Value of Hardware Unboxed benchmarking

[Shifty Geezer]

I disagree. It's just showing RT in action and how limited ray counts affect results. If they aren't obvious, viewers would watch the video and think, "what are you going on about? I'm not seeing any shimmer." If they are obvious, the viewers will see for themselves and think, "yeah, I can see that shimmer/boiling/whatever. I don't like it/can live with it/am not bothered by it."

the most egregious aspect is the title, but it's true RT does have a noise problem because there's finite power to trace rays and on ordered grid, the remedies are limited and imperfect. That's about as controversial as saying, "PS2 has a shimmer problem!" As you say yourself, it's inevitable .

Generally when you lose half or more of framerate RT is transformative to the overall rendering to a point where some artifacts it has (mostly because it has to run on slow RT h/w mind you) are vastly outgunned by the sheer improvements it brings with it

Mentioned a little in the video. RT is better overall, but has noise. Its not a video about the value of RT.

which generally results in a removal of old artifacts.

Generally? Sometimes it adds an effect which comes with additional artefacts. eg. No reflections versus RT reflections. RT reflections adds noise the non-RT version obviously doesn't have. Dynamic GI versus baked GI - baked GI comes with zero animated artefacts.

But that's all really besides the point because, again, there's no claims being made for non-RT quality. This is not claiming to be an investigation into video game artefacts. There's no suggestion non-RT is artefact free, or better. It's just putting one aspect of contemporary games tech under the magnifying glass.

 

[DegustatoR]

RT is better overall, but has noise.

RT has noise because the h/w power is limited and the amount of rays traced is low. This is an obviously solvable problem with future h/w upgrades.
None of non-RT artifacts - of which there are plenty in literally every non-RT game out there - can be solved by putting more h/w power into the old ways of rendering.
This is the main reason why this video completely misses the point.

 

[IQandHDR]

In my book, the less RT effect used, the lesser the visual impact.

The games I own that have had the biggest visual impact are:

Black Myth Wukong
Cyberpunk 2077 Overdrive
Minecraft RTX
Quake 2 RTX
Doom 2 Raytraced
Quake 1 Raytraced
Half Life 1 Raytraced

And now we have games like "Indiana Jones and The Great Circle" being made from the ground up for RT.
The bigget jumps in eg. CyberPunk 2077 has been from PT + RR
That nvngx_dlssd.dll when implemented really does a lot for the visual quality.

 

[troyan]

I disagree. It's just showing RT in action and how limited ray counts affect results. If they aren't obvious, viewers would watch the video and think, "what are you going on about? I'm not seeing any shimmer." If they are obvious, the viewers will see for themselves and think, "yeah, I can see that shimmer/boiling/whatever. I don't like it/can live with it/am not bothered by it."

Most problems go unnoticed because they do not matter. Indiana Jones is a great example how low quality rasterizing effects can be more problematic than a "noisy" Pathtracing solution:

 

[trinibwoy]

Tim mentioned several times that the overall presentation is better with RT enabled. This wasn’t some crucifixion of RT. He’s right that we need faster hardware. The biggest disappointment is that current games won’t scale up to take advantage of better GPUs.

 

[homerdog]

Generally when you lose half or more of framerate RT is transformative to the overall rendering to a point where some artifacts it has (mostly because it has to run on slow RT h/w mind you) are vastly outgunned by the sheer improvements it brings with it - which generally results in a removal of old artifacts. An example here would be any game with path tracing.
And vice versa when RT isn't bringing much in terms of improvements its performance impact is usually rather small and the artifacts it may have aren't any different from what you'd get without RT. An example: DATW where RT is a) cheap, b) doesn't have any artifacts at all and c) removes a lot of those as it substitutes SSAO and SSR.

I agree but this is subjective. Path tracing is amazing and I'd use it whenever it's available unless input lag becomes overbearing. But we're talking about a massive performance hit. From what I've seen most gamers prefer more frames to better frames, even if the path tracing mode looks way better. Especially if they have anything below a 4080 (vast majority of gamers).

In any case I see no problem with HUB pointing out deficiencies in current RT implementations. Not every analysis has to be a comparison to other techniques.

 

[troyan]

Tim mentioned several times that the overall presentation is better with RT enabled. This wasn’t some crucifixion of RT. He’s right that we need faster hardware. The biggest disappointment is that current games won’t scale up to take advantage of better GPUs.

