DDraceNetwork - general

matricks

I added some context to my old message

that's very kind

ngl game kinda paid to win

NPC with one dialogue option

1

PeepDaMeep

ngl game kinda paid to win

05:47

where to send money for r1 multeasymap?

q all, i need bind for hammer fly, who can give it to me pls?

@Cipy29 ^

3

Xrxlil

q all, i need bind for hammer fly, who can give it to me pls?

?

@Xrxlil https://forum.ddnet.org/viewtopic.php?t=2537

.

@ano hello can u help me

14:34

anyone

14:34

help me

https://dontasktoask.com

Don't ask to ask, just ask

the ? is not common to learn in the internet, just see all the facebook posts

?

1

https://dontasktoask.com/ should be a rule tbh

2

yes

@kefa

17:15

blockers getting a 10 min ban isn't gonna help the community at all

17:16

ban that mf for an hour from all of the ddrace servers, make him think about what he did

Alexander

ban that mf for an hour from all of the ddrace servers, make him think about what he did

alexander for president

i hope not justatest

@Nill Kiggers you should write your complains to an admin or in #reports

Quick Q: The fps counter in the game.. is that rendered frames or game loops on the client?

rendered frames and game loops are the same afaik

O.o I decoupled that in teeworlds from what I remember

hence, vsync makes you move less precisely

to decrease the latency

there's definetely input latency when i use vsync

18:29

@Jupstar ✪ any opinion on that?

they are decoupled (edited)

18:29

gfx_asyncrenderold 0 will sync them

so the ingame fps counter shows what?

the rendered frames

sweet! thanks

18:31

Saw that someone went overboard and did a Vulkan backend as well

overboard? almost everyones game runs faster because of vulkan kek

yeah, the opengl to metal wrapper on macos performed so poorly for the m1 chip. but sadly the vulkan to metal wrapper is even worse xDD

18:32

i wanted to know how good deen's m1 compares to my desktop pc

@Jupstar ✪ Oh noes!

Jupstar ✪

Click to see attachment 🖼️

^ fps record on my setup ryzen 3800x, most hud etc off, rx 6900 xt, linux wayland + lto'd drivers etc.

Yeah I was hitting 8k without problems

nice

wonder what the bottleneck is in general when it comes to those speeds

yeah good question. i assume RAM access is still the biggest bottleneck. But it's most probably some CPU side thing, the GPU only runs at around 90% but thats also why i wanted to know what M1 could achieve

You shouldn't have to do that much on the CPU in general... I wonder if it quicker to do a tilebased rendering on the GPU through compute in order to save vram bandwidth... stuff that have been keeping me up at night

18:39

Bascilly, do rendering old school NES styled

maybe yes, but maybe the GPUs are so highly optimized with rasterizing blending etc. that you cannot beat it with a custom compute shader which is kinda more general approach but generally since CPU is pretty idle, calculating as less as possible on the GPU is still the best for ddnet. @Patiga wrote a different renderer that kinda does what you want i think. it calculates what is rendered per pixel. so the fragment shader is more heavily used it's especially great for zooming out, but at normal zoom the overhead is bigger then normal vertices -> rasterize -> fragment approach

Oh! Cool! So someone went down that path. I've always wondered

18:47

@Jupstar ✪ Know if anyone tried skipping textures all together and just render vector graphics? (edited)

I am not sure what exactly you mean. vector graphics as SVGs?

Yeah, instead of offline render SVGs to bitmaps, just render SVGs in runtime

18:49

or process them into a more efficent vector format

i dont think somebody did that. we dont support SVGs at all. PNGs are simply so much faster to load. But i guess modern hardware could handle it at an ok speed. integrating it into multiple backends is hard tho i guess, except there is a good library that supports OpenGL, Vulkan etc.

I don't think you wanna go full on svg in the end, just convert it into something that is fast to render but still vectors, would mean you would support infinite resolutions

i think it‘s quite some effort to render vector graphics directly on gpu. but maps applications still do it

Jupstar ✪

they are decoupled (edited)

then why less fps = more input latency?

matricks

I don't think you wanna go full on svg in the end, just convert it into something that is fast to render but still vectors, would mean you would support infinite resolutions

yeah. i'd say here again with the ryzen CPU revolution it would be accaptable, since u can render multiple SVGs at once. but e.g. https://github.com/ddnet/ddnet-data-svg our data directory as SVGs takes a few seconds to render with inkscape even on my desktop

GitHub - ddnet/ddnet-data-svg

Contribute to ddnet/ddnet-data-svg development by creating an account on GitHub.

Chairn

then why less fps = more input latency?

input latency = u click -> it gets rendered

18:54

so all kind of latency is kinda input latency to the eye

but it's also less precise

@Jupstar ✪ I mean, look at modern font rendering, you can push vector graphics very quickly on GPUs with some cleaver preprocessing on the CPU

Chairn

then why less fps = more input latency?

"In video games the term is often used to describe any latency between an input and either the game or the display reacting to that input, despite the fact that this may emerge from a combination of input and output lag." thats what wikipedia says xd

if both were independent, there would be no change depending on frame number

matricks

@Jupstar ✪ I mean, look at modern font rendering, you can push vector graphics very quickly on GPUs with some cleaver preprocessing on the CPU

true, but tbf a single character is almost nothing vs a 1024x1024 or more texture

Chairn

if both were independent, there would be no change depending on frame number

what do u mean with less precise

you cant do small moves with vsync for instance

when i decrease fps the only thing i notice is, that i get more tearing

or it's way harder for some reason

vsync on windows waits

18:56

its like sleep command

so, each key press is like 1/60 second duration?

18:57

i mean rounded up to next 1/60 second duration

no but i think that makes input feel very unreliable

18:58

i'd need your config to understand what exactly your problem is. I think we do wait for the GPU as soon as the client "thinks" it has to swap the frames