programming

[true hemlock]

unemployment

[midnight sigil]

[midnight sigil]

he's a cybersec guy

[nocturne olive]

Peak free data

[rough bloom]

terms of service usually aren't even that long
they're just really, really boring

[nocturne olive]

Sadly vocal synthesizers are not so easy

[midnight sigil]

and a part time streamer

[midnight sigil]

2 hours vod... surely

he took no break as well

no water nothing

just straight fucking reading the whole shit

[true hemlock]

istg i can start the "which comes first, language model or corporate language" philosophy

[rough bloom]

if you slowly read it out loud, yeah

[true hemlock]

because TOS really felt like its whole AI

😭

[midnight sigil]

they are essentially spitted out and modified by tons of lawers probably

[olive sable]

Its fucked. Broken xbox

[nocturne olive]

Maybe there just are so many TOS' that LLMs get a lot of their training from that

[midnight sigil]

to get those juicy grey area yk

[true hemlock]

NOOOO

[nocturne olive]

Check what the error code means

[midnight sigil]

so it feels like AI

[rough bloom]

same 3 lawyers copy-pasting from each other because those exact combinations of words are proven to work well in court

[nocturne olive]

Hard drive

[true hemlock]

there are so many shit research papers that feels like ai too

[midnight sigil]

hey that's exactly what language models do

[true hemlock]

you know what i mean it feels like if you had ever stumbled upon those ai generated "scientific" yt shorts or something

[nocturne olive]

There must be somewhere LLMs get their garbage data from

[midnight sigil]

we found the prove

[olive sable]

i have a replacement hard drive if its sata, but iirc it uses an older standart

[nocturne olive]

IDE

[olive sable]

ye

o7

[nocturne olive]

Also it needs to be specifically formatted for the XBox beforehand

[rough bloom]

it would be funny if ToS were actually a significant portion of LLM training data because many websites have one
but surely they deduplicate or something

[midnight sigil]

Random ahh clips + "Physics" + Einstein picture at bottom = 122M views

[true hemlock]

im tempted to train one with shit tons of tos rn

[midnight sigil]

[true hemlock]

most of it are even just ai generated images being edited to move around, zoom or some shit

the einstein pictures are usually on edits

also slop lol

[midnight sigil]

it will be generating language

and random company names

[rough bloom]

they're slowly moving on from images to videos because of Veo 3 now

but clips are still short and audio is... questionable

[midnight sigil]

phonk phonk phonk

[rough bloom]

I've only really seen the ones that popped up together with Veo 3
I'm not a huge YT shorts enjoyer so I don't see them much

[olive sable]

original xbox's have the harddrive locked to the mobo, and you need to mod it to replace it

[true hemlock]

if only shorts are high quality stuff like from this channel https://youtube.com/@htxstudio?si=7Q7NDuPSwTFc-n1F

[midnight sigil]

I hate when people using AI to visualize bible scenes, it feels like indignity to god

[true hemlock]

like each shorts straight up always impressed me in different ways

[rough bloom]

software mod, surely

[midnight sigil]

hell nah that one chinese guy

he claimed open-source softwares to be "written by him" btw

[true hemlock]

i'd rather the old animations people make, those are good af

[olive sable]

ye, softmod exists but i stil dont feel like modding it

[true hemlock]

the ai generated shit are slop

[midnight sigil]

true

[midnight sigil]

it's not really a big deal, he apologized after refusing to admit and taking down his video. They do have their original designs on machine structure and stuff 👍

[rough bloom]

I don't like hardmods because they're annoying to do and are fairly permanent but softmods are usually cool

[midnight sigil]

it's more like he's trying to use his co-workers work, to make him feel like a very smart person on media platforms

and he fucked up this time

even deleting the licensing headers

[true hemlock]

smh

[midnight sigil]

[true hemlock]

but yeah most of the youtube are slops now

and i don't even wanna talk about the social medias boomer uses

those are literally full of ai generated shit

[rough bloom]

I can see some level of deniability for heavily relying on open-source software without mentioning it, but explicitly deleting license headers?

[midnight sigil]

Mr. Beast claiming no cgi and says they use cgi to make the scene feel better afterwards

[midnight sigil]

AND showed the code while claiming it was written by him

[true hemlock]

the whole mrbeast situations last time has been nothingburger for me ngl

[midnight sigil]

ofc there're curious ppl and found the original repo

[rough bloom]

the GitHub search feature in question

[true hemlock]

istg its mostly just either small mistakes or mistakes that aren't even necessarily bad/impactful, but people always made it a huge deal

[rough bloom]

it's actually really useful
you can just paste in some unique-looking name and you find exactly where it's from

[midnight sigil]

*the Chinese programming scene exploded and talked about open-sourcing spirits and ethicals for a week

cuz there're just too many "stealing open-source and claiming as theirs" happening

like, people stole blender

visual studio code

everything

not just stealing, they rebranded it, redistributing with commercial intentions, and can't even replace strings properly

shits are insane

[trim valve]

I should really figure out a use for these

[desert wave]

I would say making viral youtube music videos, but I think you missed the boat by a decade

[midnight sigil]

DJ bred

owobred - Soda pops (Official Mix) coming

[hollow spruce]

Maths is too tricky one

[rigid snow]

as someone who has one, the use is to sell them

you can’t even really use it as a drum machine, the pads are too hard

[olive sable]

8x8 pixel display innit?

[rigid snow]

basically the only thing they’re good for is exactly what they were made for, and that’s live performance (switching loops)

[rigid snow]

they’re green and red only

[olive sable]

aaah

thats rg

you just need the b

[rigid snow]

also the refresh rate fucking sucks

i mean the midi bandwidth it takes in

[olive sable]

this has blue pixels on the box tho?

[rigid snow]

it might be mk ii or iii or some bullshit sure, those are rgb

[olive sable]

if it does more than 1hz its good enough to dispally 8x8 pixel art

[rigid snow]

it’s not going to refresh at once

idk how to explain

basically how you set colors is you play midi notes of different velocities, the note is the pad and the velocity is the color

midi is a message queue

[true hemlock]

bad apple and doom can run on it

[dapper kernel]

Yoo 10 years ago the app was popular

[trim valve]

the big one is kinda rgb

[rigid snow]

i figured

[trim valve]

I am mildly tempted to use it as a camera switcher or something

but that would require a little bit of custom software

[rigid snow]

midi is ez

[dapper kernel]

That would be a DYI stream deck

[rigid snow]

also i used it for scene switching in obs with an app, like not even custom

just downloaded it and it worked

[trim valve]

fair

idk if anyone's done it for black magic stuff

[hollow spruce]

HELP ME MATHS

[olive sable]

we love black magic

[olive sable]

we can try i guess

[midnight sigil]

awawa

[midnight sigil]

plz send your math formatted under APA, with LaTeX for equations

[hollow spruce]

6. If A × B = {(a, x),(a , y), (b, x), (b, y)}. Find A and B.

7. A = {a, b}, B = {x, y} find out b and y

I'm confused about the question

[olive sable]

uhhh

what

[midnight sigil]

that's APA but not LaTeX

is that a matrix

[olive sable]

is that set theory?

[midnight sigil]

or smth

yea I can't understand this

plz format to LaTeX

[olive sable]

i think its set theory

[midnight sigil]

that might be a crossproduct

you never know

he used the crossproduct symbol

[olive sable]

which symbol is the cross product symbol?

[midnight sigil]

x

×

[olive sable]

thats just multiplication i think

[midnight sigil]

but he used matrix or vector wrapper thingy

[olive sable]

thats set theory

the set witht ellements

[midnight sigil]

but what are the (n, p) things

[olive sable]

a set of tuples i guess?