It is. Here is an example from Alan Wake 2:

This guy knows exactly that moving the camera would break these SSRs. So he is just moving the character. Its the definition of "hit piece".

 

[homerdog]

It is. Here is an example from Alan Wake 2:

This guy knows exactly that moving the camera would break these SSRs. So he is just moving the character. Its the definition of "hit piece".

He couldn't compare the noise if the reflection isn't present in the SSR shot.

 

[troyan]

He couldn't compare the noise if the reflection isn't present in the SSR shot lol.

Which makes it a useless comparision. Nobody need Raytracing, when you can fake everything. That is the point: Raytracing is used to archive more - for example stable reflections. These noisy reflections are better than having no reflections at all.

 

[homerdog]

Which makes it a useless comparision. Nobody need Raytracing, when you can fake everything. That is the point: Raytracing is used to archive more - for example stable reflections. These noisy reflections are better than having no reflections at all.

He is comparing the noise in the reflections so he'd have to point the camera in a way that the reflections are present in the SSR shot.

I agree that noisy RT reflections are better than SSR disappearing unstable crap, but he is specifically comparing noise. It even says so in the title of the video.

 

[troyan]

The reason these SSRs are not as noisy is just that they can used these rays for this limited part of the scene. Battlefield 5 is doing this for the raytraced reflection - sampling these rays and using is on upto 25% of the screen to get full resolution reflections.

He is comparing a dynamic full pathtraced image to a limited area of the scene. Makes absolute no sense.

 

[Shifty Geezer]

RT has noise because the h/w power is limited and the amount of rays traced is low. This is an obviously solvable problem with future h/w upgrades.

You realise that's exactly what the video is about, right? Which bit of his conclusion do you disagree with? That we need better hardware that can sample more rays, or that we also need better denoisers that can handle lower sample counts?

Most problems go unnoticed because they do not matter.

Which has nothing to do with this video. It isn't claiming anything for/against non-Rt solutions. It's just talking about the limits of RT and the affect of different RT solutions, denoising versus not.

This guy knows exactly that moving the camera would break these SSRs. So he is just moving the character. Its the definition of "hit piece".

In that example he's talking about how temporal lag on the denoising affects its clarity. That's not going to be evidenced without motion, is it? But in the scene just prior to that, when talking about the temporal denoising on the floor, he specifically states he prefers pathtracing on. So it's no condemnation of RT. It's just a look at some of the current shortcomings.

He is comparing a dynamic full pathtraced image to a limited area of the scene. Makes absolute no sense.

No, he's not. He's looking at RT. It's not a comparison to say non-RT is better. Non-RT doesn't come into it. All the complaints are looking at this video as being something it's not. It's looking at RT and the limitations that cause noise and the way noise is addressed, either by denoising with temporal issues, or just letting it be and having fuzzy issues. Which, for people who don't follow 3D tech so closely, might well be informative.

At 14:50, he highlights the shortcomings of SSR - "looks noticeably worse" - and states RT is preferable. The comparison is there just to see the difference in RT in terms of responsiveness. It wasn't a judgement on anything but an objective look at the pros and cons of the different solutions.

It's just a look at RT, which is limited, because the hardware is and the denoising is.

 

[homerdog]

The reason these SSRs are not as noisy is just that they can used these rays for this limited part of the scene. Battlefield 5 is doing this for the raytraced reflection - sampling these rays and using is on upto 25% of the screen to get full resolution reflections.

He is comparing a dynamic full pathtraced image to a limited area of the scene. Makes absolute no sense.

I think he's trying to show what noisy reflections look like. For all the problems SSR has, this example of SSR doesn't really have noise that I can see. So it's fine to point out what noisy reflections look like using the SSR shot as a control. It doesn't mean SSR is better overall (it's not).

He could also maybe mod RT to massively increase sample count to eliminate noise and use that as a control. IDK if that's possible but it would make the same point. RT reflections as currently implemented in games are noisy.

 

[Shifty Geezer]

I think he's trying to show what noisy reflections look like. For all the problems SSR has, this example of SSR doesn't really have noise that I can see. So it's fine to point out what noisy reflections look like using the SSR shot as a control.

He later uses the arcade in Cyberpunk and explicitly calls out the failings of SSR and mentions RT is better. His point and attempts to illustrate are on temporal lag in RT effects.

He could also maybe mod RT to massively increase sample count and use that as a control.

Not possible in realtime, and not necessary. There's nothing wrong in comparing a temporal solution against a non-temporal solution so long as you explain what you're looking at.

