GPU Ray Tracing Performance Comparisons [2021-2022]

Yea ;) with nothing to fall back on the choice is basically made for you. Hopefully in time a developer will discover to incorporate all of this + reflections on console.
Quarter resolution reflefctions, 1080p denoised then upscaled.
Can't see this looking good without a ML upscaling, TI based one from 1080p, I'm not hopeful at the moment. (say 1080p due to how heavy RT reflections are on RDNA2)

Will definitely be Intresting to see what devs come up with in regards to dealing with reflections once we fully drop last gen.
 
Quarter resolution reflefctions, 1080p denoised then upscaled.
Can't see this looking good without a ML upscaling, TI based one from 1080p, I'm not hopeful at the moment. (say 1080p due to how heavy RT reflections are on RDNA2)

Will definitely be Intresting to see what devs come up with in regards to dealing with reflections once we fully drop last gen.
indeed.. just thinking out loud, this will outperform my 1070 at software DXR for sure ;)
 
For BVH you could also consider things like instantiation. Is there standard way to handle instances of same objects or is it perhaps something hw can do tricks with? What about rotating, resizing etc. the instances during BVH creation? Naively one could create unique geometry out of instances(bigger BVH). Perhaps there are better ways to instantiate and also have tricks in hw to support this efficiently. Another thing is flexibility of hw. HW can be so hardcoded today that there is no point opening it up further. AMD is ahead here as they are using regular compute, on the other hand this approach seems to have a pretty huge performance hit(blender, cp2077,...)

edit. Your thought about parallelism is naive. Nvidia has separate fixed function hardware to handle bounding boxes and triangles. You need to keep all units fed or you will go slow. Parallelism is very important as is sorting for coherent rays etc. Luckily for now the black box hides a lot of low level stuff and developers can focus on the big things like how to cast rays that contribute most and how to reuse results between frames.

:) My parallelism comment was sarcasm meaning devs and IHVs should be able to innovate both at the same time - which many people do not even want to have, assuming leaving it all to IHVs / MS and waiting is good enough.

I'm just too different from state of the art engine tech, so this does not apply to me, sadly. Instancing is no option also because of that.
Actually the only way to support RT for me is to have big chunks of geometry which i'll change if LOD changes, and build BVH from scratch each time this happens. Not sure if that's practical because detail is planned to be very high.
My situation is similar to that of UE5, but even worse because they rely on instancing and so can reuse BVH, while i can not.

However, chances are we'll have DXR 5.0 until i'm done, so waiting is probably ok ;)
 
Well, the 1070 is a higher mid tier GPU launched in 2016, that gpu is closer to the 2013 consoles then the 2020/2021 ones :p
sure but it's still packing quite a bit more power than series S however.
but it really goes to show despite how weak series S may be, the RT units do indeed help sufficiently to perform at this function - in particular not even requiring the dedicated units of the RT Cores.

I think if Metro Exodus does indeed hit their marketed targets for Series S, that will re-write people's thinking around Series S entirely. At least it does for me.
 
sure but it's still packing quite a bit more power than series S however.

Yea i ment closer in actual release dates, the 1070 is closer to the oneS/PS4 then PS5/XSX :p

but it really goes to show despite how weak series S may be, the RT units do indeed help sufficiently to perform at this function - in particular not even requiring the dedicated units of the RT Cores.

I think if Metro Exodus does indeed hit their marketed targets for Series S, that will re-write people's thinking around Series S entirely. At least it does for me.

True, its still a zen2 based system, with a rdna2 gpu thats probably going to be more efficient then a 1070 (and thus be more performant in modern games, in special with RT), then we have the fast nvme ssd, 10GB's of gddr6 etc.
A very capable little machine, in special if you dont care all that much about 4k (whatever that means these days, PS5/XSX still upscale for just about all next gen games).

Edit: looked up the 1070 at techpowerup, its still a 6.5TF gpu yes.... its probably more powerfull upfront yes, but indeed the series has a leg up when RT'ing ;)

Also forgot the 1070 came with 8GB of GDDR, quite impressive for the time, and in comparison to the series s, it could have an advantage in the ram department since its entirely for the 1070 itself (aswell as its BW).

If your on gamepass and have a series s, do some comparisons :p
 
Last edited:
I'm surprised how well ancient 4670k cpu is performing in metro exodus enhanced edition benchmark. GPU used is 3070fe.

There is something wrong in benchmark results. I selected raytraced reflections but result reports hybrid. Not sure if setting is not respected by benchmark or if result page has a bug.

upload_2021-5-6_13-15-43.png

edit. Original message claimed shadows, this was brainfart, I meant reflections like the attached picture shows
 
Last edited:
There is something wrong in benchmark results. I selected ray traced shadows but result reports hybrid. Not sure if setting is not respected by benchmark or if result page has a bug.
If by "shadows" you mean reflections then they are "hybrid" RT+SSR, much like in most other titles with RT reflections.
 