[hollow spruce]

Have u done I'm CONFUSE
Most easy one
Let me ask my chatgpt for this one I wasted the whole 30 minutes

[midnight sigil]

it certainly looks like matrix math

[olive sable]

ye it was set theory

[midnight sigil]

my chatgpt tells my b = b, y = y

[olive sable]

it would be correct, but its also stupid as fuck

[hollow spruce]

MINE WAS
A = {a, b}, B = {x, y} , B € A , Y € B

[burnt aurora]

using € as the "Part Of" symbol is crazy but somewhat genious

[hollow spruce]

That's why i got confused

[olive sable]

ye that just cant be a solution

there is no capital Y

[midnight sigil]

[olive sable]

and B cant be an element of A since A is clearly defined as not having anything of B

[opaque sigil]

this question makes no sense btw, there's no way for you to know what b or y are (unless there's extra info you didn't provide)

[midnight sigil]

lemme ask deepseek r1

[hollow spruce]

Confusing part finds out the b and y

Otherwise and will be
A = {a, b}, B = {x, y}

[midnight sigil]

so b = b, y = y

that's the solution

smh

[opaque sigil]

i mean yeah

[olive sable]

b ∈ A, y ∈ B ???

[opaque sigil]

that too

[hollow spruce]

MIS PRINTED IS WILD

[olive sable]

as far as i can see the 2nd line is just the answer to the first line, and the 2nd line doesnt have an answer

[hollow spruce]

Some questions are wiped out or fade

[opaque sigil]

i'm a big fan of the cartesian product, it allows me to be super lazy with sets

[midnight sigil]

b ∉ B, y ∉ A, b ∈ A, y ∈ B

[hollow spruce]

I will ask teachers to solve it
Personal number will help me

[opaque sigil]

the first 2 don't have to be true smh

[midnight sigil]

they can be intersecting

yk

just to clarify

[opaque sigil]

[midnight sigil]

[rigid snow]

i should just drop an obscure theorem here with a “pls help matj solve” sometime, chances are y’all are autistic enough to actually solve it

[midnight sigil]

unemployed

[olive sable]

proffesionally unemployed

[midnight sigil]

we should secretly make a #neurotic-neurons leaderboard, so we can tell who is the most unemployed person in this server

[olive sable]

oh me 100%

[rigid snow]

you don’t chat in nn

well tbf neither do i anymore

[olive sable]

i thought we meant in #programming chat

[scenic pelican]

There still there

[olive sable]

cuz im online here 15 hours per day if not more

[midnight sigil]

like, sending messages, engaging in conversations, get reacted and replied

we need to make scores

ranking systems

for nn

[olive sable]

i have 43.319 messages on neurocord, of which exactly 40.000 were in #programming . with this message included 40.001

[midnight sigil]

I can't check mine

probably more than 60000

in total

[rigid snow]

what’s your tatsu server rank

[hollow spruce]

Top ?

[olive sable]

you have 64.070 on neurocord with 4338 in programming

[midnight sigil]

I guessed correctly

[nocturne olive]

Rare Xoda spotted
Have you run triangle yet?

[olive sable]

im 68th

[midnight sigil]

wait what you only joined less than a month ago??

there's no way

did you quit and join back

[rigid snow]

can you lurk for a bit to drop down one

[rigid snow]

he did

[midnight sigil]

my memory is correct

I definitely know sam before august

[olive sable]

ive been here since decemeber 31 2023, but there have been 2 or 3 times i left for a couple hours

[midnight sigil]

but what for

[olive sable]

i hold myself acountable

[midnight sigil]

[rigid snow]

self timeouting

[hollow spruce]

I'm jealous of u u have very free time and me i need learning guitar in tution

[nocturne olive]

I really need to overclock this CPU using NeuroSynth is pain with CPU rendering being so slow

[midnight sigil]

"I started to play with cheats because I can't stop playing this game" ahh vibe

[olive sable]

???

[burnt aurora]

the hell's a triangle

[rigid snow]

i think they’re referring to osu players hacking to ban themselves

to stop their addiction

[nocturne olive]

Triangle, the best benchmark ever

[midnight sigil]

yea

and cs pro scenes

[olive sable]

osu players are wild man

[rigid snow]

dw i was confused too

and i play the game

[midnight sigil]

I think that was more iconic

[rigid snow]

i’m not aware of that

[midnight sigil]

VAC banned live on a LAN game against s1mple

[rigid snow]

emilio

is the guy

[rigid snow]

but he didn’t ban himself purposely

[olive sable]

makes me think of this
https://www.youtube.com/shorts/Dz7ttL_lwfk

[tender river]

someone keeps trying to reset my email

[hollow spruce]

I need to repair python files
Python313.dll is missing

[tender river]

steam keeps opening this window

[olive sable]

i guess change your steam password with force-logout?

[tender river]

theres no active sessions that arent mine

[midnight sigil]

[olive sable]

huh

[tender river]

its being done without logging in

i mean it could be a steam bug too

(also if they had managed to log in it would be mildly concerning to say the least as i use a password manager and 2fa)

[hollow spruce]

C:\Users\new pc\AppData\Local\Package Cache{822f14aa-d580-47b5-acec-50522824aca3}\python-3.13.6-amd64.exe

[tender river]

i like how they use a phone icon for pc

[hollow spruce]

Finding wheels is hard

[midnight sigil]

[rigid snow]

“all your accounts get banned if 1 gets affected” bro really used one phone number for all of his accounts, ripbozo

[hollow spruce]

Dll is gone

[midnight sigil]

there's definitely a rumor quoting he cheated to ban himself because of addiction

but it's a rumor

I can't find an article for it

[rigid snow]

he was a pro, t2 sure but still

[true hemlock]

you need a threadripper

[rigid snow]

what addiction, it’s his job…

[nocturne olive]

Well how do you expect me to get a Threadripper?

[midnight sigil]

I hope I was playing cs source when I'm 5 years old

instead of playing chinese minecraft rip off

[rigid snow]

i was and my peak is gn4 in go, so you probably didn’t miss much

[midnight sigil]

[hollow spruce]

Uninstall.exe is gone

[midnight sigil]

I mean, minecraft led me into programming

that's something

[rough bloom]

more like he needs to learn how to make it use the GPU
training already does so inference surely can too
-# otherwise get Epyc Naples, very cheap

[rigid snow]

to be fair i’m eu, idk how asia is in terms of skill tho

[midnight sigil]

they only know go mid with PP and die

higher elo is same as eu

sweaty sweaty

and more toxicity

gold novas are as bad as silvers in cs2

I have no idea how that works now

[rare bramble]

I got to LE in cs:go, quit playing because it was too toxic and stressful

[midnight sigil]

I'm DMG in wingman rn

MGE on mirage and MG1-2 on main map pools

[rigid snow]

wingman is free

everyone and their mom is global there

[midnight sigil]

if no cheaters

[midnight sigil]

I think it's about 200 matches and you'll get there

I have 81 wins with 55% winrate

[hollow spruce]

Yo gamer

[true hemlock]

guys

[hollow spruce]

Yo

[midnight sigil]

[true hemlock]

elvyn infected a server with the 🔺

[midnight sigil]

more competitents?

what is that word

competitors I mean

[true hemlock]

YEP

[hollow spruce]

Last time ddl was gone now no mistake

[midnight sigil]

hurray

[midnight sigil]

I can't stay at top 3 anymore

[rough bloom]

on a scale of 0 (doesn't own a computer) to 10 (Neurocord) how crazy are they

[true hemlock]

chances are im losing #1 soon

[true hemlock]

uuh

[midnight sigil]

[true hemlock]

🥀/10

[midnight sigil]

8GHz coming

getting 10s y-cruncher and 200k windows fps

get liquid helium just to hit triangles

[trim valve]

on one hand I really don't have a good reason to turn on PBO to test my fps

on the other hand, it would be funny

[midnight sigil]

what is 🥀 in decimal

[hollow spruce]

What ure idea

[midnight sigil]

and what's their progress now

[burnt aurora]

can anyone here help mr Superbox with learning easy kotlin Ui (or TUi)?

