AMD Radeon RDNA2 Navi (RX 6500, 6600, 6700, 6800, 6900 XT)

Gamers should wait for the proper patches before starting to play Cyberpunk, IMO.
There, fixed it for you. So much stuff still sub-par, ugly looking or simply wrong (some things got fixed in 1.04, I admit). I think I'll pass and wait until at least the next patch or maybe the one after that, because I really want to enjoy the game, not force myself through it. And maybe, just maybe, recent graphics cards will become more readily available so I can upgrade my PC a bit.

Any update on your progress with the polygon throughoutput rate ?
Nothing new, seems like I'm not the only one who's busy.

Which doesnt say much. You do need DLSS if you want to run RT thats going to matter on a 2060. I shared a yt video somewhere here about the performances.
Even DLSS Quality introduces ugly looking articfacts to hi-frequency textures, some lower levels (Perf & Ultra Perf) are temporally unstable in places in addition to that. Observable in the intro-section of Corpo-Rat origin, when you're in the level where your boss sends you on your first mission. Not pretty.
 
Even DLSS Quality introduces ugly looking articfacts to hi-frequency textures, some lower levels (Perf & Ultra Perf) are temporally unstable in places in addition to that. Observable in the intro-section of Corpo-Rat origin, when you're in the level where your boss sends you on your first mission. Not pretty.

Thats why its good to have a choice. PC games designed for the platform do require the latest if you want no compromises at all. I think that, going after Alex's video you can get a quite good immersion.
Obviously, 2060 level RT performance isnt going to be all that amazing, and dlss is the only way to even have RT with that gpu to avoid low fps encounters frequently.
 
Surely AMD will release a demo any day now showcasing the performance and IQ benefits of their flexible implementation.
Doubt it. They are not NV who comes up with research results on any idea that might make sense. I guess they have to focus on good DXR support, competing DLSS, etc.
It's our task to do this. But AMD has to expose to make it possible.

My biggest hope is on Epic to push AMD RT. They should like this as much as i do.
 
...
Nah, no conspiracy going on. RDNA2 and Ampere (6800/3070 for example) stack up nicely in normal rasterization, with Ampere performing somewhat better. Since RNDA2 RT is subpar its where it falls apart, aside from reconstruction tech and higher resolutions to an extend.
...

You sure about that? That's not what I've seen from reviews.
 
Doubt it. They are not NV who comes up with research results on any idea that might make sense. I guess they have to focus on good DXR support, competing DLSS, etc.
It's our task to do this. But AMD has to expose to make it possible.

My biggest hope is on Epic to push AMD RT. They should like this as much as i do.

That’s why their stuff hardly ever gets used. If AMD can’t find the time, money or talent to show their products in the best light then why should anyone else.

You sure about that? That's not what I've seen from reviews.

Yeah the 6800 is well ahead of the 3070 in most numbers I’ve seen.
 
It's going to be interesting to see how they adapt their RT solutions. I'm only a layman, but seeing what needs to be reworked, cut, or simply ported over for optimal compatibility will be fun to study.
There, fixed it for you. So much stuff still sub-par, ugly looking or simply wrong (some things got fixed in 1.04, I admit). I think I'll pass and wait until at least the next patch or maybe the one after that, because I really want to enjoy the game, not force myself through it.

I agree with both and I think CDPR will put proper effort into updating the game to perform much closer to what they idealized.
According to recent news they already recouped all the development and marketing investment through the digital pre-orders alone (!!!), so a continued support is now a means to directly increase positive revenue (and employee bonuses, to which CDPR already stated they wouldn't be dependent on review scores).

Continued support is the best way for CDPR to ensure they're getting a continued revenue stream that will directly put more money into the pockets of the dev teams. Just look at No Man's Sky, a 4-year-old one-time-purchase game that still gets into top-5 worldwide sales charts every now and then, thanks to its relentless continued support.


At the moment, from the reviews I've seen I think only the following paths are optimized in the game:
- PC version with standard DX12 rasterization (across all architectures)
- PC version with nvidia-centric raytracing
- PS4 Pro version (equivalent to PC standard DX12 but scaled down)
- XBone X version (equivalent to PC standard DX12 but scaled down)


What CDPR is focusing on, at the moment, is:
- OG PS4 version
- OG XBone version
Apparently they couldn't quite get things done properly on the 2013 consoles, but according to their latest statement it seems they're pretty confident about getting good results in the short term (up until February it seems).


