[Feature request]: Low latency decoder for mediatek chips #82
Comments
|
I think that's hardware independent, maybe newer MTK chips like the 9000 series have this decoder, but older ones might not. |
It's likely we should ask to older mediatek chip owned their decoder list, but at least it could allow the recent mtk chips to play in good conditions. Parsec allows you to have lowlatency directly on all games even using the C2.mtk.avc or hevc.decoder and not the lowlatency one. On moonlight it's more complicated. You have to start the game before the stream, and above all not go through the desktop, otherwise the decode time remains high. What's more, on xbox games I manage to get down to 4 ms, whereas that's not the case on steam games. If Parsec can do it , i'm sure moonlight can also close this bug of 2 yo :). |
|
Parsec's report differs from Moonlight's, they're not actually the same thing. My own tests show that Parsec has worse performance than Moonlight, and the rest are server side things. You can try the latest pre-release and see if Warp Drive mode can improve anything. |
|
Working well for me with prerelease version + warp mode on Galaxytab S10 ultra (Dimensity 9300+) . Without the desktop bug and steam bug (tested marvel rivals on steam and age on xbox). Decoding time a little higher than parsec but you're already mentionned that report differ. Lowlatency decoder is used. |
|
Hmmmm you mean Artemis can already use the MTK low latency decoder and also got rid of other issues? I didn't think Warp Drive mode can solve so many things... |
|
Artemis can already use the lowlatency decoder without the warp mode. But I had a weird problem where to get the 4-5ms you had to :
So all that weird stuff is solved and it's why i tested with marvel rivals on steam because i can't have the low latency before, even when using the good decoder. I don't know how Artemis manages to ask the tablet which decoder to use and switch on the lowlatency one. For webrtc streams (Xcloud / GFN) Chromium browsers go directly to c2.mtk.hevc.decoder (chrome / edge), same for the native geforce now application... With bad results, of course. I'd have to look through the logcat |
|
"Also be aware that Artemis/Moonlight's status overlay doesn't show latency correctly, trust your feeling." Do you think this issue can be fixed to have a real benchmark behind the different options ? warp drive... |
|
Even with the false value it still shows a difference on my devices. Currently there's no good method to benchmark latency, but you can use the frame skipping test from TestUFO, stream by mirroring your physical display and use a third device to do high speed recording to compare how the white square advances on both devices. |
|
What I find strange is that if I limit Artemis to 30 FPS I can't get below 16 MS of decoding time. Whereas if I set it to 120 fps, I'm under 5 ms. It should be the other way round, because the more frames there are to decode, the more it loads the system? I see no difference between warp mode and warp 2. If you want to see how the lowlatency decoder is choosed on my S10 : |
|
Actually, no, lower framerate does lower system load when bitrate is also lower. The delay is basically frame queue and sync delay which cannot be lowered if streamed with lower fps. |
|
OK, so if I reach less than 5 ms with a mediatek with 120fps, the problem seems to have been solved? Nothing to do with moonlight but I'm trying to understand how it works and why it doesn't work in web-rtc with the browsers on the services... And why does artemis manage to select the lowlatency decoder when browsers go straight to the classic c2.mtk with the consequences that we know. I'm not sure that they use mediacodecinfo when the applications do... |
|
I think it's sort of resolved or workarounded. Warp mode isn't something super fancy but a hack that really works... |
|
Except that for those who don't have or can't switch to the low-latency decoder, warp mode or not, they can't get down to the correct decode time. |
|
Yeah, it's still mostly an issue from MTK rather than applications. |
Is your feature request related to a problem? Please describe.
Hello! Is it possible to use a low-latency decoder for MediaTek chipsets in moonlight? According to this post, the low-latency feature is available moonlight-stream#1241 (comment)
Describe the solution you'd like
To have access to an experimental low-latency decoding feature for MediaTek chipsets
Describe alternatives you've considered
Without an alternative
Screenshots
Not applicable
The text was updated successfully, but these errors were encountered: