4 ms·
Latency always goes up with technology. Think of streaming or video on demand services where its now up to five seconds to change.
by VLM 5y ago
Latency always goes up with technology.
Think of streaming or video on demand services where its now up to five seconds to change.
- agumonkey 5y agoIt does happen on computing too. Even on very recent machines, very recent OS kernels you can notice the layers. Very odd to see.
- hulitu 5y agoBecause the developers are lazy. Why do we need 20 layers of libraries to draw a rectangle ? Give a developer a library and he will use it.
- bitwize 5y agoGive the developer no library and they will find a way to slam the bits into VRAM. That said, the friction among modern devs to go from zero to cross-platform app is much lower with Electron than it is any other framework. So learn to love Electron because it is here to stay.
- dreamcompiler 5y ago> Latency always goes up with technology. It's certainly possible to build non-causal signal processors for broadcast TV by using multiple tuners that buffer channel N+1 while you're watching channel N. You could do it for streaming services too by using statistics (or--dare I say it--AI) to predict what movie the user is most likely to sample next. It might make the decoder cost an extra $2, and that's probably why nobody does it.
- jandrese 5y agoSeems like you'd also need a second tuner which would increase the cost a bit more. Ultimately the problem is that with old analog signals you just needed to wait for the next V-Blank (4 hundreds of a second at the most) and you were good to go. With digital signals you need to wait for the I frame which can take dozens or hundreds of frames. And this assumes you have a good solid signal and can decode everything perfectly, if the set is trying to piece together a fragmentary signal it's even harder.
- fomine3 5y agoPossibly there are some trick appliable to improve channel switching performance (like include low-res I-like frame frequently?), but no one have incentive to improve zapping performance.