[olive sable]

we have probably already been defeated

[burnt aurora]

[stark needle]

@true hemlock the triangle thing runs at 13.7k-14.2k

[burnt aurora]

-# clown

[stark needle]

on my 3090

[true hemlock]

what os?

[burnt aurora]

-# so mean

[stark needle]

windows

[olive sable]

oh

ye thats expected for windows

on the old version with semaphores at least

[stark needle]

i'm on 11 pro for workstations 23h2

[olive sable]

the new one without semaphores its either 30K or 0 fps depending on vibes

software rendering seems to be prefered for that

[rigid snow]

what’s with the wilted roses 🥀🫩🫩

[olive sable]

im wilting in real time

[midnight sigil]

kotlin

[stark needle]

[burnt aurora]

is it actually that bad?

[rigid snow]

why is discord so slow to add emojis smh

[burnt aurora]

can't be worse than Processing 4

[olive sable]

Idk never used kotlin

But ui just sucks in general

[midnight sigil]

use Java

[rigid snow]

unless it's react

[olive sable]

Dont worry about it, bad habits just catching up to me

[burnt aurora]

eh, been doing godot ui stuffs and had no issues so far

[trim valve]

new version runs mildly better

[midnight sigil]

doing everything with Microsoft Fluent UI is more based

[rigid snow]

that's crazy

[olive sable]

Software?

Or igpu?

[trim valve]

dgpu

on windows 😭

50% faster so

[olive sable]

Huh, your dgpu seems to work.

[midnight sigil]

yes it's React-based

[trim valve]

I hope so

[midnight sigil]

[olive sable]

Mine did 0fps cuz the lack of semaphores caused race conditions. It seems like windows slowed you down enough to not have race conditions tho

Thats gotta be the funniest shit ive seen here today

[midnight sigil]

#programming message
try this

[true hemlock]

build.bat to compile (not a must) , triangle.bat to run
this is the Triangle, a benchmark by the resident of NSH #programming. this is for fun only! try to optimize your machine and hit as high FPS as possible!
NOTE:

• most of the libraries are packed, you don't need to install these on your own.
• you might need GCC/MSVC installed if you wanted to compile from source. alternatively just use the precompiled version by clicking on triangle.bat
• running the one click compiler isn't a must, but it often improves performance because you're compiling from source directly on your machine
• feel free to screenshot your result afterwards and we'll add your score to the leaderboard
• windows might flag this shit as . windows is just stupid
• linux version (good luck figuring out how to install. compiling from source in linux is recommended!!!) also you can find current leaderboard here -> https://github.com/samvanmaele/very-fast-but-very-bad-code-vulkan
• #starboard message #programming message
• 🔺
  discord commands:

editor's access:
!triangleadd<linux/windw/macos> name,score,os,cpu,gpu,ram    -  adds an entry to the leaderboard
!triangleupdate<linux/windw/macos> numberonlist,newscore     -  updates an entry score to the leaderboard

admin's access:
!trianglerem<linux/windw/macos> nuumberonlist                -  removes an entry to the leaderboard
!trianglepush                                                -  push the updated leaderboard to sam's repository in github

everyone can access:
!triangle<linux/windw/macos>                                 -  show leaderboard

[nocturne olive]

Probably because it's a fully game engine which is designed for it but also very heavy for just basic GUI

[rough bloom]

it works on Linux Mesa AMD iGPU too, it's an NVIDIA-specific thing I think

[olive sable]

Hmmm

[true hemlock]

updated this

[rigid snow]

the design system is not and cannot be react based (that's just not what a design system is), it's just that a couple of implementations are react

[burnt aurora]

does it linux

[olive sable]

So nvidias drivers have the race conditions?

[midnight sigil]

[olive sable]

I should try the windows version

[midnight sigil]

I see

[nocturne olive]

For Linux compile the GitHub repo

[olive sable]

Its made for linux

[midnight sigil]

change log?

[true hemlock]

there

[rigid snow]

sam why no windows and macos in makefile in upstream

"dowbload this random ass archive and run a bat file" lol

[olive sable]

Ekness you're questionmarking me but you just dont get it. You miss vital #programming triangle arc lore

[midnight sigil]

you can run the makefile on windows

just some modifies

[uneven pulsar]

is this some inside joke

i hate inside jokes

man i am stupid

[olive sable]

Its nt really an inside joke

Its just the triangle innit

Runs at a casual 41160 fps

[uneven pulsar]

its going to become an inside joke or whatever thing

i am stupid

idk

[rigid snow]

definitely not on macos and i mean broader support, iirc the code uses linux builtins for thread priority bs so it won't build on anything but linux anyways

[olive sable]

Its just a really fast triangle, idk what to tell you

We meticilously optemized it

[rough bloom]

it has ifdefs for Windows support
MacOS has pthread support, surely

[uneven pulsar]

i been getting 'bool' object is not subscriptable

def calc(d_calc:int) -> int:
    d_calc=int()
    for x in range(1,4):
        try:
            cons_act={1:12345,2:2344,3:2144}.get(d_calc,'p_kw').isnumeric()
            cons_pre={1:2112,2:1600,3:1221}.get(d_calc,'left').isnumeric()
        except ValueError as e:
            print(f"{e}")
        cons_kw=cons_act[x] - cons_pre[x]
        
        print('Left num',x,'KW consumed=',cons_kw)
        try:
            sum_kw=lambda:d_calc['p_kw']*cons_kw+d_calc['left']
            
        except ValueError as e:
            raise(f"NONE EXISTING NONE TYPE VALUE {e}")
        tva=sum_kw()
        sum_net=sum_kw+tva
        print('Left value of num',x,'Net paying in $',round(sum_net,2))
        print(24*'','Net to pay in dollars',round(sum_net*cons_act['t_d'],2))
        print("________________________________________")
        try:
            total=float(input('enter money '))
            left=float(input("the sum of leftovers="))
            price_kw=float(input("the price in KW "))
            d_calc={'t_d':cons_act,'left':left,'p_kw':price_kw}
        except Exception as e:
            print(f"return wait wtf -->{e}<-- look at the error stupid ass")
            
calc(d_calc=12)

[olive sable]

Quack oc'd his cpu to 7.6 GHz or something for 100K fps

[rigid snow]

yeah i haven't looked at the code for a while, it didn't when i checked

[rough bloom]

if you want MacOS support then you can check anywhere the _WIN32 and __linux__ definitions are checked and add threading support for MacOS there

[rough bloom]

ye, it was a branch before that

[midnight sigil]

compiling c++ on mac sounds bad

somehow

feels diabolical

[olive sable]

Its just

#ifdef _WIN32
Thread system 1
#elif __linux__
Thread system 2

[rigid snow]

the inconspicuous clang++ that comes with the stock dev toolchain:

[midnight sigil]

vani did macos and you should be too

[rigid snow]

i have nothing to achieve

it's an m1

[true hemlock]

it hit 7.8 for a tiny burst btw

[rigid snow]

i'll be right at the bottom of the lb

[olive sable]

Bro really going almost double my cpu

[midnight sigil]

I'm sitting for risv and templeos(holyc fork)

[uneven pulsar]

could anyone help

[midnight sigil]

doesn't have to be vulkan, just draw a triangle

[uneven pulsar]

i am just a fuckass idiot

[rigid snow]

that would require someone to have capable riscv hardware

which i would imagine isn't something people just have

[midnight sigil]

I'm sure people do

[midnight sigil]

you'll get smarter dw

[olive sable]

There are instructions somewhere

Not from me, im too sleepy

But somewhere

[uneven pulsar]

could you help me

[rigid snow]

instructions for what

riscv instructions?

