Visual analysis I
Alan Wake II is the first game to support Nvidia DLSS 3.5 from the start. In addition to the technological aspect, there is also the high popularity among gamers. This gives us the basic reasons to take a close look at the performance under different settings. In diving in with the gaming performance, we will be interested not only in the visual side, but also the power consumption. Not just of the graphics card, but the CPU as well.
Disclaimer: This article was written in close collaboration with Nvidia. However, all thought processes and data are our own. Nvidia did not interfere with any of the “substance”.
You already had the chance to read the analysis of DLSS 3.5 technologies on HWC. The big new features are Frame Generation and Ray Reconstruction, which we have explained in an earlier article (linked below). In this piece, we are going to look at the impact of these features in the Alan Wake II game where we are going to test them.
The content of the article can be divided into three parts, where in the first Jan Olšan focuses on the analysis of the visual quality of various scenes. First it’s going to be the visual impact of DLSS Ray Reconstruction when used with native resolution (or better said, with DLAA) and then the impact of the various quality and speed tradeoff settings of the DLSS upscaling technique.
For all the settings, details of which can be found in the test methodology description, gaming performance is also measured. So you can easily keep track how the visuals smoothness, right off the bat on graphics cards with all GeForce RTX 4000 Super GPUs (RTX 4070 Super, RTX 4070 Ti Super and RTX 4080 Super). Lastly with the RTX 4080 Super card, and this is the third part, we also measure the power consumption associated with each setting. We look at graphics card consumption as usual, but also at CPU consumption. The latter is also dependent on the graphical detail settings, where each leads to a slightly different load. After taking all the criteria into account, which one is right for your taste?
Image comparison of the effect of Ray Reconstruction in DLSS 3.5
The first scene
The first scene should hopefully show the improvement in ray tracing quality with Ray Reconstruction. This scene shows an interesting shadow complex behind the staircase attached to the hangar. In the right screenshot, which is with Ray Reconstruction enabled, the shadows are much sharper, which could be correct due to the sharp light of the setting (or is it rising?) sun. Without ray reconstruction (left half), the shadow’s edges are much less sharp, which we assume is not the intention here, but a manifestation of how the shadow calculation works with less resolution.
By the way, without raytracing effects, the game basically doesn’t draw a shadow of this structure, or only a hint of it.
In addition, there is a difference in the reflection on the water (reflections are one of the main things raytracing effects handle). In this case, the effect that Ray Reconstruction has on the scene (in the right half of the comparison) is ironically not sharpening, but the opposite.
In the left half of the screenshot you can see that the reflection of the warehouse and staircase in the water is kind of grainy (and quite rough looking), while on the right it’s smoother. In this case it’s probably not the case that the blurriness is indicative of a lower resolution, here the smoothness of the image in the water is probably desired. The coarseness on the left should probably be a manifestation of the lower resolution of the reflection, where the combination of the individual subsamples fails to form a smooth shape properly (probably also because the image is affected by the motion of the simulated water surface) to produce a completely smooth image. Thus, we suspect that the look in the left half (without Ray Reconstruction) is more a manifestation of aliasing rather than a sharper image.
The next comparison will still be the same scene, but we will move to the right to a different part of the image.
Here again you can see that switching on Ray Reconstruction (right) smoothens the image in the water, which without Ray Reconstruction (left) shows the aforementioned graininess, probably due to aliasing caused by low resolution. This is quite evident in the comparison of the straw bale reflections, which do not match their conception above. On the other hand, not everything is ideal in this screenshot on the right, you can clearly see the heavy aliasing on the wires, which are completely smoothed out and rendered with a clear sharp line in the left image, whereas on the right the DLAA anti-aliasing failed to join them and they remain jagged and intermittent in ugly ways. However, this may not be a defect of Ray Reconstruction. These artifacts are caused by the failure of temporal anti-aliasing and reconstruction, which are handled by DLAA (and DLSS). These artifacts sometimes appear, sometimes disappear, during motion. So it might rather be bad luck that the temporal aliasing mechanism (within DLAA) faltered exactly in this time ad place.