If by "shadows" you mean reflections then they are "hybrid" RT+SSR, much like in most other titles with RT reflections.

Benchmark has 2 settings for reflections. Raytraced and Hybrid. I select Raytraced before running the benchmark and result reports hybrid reflections to have been used. I see other people noticing same discrepancy between reflection settings and benchmark result. Either the setting or report is wrong.

edit. Oh, I had a brainfart. I wrote in original post shadows when I meant to write reflections
 
Last edited:
Quarter resolution reflefctions, 1080p denoised then upscaled.
Can't see this looking good without a ML upscaling, TI based one from 1080p, I'm not hopeful at the moment. (say 1080p due to how heavy RT reflections are on RDNA2)

Will definitely be Intresting to see what devs come up with in regards to dealing with reflections once we fully drop last gen.

ML is mostly useless for this, reflections are too incoherent for the most part. There's already good temporal reprojection techniques, see Miles Morales, but it's unclear they're using it here. Quite frankly Village seems like it's launching two months or so early. Could really have used a bit more time for things like better reflections and a more stable framerate.
 
Nope, as usual in most AMD sponsored RT titles, the actual RT is balanced so it doesn't embarrass AMD. Not because of image quality reasons.

Reflections are missing half the objects in the game, GI is also limited, and is comparable to the limited UE GI used in several NVIDIA UE4 sponsored titles, however at least in those titles you are getting good RT reflections, and not the low effort found here.


It's not reasonable on AMD hardware still, fps are cut by 1/3 on their hardware despite the effects being cheap.
Have any links to benchmarks of those UE4 games with GI?
 
ML is mostly useless for this, reflections are too incoherent for the most part. There's already good temporal reprojection techniques, see Miles Morales, but it's unclear they're using it here. Quite frankly Village seems like it's launching two months or so early. Could really have used a bit more time for things like better reflections and a more stable framerate.
Possibly.
I was talking about the performance and quality shown by ML.
Was Miles Morales doing it from 1080p? Think its about 1440p.
I think once you add RT reflections onto other RT effects at the moment it will be that demanding.
Not long until we get metro to see how it holds up with its choices.

Hopefully your right though.
 
BVH you want to access is likely packed and compressed in a very hw dependent way to optimize cache line and memory usage. You would need to be able to decode the BVH on per architecture/chip basis. There is no common BVH format that you could use. This is the whole point of black box. It allows each vendor to innovate as API doesn't limit trickery hw is allowed to do.

You can bring that argument to some degree about mesh formats as well. How nice it would be for an IHV to pack and optimize meshes, sort them, quantize them, laying them out such that their prefetcher work optimally ...
I think we agree that this is not a constructive suggestion from a developers perspective. An IHV naturally has to suffer abstraction cost, because it benefits the ISVs. They are the one making the software, and they need to deal with responsibilities. Nothing is worst than an IHV that just shrugs, or doesn't even do that, because the ISV is in the end still the responsible one. Data is, and always was, owned by the software developers, and it's just natural that a black-box BVH produces friction and protest from programmers.
 
ML is mostly useless for this, reflections are too incoherent for the most part. There's already good temporal reprojection techniques, see Miles Morales, but it's unclear they're using it here. Quite frankly Village seems like it's launching two months or so early. Could really have used a bit more time for things like better reflections and a more stable framerate.

Virtually all reconstruction methods fail in those cases regardless if they're ML based or not. There's no reliable way to reconstruct our original signal (reflections) in image/(screen) space if our information used to generate the signal lies outside of this boundary. Even state of the techniques like spatiotemporal variance-guided filtering won't give you clear specular reflections ...
 
I'd imagine that better reflections and even GI possibly could've been added to PC version with ease - but it would incur a performance hit which would be too heavy without DLSS.

So basically this RT implementation is what's possible right now when consoles and RDNA2 are used as the top end target h/w.
 
I'd imagine that better reflections and even GI possibly could've been added to PC version with ease - but it would incur a performance hit which would be too heavy without DLSS.

So basically this RT implementation is what's possible right now when consoles and RDNA2 are used as the top end target h/w.
tough its under 60fps on rtx3070 in 4k so its not that geforces are bored with this game
performance-3840-2160-rt.png
 
Yes, there is a tendency developing with the last few RT titles, but we will only see the full picture a few years from now. It does seem that the 200% and 300% differences between AMD/NVIDIA we have seem in some early games do not show the full picture either.

Releasing a product with RT that is 300% faster on the competition like it seemed to be, didn't make sense imo.
 
Dont know which games you mean. Games with very light raytracing just dont hammer AMDs GPU. In Metro the number of rays scales with the resolution. So using "Ultra" in 4K is worse than in 1080p. With nVidia tracing rays nearly 4x times faster the difference will only get bigger with more rays.
 
Obviously, the lighter the use of ray tracing the closer the gap between RDNA2 and Ampere will be. And as evident, RE is very soft/light on ray tracing. The next game might paint a different picture (Metro RT, a heavy use RT game).
 
Back
Top