Introduction

Ghost of Tsushima has finally released on PC, with several enhancements over its original PlayStation 4 release, such as Intel XeGTAO, NVIDIA HBAO+, full support for ultra-wide screens and Direct Storage 1.2. The PC release also has support for NVIDIA’s DLSS Super Resolution, Frame Generation (also known as DLSS 3), NVIDIA’s Deep Learning Anti-Aliasing (DLAA), AMD’s FidelityFX Super Resolution 3 (FSR 3) upscaling and Frame Generation, and Intel’s Xe Super Sampling 1.3 (XeSS 1.3) from day one. With the latest version of XeSS, Intel introduced two new upscaling profiles: “Native AA” and “Ultra Quality Plus” with 1.3x resolution scaling, and they also changed the resolution scaling for existing upscaling profiles. “Ultra Quality” has shifted from 1.3x to 1.5x, “Quality” mode has increased from 1.5x to 1.7x, “Balanced” mode has bumped from 1.7x to 2.0x, and “Performance” mode has risen from 2.0x to 2.3x. All implemented upscaling solutions are able to use Dynamic Resolution Scaling (DRS) at 30, 45 and 60 FPS, a very welcome feature. When DRS is active, the internal resolution will scale from 100% to a minimum of 50% in more graphically demanding scenes. In order to run Ghost of Tsushima at maximum graphics settings and reasonable framerates at native resolution, quite a powerful GPU is required, which is why upscaling solutions are so important. But depending on the game, there can be differences in the implementations of NVIDIA’s DLSS, Intel’s XeSS and AMD’s FSR, so we are keen to take a look at how temporal upscalers perform in Ghost of Tsushima.

Below, you will find comparison screenshots at 4K, 1440p, 1080p, and in different XeSS, FSR and DLSS quality modes; the TAA, DLAA, DLSS Frame Generation and FSR 3 Frame Generation screenshots are also available in the dropdown menu. For those who want to see how these technologies perform in motion, watch our side-by-side comparison video. The video can help uncover issues which are not visible in the screenshots, such as shimmering or temporal instability.

Our main testing was made using a GeForce RTX 4080 GPU at Very High graphics settings. We also used a GeForce RTX 3080 and Radeon RX 7900 XT for additional testing. Motion blur, depth of field, film grain, vignette and chromatic aberration were disabled for better image viewing. DLSS Super Resolution and DLSS Frame Generation in this game shipped with version 3.7.0.

Screenshots

Side-by-Side Comparison Video

Conclusion

Right off the bat, we would like to point out several major issues that Ghost of Tsushima currently has. The first issue is a very underwhelming image quality when using the native TAA solution. The TAA implementation in this game has an extremely pixelated and shimmery look on the whole image across all resolutions, even when standing still, and this becomes much worse in motion. The amount of artifacts is unusually high to the point where we think the native TAA solution just does not work correctly at the moment, and needs to be fixed in future game updates. The second most noticeable issue is that the quality of the depth of field effect when upscaling is enabled. In our routine testing of upscaling solutions we usually disable depth of field for better visibility, however, as this game has depth of field enabled by default, we played the introduction sequence with it enabled and noticed that all objects in the distance that were affected by depth of field are extremely pixelated and have a low resolution presentation. These artifacts are especially visible on small particle effects such as fire effects, and around characters during cinematic cutscenes. Also, the sharpening sliders for upscaling and antialiasing solutions are completely missing in this game, instead, the developers have set custom sharpening filter values. TAA, DLAA and DLSS are set to all use a small amount of sharpening in their render path, FSR 3 uses a medium amount of sharpening and XeSS does not use sharpening at all, making it look noticeably softer in comparison to other upscaling and antialiasing solutions.

Intel’s XeSS comes with three upscaling kernels that are optimized for various architectures. The first is the kernel that gets used on Intel Arc GPUs with their XMX engines. This is the most advanced model too, that not only performs better in terms of FPS, but also offers the best upscaling quality, the “Advanced XeSS upscaling model.” Intel also provides an optimized kernel for Intel Integrated Graphics, and another compatibility kernel, used for all other architectures that support Shader Model 6.4, e.g. all recent AMD and NVIDIA cards. These use the “Standard XeSS upscaling model,” which is somewhat simpler, with lower performance and quality compared to the Advanced XeSS upscaling model (we use the compatibility model with our test system’s RTX 4080). If DP4a instructions aren’t available, as on the Radeon RX 5700 XT, slower INT24 instructions are used instead.