These flaws, the associated shimmering and temporal anti-aliasing artifacts are most apparent when you see the game in motion, comparing just still images doesn’t capture the whole reality of this problem.
The second scene
In the second scene, Ray Reconstruction seems to have less benefit, but this may be because although we are looking at a large reflecting surface in the scene, which should be the ideal place for ray tracing quality demonstration (and Ray Reconstruction should have big benefit), there is the problem that the reflected objects are relatively dark, low contrast, and thus there is not as much chance of noticing details in them.
Even so, it seems that with Ray Reconstruction, sharper contours can be seen on the building which can be seen in the reflection on the café window. It is noticeable that with Ray Reconstruction, the wires that are reflected on the glass are better preserved and their thin line is less visible on the left. On the other hand, with Ray Reconstruction, the aliasing (jagged lines including those mentioned wires) again appears here.
To illustrate the importance of ray tracing for reflections, here’s what this scene in Alan Wake II looks like when ray tracing is turned off.
The article continues with further chapters.
Alan Wake II is the first game to support Nvidia DLSS 3.5 from the start. In addition to the technological aspect, there is also the high popularity among gamers. This gives us the basic reasons to take a close look at the performance under different settings. In diving in with the gaming performance, we will be interested not only in the visual side, but also the power consumption. Not just of the graphics card, but the CPU as well.
Image comparison of different DLSS settings
This comparison will be about the different levels of DLSS. This upscaling has progressive levels from Quality, Balanced, through Performance and Ultra Performance. For these, the internal resolution at which the game actually runs and from which DLSS 3.5 scales the image up to the target resolution (in this case 2560×1440) is gradually reduced – for example, for Performance it should be 1280×720 pixels, for Ultra Performance it’s only 853×480 pixels (the scaling factor is 3x).
We compare the DLSS image in each quality level with the DLAA mode, which is a special DLSS mode that does not use any upscaling, but applies the same temporal filtering and reconstruction (and that includes anti-aliasing similar to TAA) as DLSS. It’s probably a fairer comparison than comparing DLSS against a “native image” which lacks anti-aliasing.
Both versions use 2560×1440 resolution, Ray Tracing on and Ray Reconstruction on. We don’t have frame generation active (neither did the previous comparisons) so that we don’t end up examining the differences between a fully rendered and a generated image, which wouldn’t be fair.
DLSS Quality
For the first scene, we picked this poster space with text that should give the upscaling a run for its money. However, you can’t really see much of the impact yet in DLSS (Quality) mode, the highest quality and lowest speed mode, which is on the right side of the image. Or rather – there will be some degradation due to the lower rendering mode, but it’s hard to distinguish visually. Instead of posters, the stumbling block turned out to be the railing on the balconies. The difference is best seen when using split-screen mode rather than side-by-side display (we can’t use that here, unfortunately, because the javascript applet that allows you to embed a “sliding comparison” doesn’t work).
In general, the more aggressive the DLSS you use, the less distinct the individual vertical bars in the railing will be – a little in the first red one, but more so in the more distant yellow railing, where the perception of being made up of individual bars is lost and it starts to look like a sort of joint mass. Albeit only slightly so for now, in Quality mode.
In the second scene, the DLSS is perhaps most evident in the pattern of the fence to the right of the van. It is apparently so subtle that any undersampling and upscaling – even in the moderate Quality mode – already leads to a large distortions, with the upscaling AI getting fooled and the pixels merging into something else instead of the original pattern (it’s possible that not only DLSS but perhaps other smoothing methods will often crash and burn on this). You can see how the details are somewhat degraded in other places. For example, you can see how the contours of the conifers on the distant horizon are distorted, but Quality mode doesn’t yet do this as significantly as faster modes.
In this scene, we also noticed aliasing on the wires and also, for example, on the outline of the curb on the corner by the van. It doesn’t seem to be the fault of the upscaling itself as it happens with DLAA too.
