, that's true enough. But if you take modern AAA games like Cyberpunk that have reasonably heavy use of RT then you're going to suffer on AMD cards. Not only do they just not have the same kind of power and so their frame rate suffers but to get it playable it has to be in low resolution as they lack (at the moment) any kind of decent upscaling. So looking at Cyberpunk benchmarks on youtube something like 42fps avg on a 6800XT at 1080p using Ultra preset with RT medium set. Where as 3080 same settings is at 71fps avg which can be upscaled to 4k for cheap.
It's not really that AMDs architecture or way of doing RT is bad, they have their own RT dedicated units on the card but they just spend less die area on them, and consequently can spend more on things like infinity cache and traditional rasterization, which is why they're bang for buck better at it at lower resolutions for rasterization. That's fine, that's just a trade off they've made, and the consoles have kinda just inherited those trade offs. They can technically do RT workloads with a lot of optimizations and that's what we see in the DF technical breakdown, really aggressive lowering of fidelity to get performance playable. I think a large part of the performance is due to the nature of dedicated hardware acceleration for things like RT, you can throw a certain workload at your dedicated cores for it and if you exceed what they're capable of then frame rates take a big hit, but if you stay under that budget then it's fine. And Nvidia just have a bit more headroom.
I think it's why AMD didn't really push RT in this generation of cards, all the reviews at launch focused on rasterization, and their comments at the time reflected that. I personally think it's because of a lack of any kind of decent upscaling. DLSS is clearly a sister technology to RT which if was missing would make RT dead on arrival. The high end gamers who want good quality aren't going to be convinced to back down from 1440p or 4k to 1080p or less for RT. Once FSR is perfected I suspect we'll see the next gen AMD cards push hard to compete in RT and no doubt sport a lot more die area on RT cores. I know I would have not dropped to native 1080p to get Cyberpunk playable in RT on my 3080 not after getting used to 4k in so many of my games, but DLSS 2.0 was an acceptable trade off for me.
I would say their way of doing RT is bad - short of some very advanced on the fly repurposing/context switching of hardware they are always going to be fighting the fundamental deficiencies/bottlenecks of doing RT that way vs a more dedicated approach.
I can see why AMD did it that way in the current circumstances but as much as fixed function and specialist/dedicated hardware is generally not the best direction in graphics ray tracing is one area where any other approach to a more dedicated one all else being equal will always be inferior and by quite a margin.
Scaled up isn't going to improve the story relative speaking unless they have a significant node advantage over any competing implementation.
Associate
Well I know the approaches are a bit different but my understanding is that RT cores accelerate the speed of calculations of the BVH and ray intersection significantly in both architectures. It seems likely to me that because it's such a significant performance increase that it'll be here to stay, even as general purpose compute increases. One of the things I've thought about a bit is what else we'd be doing with all this GPU with classical rasterization, I see AMD went with more rasterization speed and less RT performance but I kinda question what is going to get done with that in modern games what cool new effects are left to do with rasterization? Right now it seems like a good deal for people frame rate chasing but that's it so far.
I guess that's how I see the trade off right now, give up some die area to dedicated fixed function hardware that allows for new cool effects to be done, vs what? I dunno what the alternative is? How many developers are chomping at the bit to do a bunch of new cool stuff with rasterization if the only had the power? It seems like more polys, more res, more fps but little in the way of recent innovation. My only concern with RT is considering if it'll be a fad or not, I honestly don't think so, I think we're only going to see more of it in games and that means and dedication to fixed function hardware is a win, except for maybe people who refuse to use RT settings for some reason and deliberately turn it off.
I think the overall sacrifice of die area to RT cores is quite small? Someone did the calculation on the forums a while back if I remember correctly, and it's not all that big, so it seems like a good way of scaling. If they can throw 4x more RT cores into the next generation of Nvidia cards for example, that'd really kick things up a notch. That seems doable by only using some fraction of the new transistors gained in a node shrink, it's not like rasterization has to step backwards, just increase more slowly over time.
If AMD can improve their approach through new architecture then great I hope they do that but like everything else they'll get improvements just from the node shrinks and a change in priority on what the transistors are used for.
RT cores are around 3-4% of the die area. Fixed function hardware ends up very small, you also gain the advantage that the rest of the rasterization pipeline is not competing for the same resources.
The problem with traditional rasterization is it is struggling to scale efficiently. It is not like Ampere skimped on CUDA core and dedicated a large amount of die area to RT, quite the opposite happened, with a massive increase in CUDA cores. IF Ampere didn't have any RT or Tensor cores, and even if it was possible to add CUDA cores in proportion to die area, then performance in rasterization would probably only be about 5% faster.
The problem with traditional rasterization is it is struggling to scale efficiently. It is not like Ampere skimped on CUDA core and dedicated a large amount of die area to RT, quite the opposite happened, with a massive increase in CUDA cores. IF Ampere didn't have any RT or Tensor cores, and even if it was possible to add CUDA cores in proportion to die area, then performance in rasterization would probably only be about 5% faster.
Last edited:
Associate
- Joined
- 26 Dec 2020
- Posts
- 156
I wouldn't take CPY2077 as a neutral or even a somewhat ok example of RT for anything other than Nvidia cards. RT for that game was probably in development when RDNA2 was not even on paper (at least for us). Something like Metro EE would be an ok comparison for RDNA2 RT vs Ampere RT even though it is an Nvidia-sponsored game.
About RDNA2 being slower for RT, yeah that's true and I am all up for Nvidia cards being able to do more in RT titles but right now we don't have many examples where it is being given a fair chance (whatever might be the reasons). The one thing I have noticed is that Raster is not going away and when the scene demands both RT and raster AMD fares a little better given its lower RT performance. It's a situation like Hairwarks or Tessellation, sure they look good but having 64x Hairwroks or Tessellation to max (you can test in Unigine) does not improve IQ but just puts Nvidia in a better light.
Consoles have always been about playing it safe, extracting money from casuals while keeping costs down. Trends are set by PC gaming now more than ever with Twitch, Youtube, big streamers and E-Sports all dominated by pc gamers and popular pc games.