This leaves to post-February for CDPR to focus on:
- PS5 (RDNA2 Raytracing)
- SeriesX (RDNA2 Raytracing)
- SeriesS (RDNA2 Raytracing)
- PC version with AMD RT (RDNA2 Raytracing)


So CDPR will need to dedicate plenty of time to study and optimize for RDNA2's raytracing, but I don't think that will happen before March 2021. And whatever optimizations they research, at least for the SeriesX/S, should be applicable to the Navi 2x cards.



And what practical differences we'll see for the games that focus on either architecture (as in what, if any, different solutions/visual design cues become predominant considerations with developers as they focus on one architecture over the other).
I think it'll be a while until RDNA2's RT capabilities are fully understood (just look at @JoeJ 's latest post), and we still don't know anything about the FidelityFX Super Sampling.
Once we get that, we should be able to pinpoint what are the strengths and weaknesses of each architecture.



Instruction set is out, so i tried to demystify RT. To me it seems:
No traversal hardware. Intersection instructions work on BVH which is stored as 1D textures.
Bounds intersection takes 4 boxes and can return their order from hit distances.
Care to try to explain what this means for the local laymen?
No traversal hardware means higher flexibility but lower performance?



That’s why their stuff hardly ever gets used. If AMD can’t find the time, money or talent to show their products in the best light then why should anyone else.
You mean on the PC market exclusively?
At this point AMD probably has a lot more RT-enabled GPUs than nvidia out there. Just PS5+SeriesX+SeriesS accounts to what, >10 million GPUs with raytracing in the hands of consumers?
I'm willing to bet more people played PS5 Miles Morales with raytracing than all the combined RTX-enabled PC games to date.
 
You sure about that? That's not what I've seen from reviews.

Yeah the 6800 is well ahead of the 3070 in most numbers I’ve seen.

I'm assuming he is saying that relative rasterization performance between Ampere and RNDA2 is fairly in line with respect overall aggregate rasterization performance based on a broader sample size and that it isn't an outlier performance wise.

Using Techspot's numbers for instance. Their mean delta at 1440p was 1.14x for the 6800 (although I'd argue their test suite likely skews towards the 6800 based on other aggregate data, eg: 3Dcenter multi review aggregate is also 1.09x for the 6800 at 1440p) while for Cyberpunk at the same resolution was 1.09. A <5% variance from the mean is fairly reasonable, as if you look at sub game benchmarks even the closest results to the mean are typically +/-3%. At the very least in terms of just the raw numbers there is nothing to suggest that Cyberpunk 2077 is a significant outlier (eg. Dirt 5 or WDL).
 
I'm willing to bet more people played PS5 Miles Morales with raytracing than all the combined RTX-enabled PC games to date.

And the RT in that game (or anything shown so far on ps5) is way below whats available to PC users. The light RT capable gpus in consoles ofcourse will reach a larger market. High end AMD gpus most likely will see better RT capabilities.

just look at @JoeJ 's latest post

I think its always a good thing to hear from more then one person/source what the take is.
 
Last edited:
This leaves to post-February for CDPR to focus on:
- PS5 (RDNA2 Raytracing)
- SeriesX (RDNA2 Raytracing)
- SeriesS (RDNA2 Raytracing)
- PC version with AMD RT (RDNA2 Raytracing)


So CDPR will need to dedicate plenty of time to study and optimize for RDNA2's raytracing, but I don't think that will happen before March 2021. And whatever optimizations they research, at least for the SeriesX/S, should be applicable to the Navi 2x cards.

Here I'm quite hopeful. Both Sony and Microsoft are well known for rather solid developer tools. And I'm sure AMD has a team or two dedicated to console support (either through Sony and MS or directly to devs. I wouldn't know.) to get things up and running quickly. So whatever refinements that CDPR could implement with the help of Nvidia I'm quite sure they are receiving from both console manufacturers. With some luck there is already reference code that can be modified and implemented relatively quickly. Layman speculation of course, but it would surprise me if Sony, MS, and by extension AMD, didn't back CDPR fully considering what a show case title it is.

