| 22 Oct 2025 |
| raphi-debug joined the room. | 10:06:25 |
Atemu | In reply to @elvishjerricco:matrix.org wonder why that is. DP 1.4 can do 4K@120Hz without DSC, can't it? only with 8bpp | 11:41:25 |
Atemu | 10bpp is 97Hz max. | 11:41:57 |
ElvishJerricco | Is the bits per pixel the same thing as the bit depth of like SDR vs HDR stuff? i.e. Does that just mean the 97Hz limit is for HDR? | 11:45:36 |
ElvishJerricco | * Atemu: Is the bits per pixel the same thing as the bit depth of like SDR vs HDR stuff? i.e. Does that just mean the 97Hz limit is for HDR? | 11:45:49 |
K900 | Ish | 11:47:07 |
K900 | 10bpp is required for HDR | 11:47:12 |
K900 | But you can do SDR 10bpp as well | 11:47:19 |
ElvishJerricco | huh, well even at "89.98 Hz", /sys/kernel/debug/dri/0000:03:00.0/DP-1/dsc_clock_en still says it's using DSC, 10bit probably isn't the reason in this case? | 11:51:16 |
K900 | It could be using DSC anyway just to run at a lower link rate | 11:51:40 |
K900 | Because higher link rates are more resource intensive to run | 11:51:50 |
ElvishJerricco | yea, and I mean it's not like it matters. IIUC almost no one can actually tell the difference between DSC on or off | 11:52:42 |
K900 | Yeah it's designed to be "perceptually lossless" | 11:54:07 |
K900 | Which is a very interesting choice of words | 11:54:12 |
ElvishJerricco | lol | 11:54:21 |
K900 | But generally it delivers | 11:54:28 |
ElvishJerricco | there's been a few times where I saw some weird coloring that looked like compression artifacts, but the anomaly remained when I dropped to 60Hz so I'm pretty sure that means it's the software that decided to make bad colors, not DSC | 11:56:36 |
| 24 Oct 2025 |
| seapat | 20:35:18 |
| 25 Oct 2025 |
Makuru | Is it possible to force games that use opengl to upscale with AMD FSR 4? | 15:27:32 |
K900 | No | 15:34:33 |
K900 | The game needs to already support some sort of temporal upscaling solution to use FSR4 | 15:34:55 |
K900 | Because FSR4 is itself temporal and requires motion data from the game | 15:35:09 |
Makuru | Ahh, thanks. | 16:39:28 |
| 26 Oct 2025 |
| Stephan joined the room. | 09:44:22 |
| Sir_Morton joined the room. | 15:33:45 |
davidak | Makuru: the windows program Lossless Scaling does not need motion data from the game engine and just upscales the output image. there is also an open source clone that works on linux: https://github.com/PancakeTAS/lsfg-vk
here is a LTT video that demonstrates the difference in the result with DLSS and Lossless Scaling. The motion data helps a bit to reduce artifacts, but it's not that noticeable https://www.youtube.com/watch?v=lCKbt8hBYAo
| 19:58:44 |
davidak | * Makuru: the windows program Lossless Scaling does not need motion data from the game engine and just upscales the output image. there is also an open source clone that works on linux: https://github.com/PancakeTAS/lsfg-vk
there is even a nix package: https://search.nixos.org/packages?channel=unstable&query=lsfg
here is a LTT video that demonstrates the difference in the result with DLSS and Lossless Scaling. The motion data helps a bit to reduce artifacts, but it's not that noticeable https://www.youtube.com/watch?v=lCKbt8hBYAo
| 20:00:09 |
davidak | * Makuru: the windows program Lossless Scaling does not need motion data from the game engine and just upscales the output image using their own proprietary model. there is an open source tool that can use this model that works on linux: https://github.com/PancakeTAS/lsfg-vk
there is a nix package: https://search.nixos.org/packages?channel=unstable&query=lsfg
here is a LTT video that demonstrates the difference in the result with DLSS and Lossless Scaling. The motion data helps a bit to reduce artifacts, but it's not that noticeable https://www.youtube.com/watch?v=lCKbt8hBYAo
| 20:03:19 |
| 27 Oct 2025 |
woobilicious | I think technically, FSR and DLSS have 100% accurate motion data since they're calculated from the movement of objects in the GPU pipeline, while LS reconstructs motion data the same way a video codec or motion upscaler works on a 120hz TV. The Artifacts you see in LS look very similar to those found in other temporal upscaling algos | 03:05:04 |
woobilicious | Anyone know if heroic end up getting removed from nixpkgs, or did maintenance get taken over? I can't seem to find the PR | 03:11:29 |