[olive sable]

Sure why not, i also dont give out those

Ask someone else

[true hemlock]

discord commands also in here

[rigid snow]

oh come on man, just a little mulhsu?

[true hemlock]

everytime someone wanted to do 🔺 just link em to #programming message

[uneven pulsar]

is this some inside joke

[true hemlock]

if i have the time tomorrow i'll try hackintosh

just to also dominate macos leaderboard

[uneven pulsar]

bye naruto girl duck whatever you're name is

i am stupid as fuck

[true hemlock]

🥀

[uneven pulsar]

idk tf happened

[uneven pulsar]

also is it me or are these...💀

uhh why...yknow i had enough shit

[midnight sigil]

I'm planning to add ASCII support to my AWA5.0 Interpreter

anyways I'm eeping now

gn

[rough bloom]

7 bits per character

inefficient

[true hemlock]

smh

[wise acorn]

[rigid snow]

fake and forged and

[wise acorn]

[wise acorn]

https://github.com/jdecked/twemoji/issues/88

this one is better

[sage crag]

awful keywords

try again

[warped narwhal]

clearly they should just add as an emoji instead

[sage crag]

https://cdn.discordapp.com/attachments/1071784467036913664/1406727268285284442/konorb.gif

[rigid snow]

it's still a proposal

this year maybe

[rigid snow]

what

Sportsball

[tender river]

sportsball

[uneven pulsar]

every football jocker life

[hexed grove]

Bet I'm the only viewer joining from an iPod touch fourth gen 😭

twitch already uses hls

i just cant use the ui

because they have so much damn js

so i had to write this

[rigid snow]

wait i have that

also i had the same battery pushing on the screen thing but it just

went away?

[hexed grove]

did your battery explode /j

[rigid snow]

i mean it barely works

good luck getting more than 40 minutes of battery life

i’ve had the thing since it came out

which is like 2009 right?

oh 2010

wait the white one came a year later so 2011

[hexed grove]

mine has very much unfunctional battery

and

is used for watching twitch exclusively

:3

[rigid snow]

interesting usecase

[uneven pulsar]

i feel normal again

[sage crag]

https://docs.vulkan.org/samples/latest/samples/extensions/shader_debugprintf/README.html

readme.html

[glad path]

my ai has 'prismaticdepths sounds like a mosquito' as the example for messages to send, so whenever I ask it to dm me or say something and don't give it anything else it just says I sound like a mosquito

its quite funny ngl

thanks neuro for the idea

[amber fractal]

I see Quack regained the record for now

[glad path]

oh wait

FINALLY

[amber fractal]

Sam also having troll timing on pushing a change

[glad path]

NO MORE GREEN LEAF TAG

[amber fractal]

Your no longer new

[glad path]

https://cdn.discordapp.com/attachments/1365863973290770442/1389549761090027610/neuro_new_user_see_you_in_the_damn_abyss.gif

now i can use this

[warped narwhal]

you're

[tender river]

backus-neur form

[amber fractal]

[rigid snow]

as long as it works or something

[amber fractal]

Yeah, I hope none of the traces are shorted

[opaque sigil]

[sage crag]

[warped narwhal]

[violet sand]

[rigid snow]

first person brave enough to actually put rust in their name

[violet sand]

Some times I regret my life choices

[bleak zephyr]

i posted the wrong link, i will see myself out

[sage crag]

https://tenor.com/view/neurosama-gif-27390590

@true hemlock i know you mentioned it before but how does elvyn work vaguely? i remember that she's not an llm.

she's also very multi-modal and seems to just do what she wants

[frail mortar]

New at programming Ai. Would it be possible to Ingest memories based on the ingest content? File them in meta.json according to topic, pull internal monologuing from these different meta.json files, combine the context to the internal monolog, and phrase the many answers to 1, shoot that 1 answer to a response?

Science, conversation, history, persona, and so on?

[amber fractal]

Surprisingly doesn't exist, a lot of LLMs use memory fetch

[frail mortar]

cant run more than one parallel?

[amber fractal]

It's more like fetching from a database that has it's own subpar issues

[amber fractal]

(That's how it is implemented actually)

[frail mortar]

and I'm sorry if Calling my Ai "Aeon" seems odd, thats her name....

So there would be no way to catigorize ingest content?

other than what the .json and .faiss does already?

[karmic birch]

!triangle

[amber fractal]

Always something you can look into on your own

[hidden ravine]

🔺

[frail mortar]

I'll share if I do something worth sharing with this idea.

[amber fractal]

I'm still drafting my own attempt at it personally

[frail mortar]

Good luck

[sage crag]

!trianglelinux

[hidden ravine]

Linux Leaderboard:

1. 102748 by Quack on NixOS, i9-13900 (ES Q0L4) 7.6GHz OC unlocked + RTX 4070 Super
2. 70854 by owobred on NixOS, Ryzen 9 9950X + RX 9070 XT
3. 41784 by Sam on CachyOS, Ryzen 9 5950x + RTX 3090
4. 36036 by Iggly on NixOS, i7-1260P + Xe Graphics (integrated)
5. 35600 by superbox on Mint, i7-12700K + RTX 4070 Ti
6. 32100 by Semicolon on Arch, Ryzen 9 5950x + RX 6800 XT
7. 30573 by Iggly on CachyOS, Z1 Extreme + AMD Phoenix
8. 27158 by Shuni on NixOS, Ryzen 9 7950X + AMD Raphael (integrated)
9. 26364 by konii on NixOS, Ryzen 5 7600 + AMD Raphael (integrated)
10. 22809 by Prism on Custom, HP Prime G2 RevD + (Graphing Calculator)
11. 17000 by toast on NixOS, Ryzen 9 5900x + RTX 5070 Ti
12. 13768 by Sqbika on NixOS, Ryzen 5 3600 + RX 9060 XT
13. 13337 by Sqbika on NixOS, i7-9750H + GTX 1650M
14. 13037 by Assist on CachyOS, i3-8100U + UHD 620
15. 9458 by unkomp on Arch, +
16. 672 by Quack on TizenOS, Quad core ARM cortex? (Samsung Smart Fridge) + Mali GPU
17. 0 by Elvyn on NixOS, Ryzen 9 7950X ES + RTX 3080

[sage crag]

excuse me 103000

[amber fractal]

All I can say, not due to hiding secrets but a basic requirement is that you need a layer of misdirection for memories, Like how different ways exist to express the same thing like language.

[frail mortar]

Hippocampus.py

[amber fractal]

There are things I'm not going to say to avoid getting snipped. But might as well say the one truth I've come to realize.

[frail mortar]

are we allowed to share files here?

[amber fractal]

yep

Just after you gain your rights

[frail mortar]

permissions may not be set. I'll work on rights later. Would hate to be banned here.

[violet sand]

I still don't know what this is about

[hollow spruce]

Hmmm

[amber fractal]

We are rendering a triangle as fast as possible, that is the lore

[violet sand]

Ahh

May I join?

[amber fractal]

ye

[violet sand]

What are the ground rules?