Personally I'm hoping that it will also lead to some performance improvements with my 5700XT as the game should get tailored better to the WGP arrangement common to both RDNA generations :3
 
You mean on the PC market exclusively?
At this point AMD probably has a lot more RT-enabled GPUs than nvidia out there. Just PS5+SeriesX+SeriesS accounts to what, >10 million GPUs with raytracing in the hands of consumers?
I'm willing to bet more people played PS5 Miles Morales with raytracing than all the combined RTX-enabled PC games to date.

Yes, on the PC of course. On consoles AMD is providing hardware but those platforms belong to Sony and Microsoft who do the heavy lifting on the software side.
 
Doubt it. They are not NV who comes up with research results on any idea that might make sense. I guess they have to focus on good DXR support, competing DLSS, etc.
It's our task to do this. But AMD has to expose to make it possible.

My biggest hope is on Epic to push AMD RT. They should like this as much as i do.

Ampere is twice as fast with Raytracing. Ampere is twice as fast with compute performance. Doesnt sound like there is any performance left to improve performance just with "more flexibility".
 
Instruction set is out, so i tried to demystify RT. To me it seems:
No traversal hardware. Intersection instructions work on BVH which is stored as 1D textures.
Bounds intersection takes 4 boxes and can return their order from hit distances.

That's all. And so i can conclude the following:
No BVH at all, meaning we can implement whatever data structure we need. There is not even a true constraint to use 4 children per node.
Addressing the 'BVH texture' only means to address bbox coords. We can store pointers there as well, but HW only cares about bbox coords.

In short: Total flexibility, and all my dreams became true. Current game benchmarks are biased because flexibility is not used. Discussion about 'Compute vs. FF HWRT' is not over, and has just started.
... somehow annoying, haha, and too good to be true. :)

Let me know if you think i got something wrong.

Please expose those 4 instructions and give some more specs, AMD! :D
I made a thread:

https://forum.beyond3d.com/threads/rdna-2-ray-tracing.62161/

I'm very much looking forward to what console devs do.
 
Ampere is twice as fast with Raytracing. Ampere is twice as fast with compute performance. Doesnt sound like there is any performance left to improve performance just with "more flexibility".

True, but lets say a 6800XT (20TF+) class rdna2 gpu has ample power to atleast get some decent results down the road. Going forward from there, GPU power will only increase each generation of RDNA products. Usually 'fixed function' is always going to be better at the start when the tech is new. When gpus get powerfull enough things might shift, but i guess thats quite a long away.

'fixed function' because i dont think either its the right wording for even NVs solution ;)
 
That’s why their stuff hardly ever gets used. If AMD can’t find the time, money or talent to show their products in the best light then why should anyone else.
I agree in general, but here it is not really about 'AMD RT' vs. 'NV RT'. It is about asking what's better: Fast hardware solutions with problematic limitations, or tailored software solutions without them?
Also there is no AMD-HW feature which has to be introduced and supported - actually it's the lack of it.

But 'why should anyone else' is easily answered with the UE5 example. I assume UE5 with DXR is barely doable, limited and inefficient, while using AMD intersection instructions would work, acceleration structures can be reused and streamed, and fine grained LOD works as well.
Ofc. i'd like to see some demos from AMD if they get at it, but collaboration with Epic (or other developers) would give much more for less effort.
 
10 years of console devs doing amazing things because of documentation...

Lol I would hardly give AMD’s documentation credit for the incredible imagination, talent and hard work of console devs and artists.

I agree in general, but here it is not really about 'AMD RT' vs. 'NV RT'. It is about asking what's better: Fast hardware solutions with problematic limitations, or tailored software solutions without them?
Also there is no AMD-HW feature which has to be introduced and supported - actually it's the lack of it.

But 'why should anyone else' is easily answered with the UE5 example. I assume UE5 with DXR is barely doable, limited and inefficient, while using AMD intersection instructions would work, acceleration structures can be reused and streamed, and fine grained LOD works as well.
Ofc. i'd like to see some demos from AMD if they get at it, but collaboration with Epic (or other developers) would give much more for less effort.

Well that’s the thing. We need tangible evidence that flexibility produces better results today than the “limited and inefficient” DXR guiderails. Until then it’s just wishful thinking. If AMD doesn’t provide that evidence then yes we can only hope someone else does.

My view on this is simple. We already have extremely flexible RT implementations (x86, CUDA) so the experts clearly understand the trade offs of software vs hardware RT. If greater flexibility was the best option today then DXR would have been designed with that in mind.
 
Last edited:
Back
Top