DLSS Balanced
In Balanced mode, the effects already seen above are amplified. The blending of the bars in the yellow railing is more apparent, and the pattern of the red one is less clear. The text on the posters also starts to degrade a bit, but it’s not extremely sharp in the original either, which obscures the difference a bit.
But notice also how the contours on the staircase and the background visible to the right of the dark lantern column are lost. The bricks on the wall, which have also suffered a bit in Quality mode are also halfway to “fusing” together, but not quite there yet.
The second scene essentially repeats the effects already seen in Quality mode, but the breakdown of the fence pattern is more pronounced, as are the effects on the details of the tree silhouettes on the horizon.
DLSS Performance
The effects on the railings (yellow railings blending together, less clearly rendered structures) are again amplified, and again we have the effect of the lines of the staircase and other objects (bricks) disappearing into the fog to the right of the lamp post, while you can see them when using DLAA. Here you can see quite clearly how the low resolution negatively affects the quality of the text on the posters.
In the second scene we again observe the preogressive degradation of the same fence and spruce trees on the horizon, where the effect of upscaling is clearly visible and the negative impact on the quality of detail is logically greater now. But in Performance mode, the reduced detail is already noticeable more generally, even on buildings for example.
Ironically, the appearance of the wires has improved in Performance mode. They’re not as prominent, they’re thinner, less dark, and more blended into the background, so even if they’re not free of aliasing, it’s less distracting in the end. But the aliasing on that curb has seemingly disappeared completely. The question is whether this points to some problem at higher settings, or whether Performance mode simply enjoys a lucky side effect of the overall blurring caused by the lower resolution, which manages to alleviate the problematic aliasing a bit (since blurring is one way to smooth out aliasing).
DLSS Ultra Performance
DLSS Ultra Performance with its extreme scaling, which is no longer suitable for lower target resolutions (it was originally presented as a solution for gaming at the target resolution of 8K, to which it upscales from 2560 × 1440), has the worst effects on quality, of course, every adverse effect mentioned earlier is strongest here.
The quality of the text has deteriorated quite a bit, but there is a general blurring on the walls, and the railings and ladders are looking really garbled. The sharpness of the foliage on the tree has also deteriorated significantly, and you can even see the blurring of the container visible in the lower right corner in the foreground.
This image already shows a lot of signs of being upscaled, but again, you have to give it credit that the temporal reconstruction saves a lot and the results aren’t as bad as you’d expect from a 3x scaling factor. You have to keep that in mind and don’t judge the quality degradation as that much of a negative, precisely due to the fact that an unsophisticated upscaling would have turned out even worse. And it’s not that the game isn’t playable with this setup. Considering that Ultra Performance improves FPS pretty significantly, that’s probably acceptable to some people.
The second scene also has quite a lot of signs of upscaling, you can already see the worse level of detail when comparing, for example, the texture of the asphalt in the foreground, and all of the previously cited problems are again more pronounced here, the distortion or outright destruction of the fence pattern is particularly strong here. The trees with their autumn leaves have also become much coarsely-drawn. The nice thing is that the “blur” is not noticeable at first glance, but when you compare it you can see a big change, they starts to look a bit like a pointillist painting. The details and and rendering deteriorate pretty much everywhere (even on the rather high contrast traffic lights).
Ironically, the situation with the aliasing of wires has improved again, they are smoother and actually look better. However, due to the low real resolution, they often completely disappear in places, which was not as much of a problem in Quality mode.
In general, however, the quality of this Ultra Performance mode is already reduced by quite a bit (which is to be expected, of course) and doesn’t match what a 2560×1440 image should look like. Those FPS gained in Ultra Performance mode are definitely not coming free of cost, it should be pointed out. So, just as one can’t say that this mode is unusable, one can’t somehow downplay the negatives either.