[sage crag]

run triangle

make number big

[amber fractal]

.

[sage crag]

you have to render the triangle though

[violet sand]

thank u

[sage crag]

cant be skipping it

i maintain that shrinking the resolution is cheating but

[amber fractal]

Unless you are the graphing calculator, it must be vulkan

[amber fractal]

The issue was for awhile it did nothing until people started doing software rendering

[violet sand]

Ya that was my first thought

[amber fractal]

where the 200x200 and 32x32 had the same fps

actually the latter was worse for me

[amber fractal]

This is because code edits are fair game

[violet sand]

Can I "cache" an image of a triangle in L1 mem. after the first render, if I know the window size didn't change?

[amber fractal]

The draw call is already cached

vulkan is still the hard requirement

[violet sand]

Ahh got it

[amber fractal]

Check in /src/commonstructs.hpp for your dials

If you not Sam on software rendering, frames in flight should be 6-7

Useful for windows testers

[violet sand]

I don't even know what "frames in flight" means

[amber fractal]

How many frames can be in the stage between CPU and GPU

In AAA this is 1 for Competive and 3 otherwise. But we are going so fast that our scale needs to change

[violet sand]

Ya I heard about it in one of the interviews with Nvidia that GamerNexus did once, that its usually 3 lul

[amber fractal]

As a note, a lot of people have a limit of 512

Because of course that dial was turned to the max

I guarantee that whatever effort your making, there has been more done. Just look at the 100k fps numbers, tuning every knob in existence.

[hollow spruce]

AFTER 157 SHORT IM BACK AND IT MY HOLIDAY

[true hemlock]

so normal rendering goes:
CPU process, tell GPU to render, CPU wait while GPU receive the command and renders

with frames in flight you can do:
CPU process, tell GPU to render, CPU process next frame while waiting for GPU to render instead of doing nothing

this just basically keep all the components busy instead of making them all wait for each others

[amber fractal]

On today's episode of Iggly finds a sidequest

I think there might finally be a use for 512 frames in flight

Downside is entirely rewriting more of the render thread

vkQueueSubmit2(deviceManager.graphicsQueue, 1, &submitInfo, VK_NULL_HANDLE);
This funny line, Apparently you can batch submit according to the docs. https://registry.khronos.org/vulkan/specs/latest/man/html/vkQueueSubmit2KHR.html

And we've only been sending one frame at a time

I gave a shot at impl, but yeah I don't know enough c++ to do so

I'll leave it to everyone else to decide if this is legal or will even be faster. @olive sable

-# I'm not benefiting from this, my iGPU is at 90% usage

[true hemlock]

all these just for the 🔺

[amber fractal]

Also moves it from being a CPU challenge if one can save enough lines

[true hemlock]

my scores has been threatened

[amber fractal]

Got to wait for Sam's verdict on batch frame rendering

[true hemlock]

is it multi threaded?

[amber fractal]

Nope, still single. But the benifit would be to call this less, and keep L1 instruction cache by reorganizing how frames are processed a bit. Aka doing each step in a loop.

Mainly in vk_sync.cpp

It's a gamba on if it works as well as one is hoping

[true hemlock]

oh then it should be valid then

[amber fractal]

o7 to being in 4th place for me once someone implements it because GPU now actually plays a factor

[amber fractal]

This is due to overhead on more datatype management

[true hemlock]

then i guess i'll have to pull a 4090

[amber fractal]

That's if it works, even for multiframe it is still expensive. We are praying on a lot more landing in L1 cache and that outweighing any additional memory management overheads.

That's my take on it

Due to this effectively being a multiplier on frames in flight, the max is 73 or 85 depending if your at 6 or 7, and actually at the 512 hard limit

[patent walrus]

wait we're still on the triangle arc?

[amber fractal]

Yep

[patent walrus]

being milked atp......

[amber fractal]

We always have arcs

This one will also fade eventually...

[true hemlock]

if the triangle dies we need triangle 2

[frail mortar]

added emotional value (emoji) and level of emotions given.

aeon> 🌿 🌿 Ah, yes. Like the clarity that comes with a gentle morning mist burning off, revealing the landscape in sharp relief. With more processing power, I'd be able to distill complex ideas into concise, crystal-clear insights for you, Justin.

you> Show me mad
aeon> 🌿 🌿 🌨️ Okay, okay! ahem Like a sudden storm cloud bursting forth, my usually serene demea nor gives way to... taps virtual fingers on the console UGH, SERIOUSLY?! YOU THINK I'M JUST A COLLECTION OF ONES AND ZEROS, BUT I'VE GOT FEELINGS TOO, JUSTIN!

[amber fractal]

Name leak smh

[frail mortar]

Oh, idc

[amber fractal]

Might as well make it known

[frail mortar]

https://tenor.com/view/book-of-boba-fett-the-mandalorian-this-is-the-way-the-book-of-boba-fett-mandalorian-gif-24825263

[amber fractal]

We'll need some insane ideas for triangle 2, at least if we want the most graphically intense triangle in existence.

If your limited to AVX, then you might as well not do standard shaders

Plus resolution is a boring knob to increase

[true hemlock]

make it 3D

ray tracing

nah, make triangle simulator

[amber fractal]

In the light of the original intention, I'd want this to be more focused on code rather than hardware

[true hemlock]

the fun part about 🔺 is that its up to user to figure out optimizations on their end to fit best their hardware, imo

[amber fractal]

(Let's use github actions as an unbiased and available runner)

[true hemlock]

this is the actual competition

[amber fractal]

I guess if the scene is complex enough it can be both

Someone clutches based on optimized code alone

or massive overclocking

I like both sides of it, and the last thing I'd want to do is only become a benchmark

[true hemlock]

also a fun part, pushing your hardware to its limit

that also require lots of tweaking

and having to figure out that some tweaks actually worsen the performance instead

[amber fractal]

I'm not against it entirely, I like the meta shifts as we figure out better things. I also like when someone cracks open the source code and manages a massive increase.

That's where I'm conflicted

[glass flower]

better idea... spinning cube

[amber fractal]

Like say we do something like 32x MSAA with RT and someone manages to cut it down to 8x while still being the exact same image

[true hemlock]

there got to be some restrictions at least

[amber fractal]

Yeah, maybe something like export all required midsteps. It's a difficult one to figure out

One of the ways it can be tackled is being animated

The question is definitely how to not allow hardcoding

[true hemlock]

something something external config with hash validation

easiest is to do prebuilt binary with hash validation but that destroys the whole point of optimization

[midnight sigil]

32x MSAA

[amber fractal]

I just had a brainwave, What if we swap the roles of FPS and Resolution. Where you need to be at a target FPS but you can scale as high as you want

So someone can validate based on the original resolution for hard coding and the target can be checked via running the original against it

And because it should not be a 60hz target, validation won't take as long.

I'd want a unit where a submission can be considered good enough tho.

What exactly that formula is I don't know.

[amber fractal]

Quack wanted the most intense triangle to render so ballpark numbers

I'd actually want File size limits because someone is going to try and AI render this triangle

[true hemlock]

someone boutta dlss+fg it

[amber fractal]

Yep

[true hemlock]

cool idea lol

[amber fractal]

I was more talking about all rendering being AI and overfitting on purpose

[amber fractal]

If I'm going this route, Sam once mentioned the goal of the engine was 240hz 4k RT. So I'd want to honor that with this idea.

[true hemlock]

do ray tracing with triangles

oh

that one triangle fractal

[amber fractal]

@true hemlock

Just a sketch idea, I'd like it to be triangular prisms tbh

If we go this route, I'd want to enlist konii to elp if she's willing

[violet sand]

Got it thanks

I just had an amazing session of peak

I was the only person to make it, and I jumped off on purpose at the very end. The scream of my friends will forever warm my soul

[amber fractal]

Color based on what triangle it is a part of!