Ghost of Tsushima’s game world is surrounded by different varieties of trees and vegetation, many of which are always in motion due to a dynamic wind system and other weather effects. In this game, FSR 3 is the only upscaling solution that struggles to retain details in moving vegetation, resulting in shimmering and pixelation. However, the amount of shimmering in motion is still lower when compared to the native TAA solution or other FSR implementations that we’ve tested previously, and the shimmering is most visible at low resolutions, such as 1080p. The FSR 3 image also has a degraded quality of particle effects, such as fire effects or waterfalls, and the typical disocclusion artifacts with pixelation in motion are also present, especially around Jin when moving through the world or during combat. Running FSR 3 in “Native AA” mode will reduce the visibility of these artifacts, but unfortunately won’t eliminate them completely. On the other hand, XeSS in its DP4a mode handles all of these small details much better, producing a very stable image in motion, however, because of the major difference in the developer set sharpening filter values between FSR 3 and XeSS, the overall softness of the XeSS image in this game may not be very appealing for some people.

For those who been looking for a good balance between image quality and performance—DLSS Super Resolution is the best option. The DLSS implementation in this game delivers the most stable and clean image when using upscaling across all resolutions. Compared to the native TAA image, even 1080p DLSS Performance mode (540p internally) will have a much better image quality, which further indicates that something is definitely wrong with the native TAA solution. With DLSS enabled at 1440p and 4K you can expect an improved and stable level of detail with particle effects, along with superior fidelity of tree leaves and vegetation rendering, providing a more comprehensive and immersive presentation, particularly during motion, when compared to the native TAA or FSR 3/XeSS solutions. With DLAA enabled, the overall image quality improvement is even higher, offering the best graphical experience overall when compared to the TAA solution, FSR 3, DLSS or XeSS.

Surprisingly, Ghost of Tsushima is the first game that officially allows FSR 3 Frame Generation to be used in conjunction with any available antialiasing or upscaling solution, including DLSS or DLAA, which grants us the ability to test AMD’s Frame Generation solution without artifacts from FSR upscaling and directly compare it to NVIDIA’s Frame Generation. When using DLSS as the base image for both Frame Generation solutions, we’ve noticed a few major differences: AMD’s Frame Generation image has some weird small black pixels around trees in the distance that are constantly flickering when standing still, and half of them will disappear when you start moving the camera. These artifacts are present with any antialiasing or upscaling solution as the base image, which indicates that they are caused by AMD’s Frame Generation alone. The second most noticeable difference is stability of shadows in motion: with AMD’s Frame Generation enabled, the player’s shadows are extremely jittery in motion, and it becomes significantly worse when the shadows are interlacing with the grass. The amount of these artifacts varies depending on your framerate. For example, running the game at 60 FPS with AMD’s Frame Generation will have significantly more jittery shadows compared to running the game at 120 FPS. NVIDIA’s Frame Generation has the same issue, however, these shadow jittering artifacts occur only when interlacing with the grass and the amount of jitter is also lower compared to AMD’s Frame Generation. Also, AMD’s Frame Generation solution has a slightly softer overall image compared to NVIDIA’s Frame Generation. To alleviate any concerns over the GPU hardware used, we tested FSR 3 Frame Generation in this game using not only a GeForce RTX 4080 GPU, but also a GeForce RTX 3080 and Radeon RX 7900 XT, to see how FSR 3 Frame Generation would perform on different GPU architectures, but we had the same results on those GPUs as well.

Speaking of performance, the DLSS, XeSS 1.3 and FSR 3 implementations in Ghost of Tsushima offer a solid 35% performance boost in “Quality” mode compared to native rendering at 4K resolution, and around 30% at 1440p and 1080p resolutions. With DLSS Super Resolution in “Quality” mode and DLSS Frame Generation enabled, you can expect doubled performance across all resolutions compared to native rendering, with AMD’s Frame Generation producing slightly more frames in GPU limited scenarios than NVIDIA’s Frame Generation solution, albeit with a worse image quality. The DLAA solution has a performance cost of around 2%, compared to the TAA solution, but offers the best graphical experience overall, and as mentioned, can be combined with DLSS Frame Generation and even with FSR 3 Frame Generation for a better than native image quality and FPS.