Artifacts are the price of higher performance, not the end of the world
Perhaps we should clarify that all these DLSS modes have their justification precisely because of their performance benefits. It has to be clarified that we’ve been focusing on finding the differences here, and thus the talk has always been about the downsides, se perhaps it sounds as if DLSS has nothing but negatives. That is not the case. But you have to remember that its goal is higher performance, which means for example that you can play the game at smooth FPS where without DLSS, you simply wouldn’t be able to. So those flaws in the image should be considered while keeping in mind that you are getting something (performance) in return. And in that context, they may be acceptable to you (the alternative, which is to flat out lower the output resolution of the game, also degrades the quality). So we don’t want to denigrate DLSS, you can see the positive side in the performance graphs in the next chapters.
It’s probably wise to avoid the Ultra Performance extreme mode if you can keep the game at playable FPS without it. Though even the negative impact of Performance mode is quite noticeable, it’s still not as bad for the visuals. But the higher you go with the settings, the less impact DLSS has, and proportionally, you don’t have to worry as much about Balanced and Quality.
The lesson here is unsurprising: If you want to enjoy the visuals and you’re not playing competitively against anyone in a multiplayer shooter or rally suffering from not having a smooth framerate, it’s a good idea stay away from unnecessarily aggressive upscaling modes than you need (just in order to get a higher number on the FPS counter in the corner), instead try to use the least aggressive DLSS mode you can still afford with your GPU performance. Which is probably obvious, and I don’t expect anyone needs to be reminded of that…
Alan Wake II is the first game to support Nvidia DLSS 3.5 from the start. In addition to the technological aspect, there is also the high popularity among gamers. This gives us the basic reasons to take a close look at the performance under different settings. In diving in with the gaming performance, we will be interested not only in the visual side, but also the power consumption. Not just of the graphics card, but the CPU as well.
How we tested
The measurements were performed in the same conditions as our standard graphics card tests. This means with active system cooling with strictly controlled air temperature (21-21.3 °C) at the wind tunnel intake. The accuracy of the measurements depends on it as well.
The general settings are set to the highest possible, as you can see in the screenshots below.
Only the inclusion and mode of DLSS 3.5 technology changed from test to test (Frame Generation and Ray Reconstruction) and rendering resolution, where we went from the highest, native (with DLAA) smoothly (through DLSS Quality, Balanced, Performance) to the lowest (DLSS Ultra Performance). We also have the results without ray-tracing graphics for comparison.
All measurements are performed in three passes, from which we plot the arithmetic average in the graphs, at 2160p, 1440p and also at 1080p. The test scene is in Bright Falls, starting at the beginning of the street where the heroine runs to the door of the diner at which the measurements end.
For the tests, the graphics cards used are the Nvidia RTX 4070 Super FE, the MSI RTX 4070 Ti Super 16G Ventus 3X and the Gigabyte Aorus RTX 4080 Super Master 16G (with th „OC“ BIOS). With the most powerful RTX 4080S, we also measure in detail the power consumption curves of both the graphics card and the processor. We do this traditionally using a combination of current clamps and a custom PCIe Express shunt.
The processor of the test build is the Intel Core i9-13900K without power limits cooled by the Noctua NH-U14S and with Patriot Viper Venom DDR5 memory (2×16 GB, 6000 MHz/CL36). Underpinning everything is the Asus ROG Strix Z790-A Gaming WiFi II (BIOS v0801) with ReSizable BAR enabled.
Graphics drivers are Nvidia GeForce 551.76 Game Ready. For a PSU, we have the BeQuiet! Dark Power Pro 12 (1200 W) and for completeness, SSDs – these are the Patriot Viper VP4100 1 TB (system storage) and the Patriot VPN100 2 TB (Alan Wake II is installed on this one). The operating system is Windows 10 (22H2).
Alan Wake II is the first game to support Nvidia DLSS 3.5 from the start. In addition to the technological aspect, there is also the high popularity among gamers. This gives us the basic reasons to take a close look at the performance under different settings. In diving in with the gaming performance, we will be interested not only in the visual side, but also the power consumption. Not just of the graphics card, but the CPU as well.
Game performance tests…
Alan Wake II is the first game to support Nvidia DLSS 3.5 from the start. In addition to the technological aspect, there is also the high popularity among gamers. This gives us the basic reasons to take a close look at the performance under different settings. In diving in with the gaming performance, we will be interested not only in the visual side, but also the power consumption. Not just of the graphics card, but the CPU as well.