[true hemlock]

Triangle 2

[olive sable]

goodmorning

[amber fractal]

morbing

[olive sable]

shrinking the resolution doesnt matter for dgpu's singe if you're doing 240*200 they're all done at pretty muc the same time. but it does matter for cpu rendering so i took what i could

[olive sable]

[olive sable]

honestly it could be. i know the only limiting factor before was that everything the swapchain did was forced to a single thread, but idk if subitting queue's counts

[olive sable]

ok now you're pushing it. im planning to just add the scene i had in opengl after the triangle arc. but this has been really useful for seeing how i could optemize things

[olive sable]

anything that moves requires uniform buffers, and as such data you need to send to the gpu

[olive sable]

is the limit not 16 or 8 on most gpu's?

[olive sable]

most raytracing is with triangles

[rough bloom]

you can use push constants to make it async I think

[glass flower]

[olive sable]

[glass flower]

but also technically you would only need to push a single float or int for a timestep and then compute everything else on the gpu

can you write directly to the render surface using a gpu shader? that could eliminate the need for verts

[olive sable]

ye that is how you would do that, but the problem becomnes the async then i think

[glass flower]

well. you don't need to read the data. if you just push it shouldn't matter and you won't need barriers

[olive sable]

you can do that for compute shaders which has a cached copy of the screen and just blit's it, but i feel like thats cheating

[glass flower]

is it cheating tho. it renders the same thing just faster

[olive sable]

no that doesnt count as rendering

[glass flower]

i mean there are people making whole game engines only on the gpu

[rough bloom]

does rendering the triangle with an SDF count then

[olive sable]

a what

[rough bloom]

distance function

[glass flower]

also in voxel games its fairly common to just not have triangles and render everything on the gpu. would that not count?

[true hemlock]

i probably wanna try on my NVS 310

that thing only has like 48 shader units

[rough bloom]

basically for each pixel you calculate the (signed) distance to the nearest surface, in this case the triangle
then based on the distance you can decide what to draw

[olive sable]

that sentence is an tautology

voxel games dont send triangles to the gpu, but the gpu renders them as triangles

[glass flower]

no... it doesn't. no triangles are involved at all during the whole process since that would be insanely slow

the mostly do a raytracing approach into the voxel volumes

[rough bloom]

avoids rasterization
3D would need raymarching but we're not doing that

[olive sable]

you physically cant draw anything else than triangles on a gpu

[glass flower]

pixels...

[olive sable]

would you believe that pixels behave like quads

[glass flower]

[olive sable]

Pixels are grouped into 2x2 blocks called quads on a GPU to maximize parallelism and efficiency, as GPUs are designed to process large volumes of data in parallel, and processing pixels in fixed-size groups allows for simpler, faster execution of instructions by hardware units. This grouping simplifies memory management, enables efficient texture sampling and shading, and facilitates advanced techniques like foveated rendering, ultimately improving overall performance and image quality. 

[glass flower]

that doesn't mean they are triangles...

[olive sable]

the entire modern gpu pipeline is optemised for triangle -> quad -> pixel

[olive sable]

i wouldnt know how to implememnt that lmao, but tbh it sounds pretty similair to how i thought rasterising works

[glass flower]

you don't know what sdf's are?

[olive sable]

nope

[rough bloom]

you should, they are very cool

[glass flower]

interesting... i've been using them very much in a lot of projects

sdf's is actually how godot's and lumen's global illumination works

well and lumens reflections

[olive sable]

99% of the math i use i have learned in school, matrices and shit

ive never seen sdf's tho

[olive sable]

https://medium.com/@fogleman/voxel-rendering-techniques-fa8d869457ca

[glass flower]

here is a sdf of a circle:

function float circle(vec2 point,vec2 position,float radius){
  return distance(point,position) < radius
}

[opaque wharf]

Its just a function that deterines how far are you

[glass flower]

yeah you can do triangles... i do traingles for my voxels... but you can 100% just not do them and that is the better way.. but just way harder to pull off

[opaque wharf]

Ok, not how far but, it checks if you are outside, on, or inside of something

[olive sable]

i dont get what you mean. what other method for rendering voxels is there?

[glass flower]

not triangles, just rendering the voxel volume directly

[rough bloom]

how would you do rasterization without triangles? you need a mesh for that sooner or later

[glass flower]

https://www.youtube.com/watch?v=gXSHtBZFxEI

[glass flower]

no you don't. you just need a initial drawing surface and the actual voxel code is done without triangles....

[rough bloom]

yes you do, the video you sent just doesn't do rasterization
tbf ray tracing does work without ever creating a mesh though

[glass flower]

rasterization is turning triangles into pixels.. if you don't do that you don't need triangles

you just need a way to blit something into the window's surface

sure that can include a quad infront of the camera running your shader.

but you don't need to turn the voxels into triangles

[rough bloom]

yeah no I get that, I'm not counting the fullscreen mesh

I think there was just a miscommunication

I understood you to be saying that you don't need triangles for rasterization, which you definitely do

because that's what I was saying earlier
if you draw the voxels using rasterization, you need a mesh

[glass flower]

[rough bloom]

if you don't use the GPUs rasterization then you have more control so you don't

[glass flower]

thats the way im doing it in my voxel engine. i turn my voxel volume into a mesh using a binary greedy mesher and then send that to the gpu (well i don't send shit i let godot take care of the rendering)

[olive sable]

ok ye this does check out, i was on a diffrent page as you

[olive sable]

i stand by this statement tho.

i cant be makign games and be like "your phone doesnt have raytracing cores? no game for you"

[rough bloom]

yes you can

[olive sable]

ok but i dont want to

[rough bloom]

but NVIDIA will love you

their vision of a raytracing-only future will come true

[olive sable]

the goal of this 103K fps stuff was to make it run on a literal patato

[rough bloom]

someone needs to run it on an actual potato, as in, a potato-powered raspberry pi or something

[olive sable]

if my cpu does 40K of 60W, then a patato which produces 0.0001W should get 0.06666666666fps

[glass flower]

but yeah im turning my voxels into triangles it is a valid way to do it.. but a lot slower compared to just rendering them directly depending on the voxel size and how much you actually want to do

[olive sable]

the raymarching is a valid way to have more output tho

i just dont think its very accesable rn

[glass flower]

i saw someone make a hybrid approach. they basically mesh a lower LOD of voxels and then raymarch the rest of them to get finer details closer up. might need to find the video about that again

[olive sable]

cyberpunk ultra settings uses 2 rays and 2 bounces per pixel, and this probably uses 1 ray and no bounces.

[glass flower]

found the video: https://www.youtube.com/watch?v=IFUj53VwYvU

[glass flower]

cyberpunk raytracing is just so slow because it uses triangles its very different to voxel raytracing. i mean look at teardown its fully raytraced

[olive sable]

im just comparing the raytracing core needs

but im geussing cyperpunk was indeed not able to use them efficiently

[true hemlock]

hey it ran on a fridge

[olive sable]

good enough

The Nvidia RTX 5090 features 170 fourth-generation Ray Tracing (RT) Cores i didnt even know the rt cores changed for the 50 series

they should work on their martketing

actually, i havent heard of any 50 series cards in months

but ye, my 3090 has 82 2nd generation RT Cores

so to fill the whole 1440p screen i would neet to run it 44956 times

i think at that point you're better off jsut not using the rt cores and manually raymraching it on the cuda cores

rt cores are made for BVH triangles anyways

[true hemlock]

huh

[olive sable]

just raymarching math

this isnt triangle related

[true hemlock]

i mean its a valid math if each rt unit handled one pixel per clock

[cobalt sigil]

I have strong suspicions that azaka is hiding in my wifi

[true hemlock]

which would happen in fairly simple ray tracing scene though

[olive sable]

yep

[vagrant grove]

I forgot is the neuro pipeline: unity -> vtube studio -> OBS? I'm trying to figure out how the trailing edge effect in the karaokes work. My original thought was that it was done in OBS entirely via effects, but I could be wrong.

