Poll: Do you care about ray tracing / upscaling?

[Heartbreaker]

Can you see why other consumers would go the opposite way though? If you don't use upscaling (or are fine with the small decrease in quality with FSR) and don't care much for RT-effects, having higher raster performance for years is a compelling argument. Especially when the cost difference is closer to 15-30%.

Most choose the greater feature set though.

IMO, minor differences in raster are kind of irrelevant compared to image quality differences that can be rather blatant.

But others are entitled to choose a lesser feature set, for a few more FPS...

 

[igor_kavinski]

IMO, minor differences in raster are kind of irrelevant compared to image quality differences that can be rather blatant.

Funny you say that, all the while ignoring the fact that some games degrade textures automatically when faced with lesser available VRAM.

 

[GodisanAtheist]

As someone who more or less exclusively buys used, AMD cards depreciate much much faster than NV cards (they certainly did on this last round) which makes the difference in features and capabilities more of a value proposition than it would be with MSRP pricing.

Presumably in the next 4-6 years when its time to replace my 6800XT both AMD and NV will have better parity on RT features (not unlike what happened with tessellation) while games will have simultaneously gotten better about leveraging the feature as well.

 

[Heartbreaker]

As someone who more or less exclusively buys used, AMD cards depreciate much much faster than NV cards (they certainly did on this last round) which makes the difference in features and capabilities more of a value proposition than it would be with MSRP pricing.

For new product buyers, faster depreciation, only makes AMD worse...

 

[jpiniero]

There's more rumors out there that the PS5 Pro's main focus is on RT. And AI upscaling (heh)

Whether Sony can get developers to make better use of RT is definitely questionable.

 

[mikeymikec]

I think upscaling may well be of importance for older hardware, but I can't help but think that it's going to be used as an excuse for no real innovation, ie. graphics cards being sold that are no better than their predecessors and what you're really being sold is a bit of software that could run on older graphics cards but that would undermine the point of you buying new hardware, so there's a bit of hard coding in the software that says DISABLE_ON_PREV_GEN=1. Then when a gen becomes "too old" according to the chip manufacturer, disable that feature in the last driver, then yank older versions of the driver from their website.

I've got a bit of a love/hate relationship with the whole idea. I mean, for starters I buy Blu-Rays and then save them to files using Handbrake. In the process I'm cutting the filesystem footprint by about 90%. Surely I must have lost something in the process, but I can't tell what it is. If therefore I can play a game while using half the power footprint and the game looks the same without upscaling, that's great surely?

Overall my current conclusion is to play without upscaling on my new setup and probably only eventually and grudgingly will I enable it.

 

[Heartbreaker]

Overall my current conclusion is to play without upscaling on my new setup and probably only eventually and grudgingly will I enable it.

The thing is that DLSS isn't just upscaling.

It's upscaling, and superior AA, and if the game includes DLAA, it's native with superior AA.

That's the main reason I wanted DLSS, not performance. TAA/FXAA just sucks, and that is pretty much what modern games have.

For Older games I love combining MSAA, and DLDSR. That is a really great AA combo.

 

[psolord]

New quick test with a twist. The twist being native dlaa vs ultra performance only. Game is Gangs Of Sherwood, maxed settings.

4k native dlaa


4k dlss ultra performance


Now here's the crazy part. Crazy because it's just dawned on me.

According to this videocardz, article, the scale factor is 3X for ultra performance, to 4k.

NVIDIA preparing Ultra Quality mode for DLSS, 2.2.9.0 version spotted - VideoCardz.com

NVIDIA DLSS Ultra Quality Redditor spots a new DLSS (Deep Learning Super Sampling) mode. The new “Ultra Quality” mode is listed as a placeholder in the UE5 documention. According to the footnotes, it should not be visible to end-users, which implies that the mode is still under development. The...

videocardz.com

However this is per dimention. The real difference is 3X3 for the total pixel count.

3840​	2160​	8294400​
1280​	720​	921600​
		11,11​

Ultra performance, is only 11% of the native 4k image and you see above their difference. Sure 4k native (+dlaa mind you) is quite better, but come on, it sure as hell does not have only 11% of the initial quality. That's some black magic right there.

Also yeah, the power draw gets less than 1/3. You'd expect less after that 11% pixel count, but them tensor cores don't run on atmospheric air it seems. The motion part of this scene was not too bad either.

 

[CP5670]

I like having scaling available but do notice the impact of it, mainly in games where you can see stuff very far away. Often trees, NPCs, etc. in the distance get blurred slightly.

As others have said, the big issue is both Nvidia and AMD will just keep selling new cards with the same performance at the same price, and using this to justify buying a new card.