Power consumption tests
Alan Wake II is the first game to support Nvidia DLSS 3.5 from the start. In addition to the technological aspect, there is also the high popularity among gamers. This gives us the basic reasons to take a close look at the performance under different settings. In diving in with the gaming performance, we will be interested not only in the visual side, but also the power consumption. Not just of the graphics card, but the CPU as well.
Conclusion
The key DLSS 3.5 technology – Ray Reconstruction – has a significantly greater impact in Alan Wake 2 than it does in Cyberpunk 2077. At higher image quality we measured a 13–21% higher framerate with RR. In Cyberpunk 77, the gain from this technology was only 1–4%, and that was with an even more powerful graphics card (the GeForce RTX 4090) than the ones used for the tests now – the RTX 4070 Super, the RTX 4070 Ti Super and the RTX 4080 Super.
Additionally, without Ray Reconstruction, Alan Wake II’s performance dips significantly in certain situations. The RTX 4070S at max graphics settings without RR can’t handle things at native 1440p anymore. From an average of about 41 fps, the smoothness drops to an unplayable 6 fps when Ray Reconstruction is turned off. The RTX 4070 Ti Super and RTX 4080 Super behave similarly, but only at 2160p resolution. These graphics cards feature the same 16GB of memory, while the RTX 4070 Super only has 12GB. So the bottleneck here may or may not be the VRAM size. Whatever the reason, this is where Ray Reconstruction comes in handy. With it, the RTX 4070 Super can run Alan Wake II fairly smoothly even at 1440p with DLAA, i.e. without upscaling where fps drops below 60 only sporadically. However, Frame Generation must be active for this to happen.
Combined with DLSS Quality, average fps is already above 100. This is, of course, where lower image quality comes in. If the goal is to keep it up while still achieving fast framerates at higher resolutions, there’s the 24% faster RTX 4070 Ti Super. Or then the RTX 4080 Super. Even with this graphics card, though it’s the most powerful of the trio tested, don’t count on completely smooth gaming at 2160p. For this it is necessary to reach for DLSS, or alternatively to reduce the graphical detail.
With DLSS Balanced though, you’re already averaging at a rather nice 90 fps with the RTX 4080S, and whether that will still look good enough you’ll have to judge for yourself. Flickering of distant objects in motion (such as power line cables) persists in some form in DLSS 2.x, but only very low resolution settings suffer from it. For example, 1080p in DLSS mode on Performance or Ultra Performance or 1440p on Ultra Performance. The “2160p” (16:9) render with Balanced is already relatively high resolution (2227×1253 px).
Beyond visual quality and speed (fps), operating power consumption can also be a factor when deciding on the optimal setting. There is also some variance in this, also with regard to CPU power consumption. It can’t be generalized, but the consumption of a graphics card usually goes up with higher resolution, and it’s the other way around for the processor. Its consumption increases towards lower resolutions. We have also mapped this in our measurements.
Unless you are trying to achieve a lower power consumption of the graphics card (e.g. to achieve lower noise level of its cooler), the total power consumption – GPU + CPU will usually be the guideline. There, the differences are already smaller, with the RTX 4080 Super (with Core i9-13900K) fitting in around 40W across settings. The highest power consumption, when wringing the maximum possible out of the hardware, is about 399 W, at 1440p natively, with DLAA, both Frame Generation and Ray Reconstruction enabled (with ray tracing on “high”).
The lowest consumption, unless you count the setting with RR off, where (for extremely low fps) a strong bottleneck is evident, is native 1080p with DLAA, with DLSS on Ultra Perforance, with Frame Generation on, at full raytracing graphics with RR. In this setting, the highest fps is achieved, albeit at the expense of the lowest visual quality. Users looking for a balanced setup will consider a different option, especially with an upper-midrange GeForce graphics card.
So under what circumstances does Alan Wake II perform best, optimally? Everyone will probably see it a little differently.
English translation and edit by Jozef Dudáš