 

[homerdog]

He later uses the arcade in Cyberpunk and explicitly calls out the failings of SSR and mentions RT is better. His point and attempts to illustrate are on temporal lag in RT effects.

Not possible in realtime, and not necessary. There's nothing wrong in comparing a temporal solution against a non-temporal solution so long as you explain what you're looking at.

The comparison to SSR is causing a lot of confusion and I think that's causing people to miss the point. He's saying "This is a shortcoming of current RT implementations." It is what it is.

BTW it's impossible for me to see the noise when watching the video on my phone. So this problem is clearly solvable with more samples.

 

[Shifty Geezer]

Some of the examples are hard to see and benefit from being zoomed in on to see what's happening. I think high frequency noise is less egregious than long-duration smearing. Big bubbling can be pretty off-putting too, but hard to notice in low brightness areas. Might be able to skew samples based on brightness to get more samples where they are most needed. It's perhaps worth noting that film grain as 1) been accepted for a very long time and 2) deliberately added to games! So managing to affect the quality of noise towards something film-grain might be useful.

Indeed! How about adding more noise?! Deliberately turn the low noise into high frequency noise and break it up. When I think of mobile phone cameras and their heavy handed denoising, I'm much more appreciative of the aesthetic of natural analogue film grain. Perhaps denoising isn't the only path to take and artistic noise is an option?

 

[troyan]

Which has nothing to do with this video. It isn't claiming anything for/against non-Rt solutions. It's just talking about the limits of RT and the affect of different RT solutions, denoising versus not.

In that example he's talking about how temporal lag on the denoising affects its clarity. That's not going to be evidenced without motion, is it? But in the scene just prior to that, when talking about the temporal denoising on the floor, he specifically states he prefers pathtracing on. So it's no condemnation of RT. It's just a look at some of the current shortcomings.

No, he's not. He's looking at RT. It's not a comparison to say non-RT is better. Non-RT doesn't come into it. All the complaints are looking at this video as being something it's not. It's looking at RT and the limitations that cause noise and the way noise is addressed, either by denoising with temporal issues, or just letting it be and having fuzzy issues. Which, for people who don't follow 3D tech so closely, might well be informative.

At 14:50, he highlights the shortcomings of SSR - "looks noticeably worse" - and states RT is preferable. The comparison is there just to see the difference in RT in terms of responsiveness. It wasn't a judgement on anything but an objective look at the pros and cons of the different solutions.

It's just a look at RT, which is limited, because the hardware is and the denoising is.

No, this video is talking about problems which do not matter. He is specific comparing SSRs to RT-Reflections.

Here is an example from Battlefield 5:

Do you really think the "noisy problem" of Raytracing is really so important that someone think that SSRs are better? I can fake a rasterizing image to be more stable, but this is still a fake. Reality is that SSRs will be more noisy than Raytracing.

 

[homerdog]

Some of the examples are hard to see and benefit from being zoomed in on to see what's happening. I think high frequency noise is less egregious than long-duration smearing. Big bubbling can be pretty off-putting too, but hard to notice in low brightness areas. Might be able to skew samples based on brightness to get more samples where they are most needed.

It's funny because I notice more problems (especially ghosting and "boiling") in dimmer, indirectly lit areas. There is a lot of subjectivity involved here.

 

[DegustatoR]

You realise that's exactly what the video is about, right? Which bit of his conclusion do you disagree with? That we need better hardware that can sample more rays, or that we also need better denoisers that can handle lower sample counts?

I disagree with the part that we "need" all of that. We don't. RT as it is is improving graphics beyond what's possible without RT while simultaneously lowering the amount of artifacts we get in non-RT rendering.
Having better RT h/w is welcome of course, as well as more AI being used for denoising and in more advanced ways. Do we "need" it though to use RT right now? No.

The comparison to SSR is causing a lot of confusion and I think that's causing people to miss the point. He's saying "This is a shortcoming of current RT implementations." It is what it is.

It's a "shortcoming" of any sampling algorithm which does lower than 100% sample coverage. There are multiple examples of noisy SSR implementations out there. This whole idea that "RT has a noise problem" (sic) is just utterly wrong.

 

[homerdog]

It's a "shortcoming" of any sampling algorithm which does lower than 100% sample coverage. There are multiple examples of noisy SSR implementations out there. This whole idea that "RT has a noise problem" (sic) is just utterly wrong.

It is a problem in just about every RT game I've played. Some more than others. Unlucky