 

[SolidQ]

Need more dark with RTX!
Half Life 2

 

[Ranulf]

I guess it makes the head crab harrys more scary or something.

 

[coercitiv]

New quick test with a twist. The twist being native dlaa vs ultra performance only. Game is Gangs Of Sherwood, maxed settings.

There are diagonal stripes all over the water surface in the DLLS Ultra Perf screenshot.

 

[BFG10K]

Need more dark with RTX!

This was one of the biggest lies of ray tracing, namely "just flip a simple switch and it automatically works everywhere without developer effort!"

In reality you need properly placed light sources or areas look pitch black when they shouldn't. You can easily see this when it's added to old games that were never designed for it.

You also need properly designed materials, surfaces and textures, or the lighting looks completely wrong, like a rough wooden surface that ends up looking like wet glass.

 

[Gideon]

Ultra performance, is only 11% of the native 4k image and you see above their difference. Sure 4k native (+dlaa mind you) is quite better, but come on, it sure as hell does not have only 11% of the initial quality. That's some black magic right there.

Also yeah, the power draw gets less than 1/3. You'd expect less after that 11% pixel count, but them tensor cores don't run on atmospheric air it seems. The motion part of this scene was not too bad either.

Upscaling is nice and definitely useful but IMO comparing still screenshots is quite deceptive, as by the nature of this tech, the image stabilizes over multiple frames. Still comparisons are a sort of the best case for DLSS / FSR.

I usually have very few issues with even DLSS Balanced at 1440p (provided a good implementation), evenscrutinizing small details usually holds up very well, while the image is relatively still. Yeah there might be some instability here and there (particularly with fine detail lines or objects far away) but it's surprisingly decent.

What I don't like at all s how upscaling handles motion in most games. Just look at the lights / lampposts and all-around jaggies here while in motion (timestamped):

And that is with huge YT compression smoothing things over. It's definitely a lot worse in game (to me).

I know, some people apparently can't see it at all, but it's hugely distracting to me (along with LOD pop-in in other areas) . In fact I find it only borderline (barely-barely) bearable at 1440p DLSS Quality, forget lower settings. I think it would take 4K + Quality DLSS for me to not be distracted by it at all.

Another take. Take a look at the overhanging wires here, DLSS performance vs quality:

I honestly, immediately see these things, and it takes me back to running HL2 with AA off in 2004.

I guess CP2077, AW2 and other RT-heavy games are some of the worst offenders (due to a lot of denoising among other things) but these games are also the ones that require upscaling the most.

 

[coercitiv]

Upscaling is nice and definitely useful but IMO comparing still screenshots is quite deceptive

Even still shots can have issues. In the case above the DLSS screen capture is full of diagonal lines on the water surface. This might simply be a temporary engine bug, but the fact that @psolord failed to spot the issue while comparing the two outputs is a red flag in itself.



Upscaling is great for people looking to improve performance or lower GPU power consumption. Presenting it as the equal or near-equal to native rendering is a false narrative though, there is no free lunch to be had here.

 

[igor_kavinski]

This might simply be a temporary engine bug, but the fact that @psolord failed to spot the issue while comparing the two outputs is a red flag in itself.

He's a swell guy. Generous to a fault. I bet his kids love him!

 

[BFG10K]

...but the fact that @psolord failed to spot the issue while comparing the two outputs is a red flag in itself.

All of this could've been avoided if "proper settings" were used, yo.

And people seem to forget basic res upscaling with a simple filter has none of the "aye eye" issues like ghosting, it just looks blurrier than native.

 

[Heartbreaker]

Even still shots can have issues. In the case above the DLSS screen capture is full of diagonal lines on the water surface. This might simply be a temporary engine bug, but the fact that @psolord failed to spot the issue while comparing the two outputs is a red flag in itself.

View attachment 91617

Upscaling is great for people looking to improve performance or lower GPU power consumption. Presenting it as the equal or near-equal to native rendering is a false narrative though, there is no free lunch to be had here.

It's misleading to use an image from "Ultra Performance" mode to shoot down the near-equal to native argument.

That argument really only about Quality mode, and in Quality mode it generally holds up. Here HWUB does extensive testing. He's finds that about 40% of the time DLSS Quality is not only as good as Native, it's better:

Given that, it's not unreasonable to call it near-equal when nearly half the time it's actually better.

A feature that renders better than Native 40% of the time, and also improves performance at the same time, is one I definitely want in my toolbox.