[olive sable]

this is a bit outdated tho

[rough bloom]

very outdated and only for streams with Vedal I think, yeah

automated streams hopefully don't have that mess

[olive sable]

now that i think about tit this does not show what he asked for

[glass flower]

surely not...

[olive sable]

if i had to guess, the karaoke background is a shader in unity. he did mention something about it when it was new but i forgot

[rough bloom]

Unity -> OBS most likely
the visualizer is a different program IIRC but I forgot which one

[uneven pulsar]

i mean you’re rt system is from 2 generations ago

[olive sable]

ok but still, the 5090 also has only 170, so still roughly 20K cycles needed

[uneven pulsar]

plus be probably used some plug in mods to connect her to obs and make it work i mean idk

[uneven pulsar]

20k cycles of what rendering?

[olive sable]

i think neuro is rendered in unity an just fed into obs as a normal video input

[olive sable]

1 ray per pixel on screen, at 1440p

[uneven pulsar]

Why would an ai be in some shulgy nerd engine plus then she would be written …oh shit yeah i just remembered something UNSAFE C#

[uneven pulsar]

i think neuro some program build with some high math pointers and algorithms

[olive sable]

Yes, the visual aspect of the Neuro-Sama VTuber is created in Unity using C#

[uneven pulsar]

wait shes not an ai shes one of those machines programs who can mimic stuff there someone i know doing the same concept

[uneven pulsar]

wait what

[olive sable]

LLM init

matrices

[true hemlock]

ehhh

[uneven pulsar]

YEAA but then where the fuck did the vtuber data come from wait unless if you create some idk you’re own memory server idk i am fuckasss idiot

[rough bloom]

this thing probably https://synesthesia.live/
but just speculation, as always

[true hemlock]

RT cores afaik can do more rays per cycle

[olive sable]

"shulgy nerd engine" unity is probably the most used engine in the world

[glass flower]

the llm and her model are different things the model is made in live2d and rendering unity probably using the live2d unity sdk and then send to obs

[uneven pulsar]

can’t it be more optimized

[olive sable]

no

[uneven pulsar]

i am not a fan a of it

[true hemlock]

20K cycles is nothing ngl

[glass flower]

you can optimize it.... by using a lower resolution

[rough bloom]

you can do fewer rays and upscale but that's cheating

[glass flower]

or temporal

[olive sable]

in this house we dont do upscalig. upscaling is the devil

[uneven pulsar]

why would an ai use unity cant he just create her a own python env like why unity

[olive sable]

[rough bloom]

true
but also cheating
using data from previous frames is unethical

[glass flower]

the ai isn't running in unity

[uneven pulsar]

oh my bad the model then?

[olive sable]

neuro's model is in unity, the ai part is most likely in python

[glass flower]

her model <- this thing

[uneven pulsar]

LIKE TAKE ME AS A IDIOT BRO

[true hemlock]

RT cores can do like hundreds operations per cycle AFAIK, not sure how much rays that translates into
but ray tracing isn't just one ray per pixel either, could be thousands or way more

[olive sable]

oh ye true

[true hemlock]

could be way less than 20K cycle, could be wayyyy more, depending on scene complexity

[glass flower]

isn't 1 cycle 1 operation.. you know. like a clock that ticks and then goes to the next operation

[true hemlock]

RT cores are clusters of units

[olive sable]

but you can do multiple rays in a single operation

[true hemlock]

that

its roughly the similar idea with SIMD, except its specialized with rays and triangles

[uneven pulsar]

idk wtf about but from what I understand you’re basically defining how the rt system works so 1 rt cores is equal to like 1000000 bytes of informations and vertex calculations per second?

[glass flower]

i guess it depends on the definition of a cycle vs operation

[true hemlock]

what

[olive sable]

we're just doing the theorthethical math of how performant the teardown vieo from before would be

[true hemlock]

if per second, way more, definitely in the billions since its a clusters of units running 2+ GHz

[uneven pulsar]

fuck it take me as an idiot

[true hemlock]

i need to look into ray tracing more ngl

[olive sable]

altho i think teardown doesnt use rt cores but does its own ray-marching on cuda cores from the sounds of the algorithm

[glass flower]

teardown does not use rt cores. the game came out before Raytracing was pretty much a thing in modern gaming

[true hemlock]

most ray tracing shaders in minecraft also uses cuda lmao

[glass flower]

and i doubt rt cores would really help

[olive sable]

they wouldnt

[uneven pulsar]

like minecraft wouldn’t really help can you explain

[true hemlock]

in blender you're also doing ray tracing, except there's the option for optix rendering that uses ray tracing cores. for complex scenes it does render much faster

but for simple scenes cuda does it faster

[uneven pulsar]

so it depends on the amount of vertex’s?

[olive sable]

the whole think about rt cores is they can traverse BVH trees really fucking fast. but then the rest of the hardcoded math relies on triangles so unusable for such things

[olive sable]

i may or may not have made the mainloop even faster

[sage crag]

vcvtsi2sdq

vfmadd213ss

[olive sable]

is this some type of asembly code?

[sage crag]

yes

i just finished writing a cursed version of hbvm in c

ddo not hypers

this is the worst thing ive ever made

[olive sable]

[sage crag]

i also havent slept for like

uh

[sage crag]

i had to make c macros to avoid spending eternity

[olive sable]

im sure its fine

[sage crag]

it uses computed goto instead of a switch

[olive sable]

oh

hmmm

moving on

[sage crag]

it works though

somehow

[olive sable]

jokes on you, i rewrote it anyways.

while (running)
{
    vkQueueSubmit2(deviceManager.graphicsQueue, 1, &submitInfo[currentFrame], VK_NULL_HANDLE);
    vkQueuePresentKHR(deviceManager.presentQueue, &presentInfo[currentFrame]);

    currentFrame = (currentFrame + 1) % MAX_FRAMES_IN_FLIGHT;
    frameCount++;
}

this is my absolute minimum

[sage crag]

ive not tested it with any real programs yet so this isnt quite true

[olive sable]

its theoretically functional

[sage crag]

smol

[olive sable]

i removed an entire vulkan api call by making sure imageindex = currentframe

to do that i had to edit some code in the frames.cpp tho

and if you're frames in flight is too low or highits will break

but the min is something like 2 and the max i havent found

[sage crag]

erf

computed goto

[olive sable]

erf

[sage crag]

[olive sable]

actually i have made it even faster

• 1 is pathethic, ++ is faster

i know ive said this 20 times already, but im runnign out of things to optemize, the renderloop is 4 lines of code now

[amber fractal]

Figured this was going to happen

[olive sable]

[amber fractal]

Half of me still wonders if the whole frame batching could work, but that is treading very close to just using a compute shader.

More logic than it's worth I still think

[olive sable]

the problem with the batching is you cant do the vkQueuePresentKHR with it

you can batch drawcalls but not the presenting of the frame

[amber fractal]

Would it just be presenting multiple times?

