安装 Steam
登录
|
语言
繁體中文(繁体中文)
日本語(日语)
한국어(韩语)
ไทย(泰语)
български(保加利亚语)
Čeština(捷克语)
Dansk(丹麦语)
Deutsch(德语)
English(英语)
Español-España(西班牙语 - 西班牙)
Español - Latinoamérica(西班牙语 - 拉丁美洲)
Ελληνικά(希腊语)
Français(法语)
Italiano(意大利语)
Bahasa Indonesia(印度尼西亚语)
Magyar(匈牙利语)
Nederlands(荷兰语)
Norsk(挪威语)
Polski(波兰语)
Português(葡萄牙语 - 葡萄牙)
Português-Brasil(葡萄牙语 - 巴西)
Română(罗马尼亚语)
Русский(俄语)
Suomi(芬兰语)
Svenska(瑞典语)
Türkçe(土耳其语)
Tiếng Việt(越南语)
Українська(乌克兰语)
报告翻译问题
Another issue is that each setting for FSR when you are scaling, actually physically changes the render window at will, meaning you are just going to see the game shrinking resolution and enlarging itself as if you are zooming in and out.
Basically the output setting, the 'scaling' that the app does, has to render the actual individual pixels that comes from the source game, in a enlarged format good enough to see said pixels on your monitor. If the render window is locked to your monitor's screen, you won't see any improvements, and it'll just act like Supersample Resolution on a smaller resolution for better management of aliasing.
Basically, turning 720p into 1440p will still show it as a giant 720p on your monitor.
The idea behind the tests was this: FSR takes some time to do it's thing. So I thought, what if it was possible to use a slightly higher resolution and apply CAS instead of using FSR. And the main purpose was to compare: 4K vanilla - 1800p CAS - 1620p FSR.
Here is the test environment:
- AMD Radeon 5700XT
- AMD Ryzen 7 3800X
- 16 GB 3600 MHz RAM
- 500 GB Sata 3 SSD
Test conditions:
1) Vanilla resolution: 4K
2) CAS (0.25-0.50) applied on top of 3200x1800 image with classical GPU upscaling.
3) FSR Lite applied on top of both 3200x1800 and 2880x1620 images.
4) FSR applied on top of both 3200x1800 and 2880x1620 images.
Resolutions:
- 4K (1:1 scale)
- 3200x1800 (1.2x scaling)
- 2880x1620 (1.333x scaling) (Lossless Scaling should really allow for 3rd fractional number, we can only enter 1.33 on scaling area :)
Facts:
- 4K image is vanilla with no scaling or sharpening applied.
- FSR and FSR Lite tests are done at RCAS 0.2 sharpness based on AMD's recommendations.
- For 1.2x scaling, in-game CAS value was 0.250
- For 1.333x scaling, in-game CAS value was 0.380
- in-game CAS was added with Reshade as it is more up-to-date and easier to control in photo mode. The reason for adding a second sharpening is to compensate for the lack of LOD bias control like "native" FSR does.
- Benchmarks were repeated 3 times for each condition.
- Both games have resizeable windows when in windowed mode. So, "Force Resize" option is enabled and the scaling value is manually entered.
- RTSS and Process Lasso were used to prioritize and FPS cap both the game and Lossless Scaling app.
Here are the numbers:
Far Cry 5 benchmarks:
1) 4K: 50 FPS
2) 1800p with only CAS: 69
3) 1800p with FSR Lite: 60
4) 1800p with FSR: 58
5) 1620p with only CAS: 83
6) 1620p with FSR Lite: 72
7) 1620p with FSR: 67
Here, 1800p with CAS + regular scaling is equal to 1620p with FSR in terms of performance. But surprisingly, FSR has better picture quality, even though it is at lower resolution. And 1620p with FSR is almost equal to 4K in static images. In motion, there may be slightly more aliasing but that's really not noticeable at high resolutions like these. FPS increase, on the other hand, is much more noticeable and welcome.
Let's look at Horizon Zero Dawn:
1) 4K: 39 FPS
2) 1800p with only CAS: 52
3) 1800p with FSR Lite: 47
4) 1800p with FSR: 45
5) 1620p with only CAS: 62
6) 1620p with FSR Lite: 54
7) 1620p with FSR: 52
Interesting! Here, the performance of 1800p with regular scaling is equal to 1620p with FSR, again. How about picture quality? This time you can see it, too. In Far Cry 5 screenshot app ShareX had problems but I managed to take photos in HZD. Here is the comparison between 1800p bi-linearly upscaled to 4K and 1620p with FSR upscaling.