 

[psolord]

Even still shots can have issues. In the case above the DLSS screen capture is full of diagonal lines on the water surface. This might simply be a temporary engine bug, but the fact that @psolord failed to spot the issue while comparing the two outputs is a red flag in itself.

View attachment 91617

Upscaling is great for people looking to improve performance or lower GPU power consumption. Presenting it as the equal or near-equal to native rendering is a false narrative though, there is no free lunch to be had here.

I said that native was better. It always is. But for 11% of the total pixel count, the end result is great. I repeat this was an ultra performance test, 720p upscaled to 4k. Just an example.

 

[Aapje]

I said that native was better. It always is.

Rendering the images at up to 24 hours per frame like Hollywood does would be even better.

But any realistic comparison has to compare the same load on the system. It doesn't really matter if spending twice the GPU resources has a better result.

Also, native often has serious aliasing defects which is why people almost always use it with anti-aliasing. So what you call native is probably not native, but native + anti-aliasing.

 

[Timorous]

It's misleading to use an image from "Ultra Performance" mode to shoot down the near-equal to native argument.

That argument really only about Quality mode, and in Quality mode it generally holds up. Here HWUB does extensive testing. He's finds that about 40% of the time DLSS Quality is not only as good as Native, it's better:

Given that, it's not unreasonable to call it near-equal when nearly half the time it's actually better.

A feature that renders better than Native 40% of the time, and also improves performance at the same time, is one I definitely want in my toolbox.

As you stated earlier DLSS is not just upscaling, it replaces the AA and as we know in game TAA is often rubbish.

If you want to do a true comparison of a native image to an upscaled image you need to do the test as DLAA vs DLSS. Now if you do that test it will never be better than native, it might on occasion be close enough but it won't ever be better.

 

[Heartbreaker]

As you stated earlier DLSS is not just upscaling, it replaces the AA and as we know in game TAA is often rubbish.

If you want to do a true comparison of a native image to an upscaled image you need to do the test as DLAA vs DLSS. Now if you do that test it will never be better than native, it might on occasion be close enough but it won't ever be better.

That not native unless Native only runs on NVidia cards...

It's essentially comparing DLSS Ultra Quality vs DLSS Quality. No surprise that DLSS Ultra Quality is better than DLSS Quality.

 

[Timorous]

That not native unless Native only runs on NVidia cards...

It's essentially comparing DLSS Ultra Quality vs DLSS Quality. No surprise that DLSS Ultra Quality is better than DLSS Quality.

Initially native just meant the screen resolution and the image resolution match and it only became a thing because running below an LCDs design resolution looks like trash where as an old CRT is fine with various resolutions so CRTs did not really have a native image. You could change it for performance reasons and did not have the huge IQ downgrade you get on LCDs when you run below screen resolution.

Arguably by that initial definition DLSS is 'native' because the output image is the same resolution as the display so the term has now evolved to mean input resolution matches the display resolution.

The issue is plenty of games have TAA which you can't turn off so you have now introduced an additional variable because FSR and DLSS use their own TAA method when you turn them on. Comparing DLAA to DLSS or Native FSR to FSR removes the TAA as a variable so the comparison can focus on the quality of the upscaling in isolation.

If you have the performance to run at native you should always turn on DLAA if available IMO because it is a free IQ upgrade. This is something Alan Wake 2 actually does as it uses the FSR or the DLAA TAA solution depending on GPU rather than using its own TAA. Performance is identical in both cases although DLAA IQ, especially in motion is better
because the NV TAA is by far the best you can get at the moment and in game TAA or AMD / Intel needs to catch up.

 

[Rebel_L]

Do people actually play games with their AA turned way up? I only ever turn it up just enough to get rid of any obvious shape problems but no more so that fine detail edges dont start blurring on me.

 

[Heartbreaker]

Do people actually play games with their AA turned way up? I only ever turn it up just enough to get rid of any obvious shape problems but no more so that fine detail edges dont start blurring on me.

What kind of AA are you talking about?

TAA, DLAA are either ON or OFF. They don't have levels. Super Sampling, MSAA had levels, but higher was just better.

I can't remember any AA method that had levels where they higher levels were worse in any way.

I HATE Aliasiing artifacts. So I've always run the best/most expensive AA my GPU can handle.

Now I can even double up on AA.

By far the most impressive AA I've seen in game so far is from combining 8X MSAA with NVidia DLDSR 2.25X.

It's pin sharp and pretty much completely artifact free, and the kicker is, it stays pin sharp and artifact free in motion. I see a bunch of small branches in the distance and walk toward/away/around the tree, and everything is just sharp clear and fully anti-aliased in motion.

I wish I could do that in more modern games.

TAA is hot garbage in comparison.