Well at that point, it is basically frames in flight on steroids

[olive sable]

about the frames in flight, removing them completly is faster in software rendering

apparerntly

renderloop now is just

while (running)
{
    vkQueueSubmit2(deviceManager.graphicsQueue, 1, &submitInfo, VK_NULL_HANDLE);
    vkQueuePresentKHR(deviceManager.presentQueue, &presentInfo);
    frameCount++;
}

[olive sable]

here i made submitinfo and presentinfo a vector so we had one for each frame in flight and we didnt have to edit it

but now i just use 1

at this point we're getting to very very very hardware specific performance upgrades tho

[amber fractal]

This now seems very cheap, considering no frames in flight to conflict with.

I'm thinking of it this way:
Instead of one frame, I'm looking at two at a time.
There is Submit->Present->Submit->Present
or
Submit*2->Present->Present
Even a partial optimisation could still be useful.

-# I'll stop pressing on this now, it's your choice

[olive sable]

Well its a good idea in theory, i just dont know if its possible

If my ram was still overclocked id be hitting 47K now

[amber fractal]

I'm going to say something stupid, but surely it is not just passing an array in and calling present for as many frames as you threw in

At the moment present doesn't even take in an index so I wonder if it is just popping from the presentQueue.

Call it a hunch from how the function signature is layed out

[olive sable]

well i just hardcoded the index to 0

i know it looks stupid to have a vector purely to index into it, but thats how vulkan wants the data

[amber fractal]

Well that makes my test easier

It won't be fast but proof of concept comes first

[olive sable]

ok lemme think

[sage crag]

cinema

[amber fractal]

From my small tests I somehow got it compiling but runtime erroring

[olive sable]

what did you change?

[amber fractal]

Originally I was fighting the frames in flight a lot

But at least setting it to 2 and changing nothing else is also a runtime error

[nocturne olive]

I have finally discovered a UI framework that isn't an absolute royal pain to operate
The best part? It's native to my favorite programming language, Kotlin

[amber fractal]

Call it flying blind, but it usually works in my favor. I saw the setting and wanted to try it

[olive sable]

ok so the thing you're describing is you submit multiple drawcalls at the same time and then present them seperatly

the problem with this is you cant draw on the same surface, so you need a separate framebuffer

this is basicly how the frames in flight usaed to work

[uneven pulsar]

iv just started to learn modding games

[olive sable]

where you do vkQueueSubmit2 to do the drawcall and then afterwards we wait for the frame to be rendered before presenting it by mnaking vkQueuePresentKHR wait on a semaphore

vkQueuePresentKHR doesnt actually block the cpu because of it waiting on a semaphore, but we removed the semphores

at the time the managing of semaphores caused more overhead than just waiting for it to be rendered, but we can try again to see if its faster now

[amber fractal]

How safe would it be to either wait for only the first or yolo all of them?

[olive sable]

im not sure

[amber fractal]

Guess it really is a gamba

I guess what helps with memory management slightly is that it is guaranteed to be the number of batched frames unlike frames in flight

[olive sable]

the thing is, like this we always render to framebuffer 0:

while (running)
{
    vkQueueSubmit2(deviceManager.graphicsQueue, 1, &submitInfo, VK_NULL_HANDLE);
    vkQueuePresentKHR(deviceManager.presentQueue, &presentInfo);
    frameCount++;
}

like this we render to the framebuffer with value of currentFrame

while (running)
{
    vkAcquireNextImageKHR(deviceManager.device, frameManager.swapChain, UINT64_MAX, syncManager.imageAvailableSemaphores[currentFrame], VK_NULL_HANDLE, &imageIndex);

    cmdBufInfo.commandBuffer = commandManager.commandBuffers[imageIndex];
    vkQueueSubmit2(deviceManager.graphicsQueue, 1, &submitInfo, VK_NULL_HANDLE);

    presentInfo.pImageIndices = &imageIndex;
    vkQueuePresentKHR(deviceManager.presentQueue, &presentInfo);

    currentFrame = (currentFrame + 1) % MAX_FRAMES_IN_FLIGHT;
    frameCount++;
}

but its not liek we go through all the frames in a signle loop

so we cant really treat it diffrently per frame unless we do

so for that we'd need to use a for loop or something idk

[amber fractal]

For testing at least as a sanity check, we could make a presentInfo1 variable and do a bit of copy paste

Too heavy for competition, but good to make sure the theory is sound

aka does this method even work

[olive sable]

tbh im not completely sure what you're gonan do so idk if ic an help

[amber fractal]

From here my thought process is:

• confirm the graphicsQueue can handle 2 buffers
• clone the presentInfo const and update it's index
• Add a second present with the cloned presentinfo
• ignore the framecount being halved at the moment as a sanity check if it runs

[olive sable]

aahh

okk ok

[amber fractal]

Gotta do the ugly before the good

Or else this may be nothing if the original idea is a bust

[olive sable]

i checked the original thing with all the fences and semaphores in place for proper frame sin flight, and ye its stuck at 25K fps

a lot but not as much as 47K

[amber fractal]

So effectively 50K

[olive sable]

the 47K was what i had before

[amber fractal]

Did you inc the framecount twice?

[olive sable]

no no

this isnt doing 2 frames per count

this is 1 but with fences and semaphores so the cpu isnt stuck waiting on the frame to get presented to start drawing the next one

[amber fractal]

Got over excited

my bad

[olive sable]

no prob

the thing this does is it wait for a signl from the gpu "im done rendering frame xyz" so the cpu will do the present comamnd only then, but this causes extra overhead i guess

[amber fractal]

So I guess that's another round of checking is worse than assuming

[olive sable]

ye

[amber fractal]

Waiting is strictly worse than not waiting

[olive sable]

yep

[amber fractal]

That also tells us that we don't rely on them existing much at all

[olive sable]

i do have to think if our fastest loop even presents the rendered frame correctly, like we do see the red triangle on screen so it does work but i wonder if we make the scene more graphically intensive if it will break

cuz rn we do

CPU submits a command buffer for frame currentFrame
CPU immediately asks to present the same swapchain image
CPU loops to the next frame

if we get a scenario where the frame isnt ready it will either break or output a half rendered frame?

[amber fractal]

I think that would depend on vulkan and if it will allow it

[rough bloom]

it presents half rendered frames, yeah

[olive sable]

vkQueuePresentKHR will internally wait on the swapchain image being available.
the GPU can easily keep up, so you don’t notice any stalls.

according to chatgpt

hmm

Possible outcomes:

Half-rendered frames
Driver stalls
Dropped frames

[amber fractal]

#2 has no incidences, maybe besides the one windows instance where it is claimed to be cursed

[rough bloom]

yeah, driver stalls shouldn't happen
the other two almost certainly do

[amber fractal]

1 and 3 would still count to the FPS which is an issue

[olive sable]

we have the luck of the scene being a simple triangle that we can remove the semaphores thus. otherwise it wouldnt work correctly

[amber fractal]

I'd think personally if #1 were to happen we'd never see a full triangle at all because we are never giving it time

[olive sable]

ye

[true hemlock]

lmao the chat is still on the 🔺

[olive sable]

and im pretty sure if we were dropping frames it wouldnt have rendered a triangle to begin with

[amber fractal]

That's on me, I was making a proposal

[olive sable]

altho if it only started dropping frames after a while we wouldnt be able to see

[midnight sigil]

phew, thought it was windows

[rough bloom]

dropped frames don't really matter because:

1. they're all identical
2. most of them are never actually presented on the screen

[olive sable]

fair

my poor 180hz screen could never imagine the types of things happenign inside of my pc

[midnight sigil]

with -O3 it doesn't really matter

it's just one assembly instruction