1800p CAS vs 1620p FSR [imgsli.com]
I told you, FSR does it much better. Even though HZD has this infamous developer trick which reduces LOD texture quality when the resolution is lowered, just like Metro Exodus..
(the sharpening may not be to everyone's taste but I overdo if for the sake of motion)
The other comparison pictures are below, at the end of the post.
Here are my findings and advises:
AMD gives you 4 options, in our case that may be too much as we can't embed Lossless Scaling into our games. I recommend:
1.333x scaling for best quality
1.5x scaling for more performance
1.667x scaling for absolute needs.
2x scaling is a "NO". FSR does a couple of shader passes and then leaves the scaling part to Bilinear filter and 2x is a hard limit for Bilinear filtering. Your HUD and in-game elements wouldn't look good enough to justify using a resolution as low as that. (Maybe if you have 5K or 8K screen)
Recommendations:
1) For 1.333x scaling: use FSR Lite, you can see in the pictures there is almost no difference (I couldn't find any). Also add 0.380 CAS sharpening with either Radeon drivers or Reshade. Adjust Lossless Scaling sharpening if you need to.
2) For 1.5x use FSR. Also add 0.580 CAS sharpening with either Radeon drivers or Reshade. Adjust Lossless Scaling sharpening if you need to.
3) For 1.667x scaling, use FSR and 0.720 CAS sharpening with Reshade and add FXAA before. You may need it at those resolutions.
4) If your resolution is going well below 1920x1080 for modern games, try to lower other settings before resolution. Actually, before trying every FSR step, first try to trim "ultra"s and then "very high"s and then "high"s... You get the point. There is no meaning of playing ultra at 1280x720.
And here are the pictures after -probably- my longest post:
4K vanilla vs 1800p with only CAS [imgsli.com]
4K vanilla vs 1620p FSR [imgsli.com]
1800p FSR vs Bilinear Scaling[imgsli.com]
1620p FSR vs FSR Lite[imgsli.com]
1800p CAS vs 1620p FSR - equal performance point[imgsli.com]
Ok, it's fine now. I fixed it somehow. Now my issue is I have low GPU usage. It seems that FSR is only using 38% to 40% of my GPU and I have a 3070 ti. It's core clock is at 780mhz and memory clock at 4500mhz. It goes back to normal if I turn off FSR. I can't seem to make it boost any further. This is upscaling from lower resolution to 1080p btw.
You mean your game is using 40%? This app should use less.
Under-utilization of the GPU means that this is not necessary. Disable vsync or increase the resolution.
I chose that resolution for many of the retro Final Fantasy games because it gives an opened-up ultra-widescreen presentation while retaining the crispness of 720p (without any blurry UIs or pixel shimmering in 1080p). I just turn on FSR so that it would still look good and uniform in my 1080p monitor.
I play the games letterboxed in a 1920x1080p monitor and the images should scale to 1920x824p, but they are instead cropped by two pixels at the bottom and I think this is a Lossless Scaling quirk.
I hope this can be fixed and that this can properly scale to 1680x720p with no lost pixels and a properly scaled and balanced letterboxed presentation.
Just hitting Alt+Enter in Unity just doesn't cut it because it has no FSR enhancements and the 3D Remakes don't have this since they are not Unity ports.
Here are my comparisons:
https://imgur.com/a/VuCbniI
I am also playing the old versions of Final Fantasy V and VI with FSR in resolutions of 640x384 and 480x320 respectively as well as Final Fantasy IV: The After Years in 1280x720p. They all scale pretty well to 1080p all things considered. I am just having trouble playing 1680x720p CinemaScope games.
I mean my GPU is only using 40% GPU power. It won't go above that. I have Vsync disabled and wanted to target 165 fps on games.
Hi, it might be an LS quirk when capturing a game or some calculations, but I'm pretty sure it doesn't break the aspect ratio of the game and the integrity of the picture, but just cuts it off a bit.
Why are you using FSR for these games instead of integer scaling?
My FPS before was around 120 to 150 fps at Horizon Zero Dawn my GPU core clock maxes out at 1920 mhz. Turned on FSR, then I get around 60 - 71 FPS and core clocks maxes out at 780 mhz.