programming

[umbral wigeon]

I spitting shit 12000 LOC on pr github
https://github.com/weenachuangkud/FastCast2/pull/39

And I want to show you guys that please dont vibe code

[stark needle]

Advice of the millenium

Don't vibe code✅

[umbral wigeon]

Yes it's working but I've not implemented everything yet

[rough bloom]

you can't stop me

-# (my wallet and sanity will)

[umbral wigeon]

(If you don't have money for Ai)

[jagged turtle]

nintendo and DMCA, name a better duo

[umbral wigeon]

(Opencode feels so good ngl bro)

2000000~ tokens used

[jagged turtle]

also to whoever said their manager doesn't know github, I really hope that manager isn't your manager soon

[obsidian mantle]

Its normal here

We arent really IT

[jagged turtle]

?

[umbral wigeon]

I think manager will not understand my PR about Motor6D, SoA, AoSoA, parallel luau, load balancing robin-round cycling, queueEvent, Serial/Parallel supports
(This is summary of my PR)

[obsidian mantle]

Somewhat between normal engineering and IT
All our projects are one-guy only

[jagged turtle]

oh looooooool

[obsidian mantle]

The most programming i did in my life is derived from this server

Chess tournament and skyrim integration thats it

[jagged turtle]

[obsidian mantle]

What i do at my job is trash compared to this

[jagged turtle]

what do you do at your job then

[umbral wigeon]

Me scripting legos for 3 years

[obsidian mantle]

They give you a programmable controller that is currently in meta (in meta means its not banned yet and its cheap enough)

[umbral wigeon]

I love lua

I wanna kiss lua

[jagged turtle]

I wanna see your lua shrine

[jagged turtle]

uh huh

[obsidian mantle]

It has some ide for it
You make program that takes info from sensors, processes it, lays it out in registers for the server to read

[jagged turtle]

and

oh

what is it called

[obsidian mantle]

Idk like middle level

My position is called engineer-programmer

[jagged turtle]

why don't they just call you software engineer

or just engineer

[umbral wigeon]

It's in the name ig "engineer-programmer"

[obsidian mantle]

Because sometimes i have to go to some forest and modify my program and do a lot of shit with wires

Screwdriver work too

Like that

[jagged turtle]

oml

[umbral wigeon]

1 day left and I'm about go robotic engineering cus I wanna build some cute girls to cuddle

[obsidian mantle]

Currently i am waiting for the guys to let me into their computer so i can configure it so the data is shown on their screens

They are not letting us in because no licence

Licence is impossible to get rn

[umbral wigeon]

Pray for me

https://tenor.com/view/wallahi-say-wallahi-say-wallahi-bro-ishowspeed-tiktok-gif-14576999865344060530

[obsidian mantle]

Its like
Shit tier programming

No github needed

[umbral wigeon]

Wait, where do you store your code?

[obsidian mantle]

I just put my version control inside of source code "version 1.2.2 - fixed bug that crashed the device"

On my laptop and copy it sometimes on company server

[umbral wigeon]

https://tenor.com/view/cat-sticker-line-sticker-bye-sticker-goodbye-sticker-hand-wave-gif-2765574846828776801

[obsidian mantle]

Basically we face new shit all the time
Its not about what you know
Its about how you can work with shit you dont know

And since im an uneducated fuck who doesnt know shit about programming, it suits me

Random ides, sometimes you need to find ancient program to configure some radio
Ride to some forest shithole for a week
Find spot where there is cell signal to find some manual for a device that wasnt declared in the project but its actually is there and you must configure it

All kind random engineering and programming shit

Thats why my position is called like that. Technically correct name

Work trips are nice you get 12$ a day as a bonus (its absolutely not worth it i hate my job )

Advice: dont get born in eastern europe

[jagged turtle]

oh brother

[obsidian mantle]

Its enough

Not like anybody else will read my code anyway ever

[jagged turtle]

true

[mighty thorn]

Okay

Rich
California
Rich

[glad path]

i don't think that's what you do in robotic engineering but I could be wrong

[maiden geyser]

just be in a right place in a right time in a right mindset with a right set of competences
clearly, the homeless are just lazy for not buying a house

[mighty thorn]

https://youtube.com/shorts/68QlDd8_jaE
how to avoid seeing other opinions 101

[olive sable]

https://tenor.com/view/why-what-in-the-hell-oh-god-pizza-house-on-fire-gif-15536410

[opaque sigil]

[olive sable]

hi

i was playing clone hero

and missed the bi-weekly dramas

[opaque sigil]

just another day in #baking

[olive sable]

[opaque wharf]

Sam. I found a way to cheat during exam

[olive sable]

okay

[opaque wharf]

Try this one simple trick

[olive sable]

you could have studied instead

[mighty thorn]

How much did he pay you to vote no

$50k?

100k?

[olive sable]

https://tenor.com/view/nothing-cheese-with-nothing-nothin-nuttin-polly-o-cheese-gif-319919207048056034

the sparks aren't his iirc

wont leak too much tho

[maiden geyser]

holy smokes, and an NDA?

[olive sable]

i wish

im poor rn

[jagged turtle]

hello omemga

[olive sable]

[jagged turtle]

:omemga

[fast pagoda]

:omemga

i am experimenting with zig and llm training

very exciting

[jagged turtle]

nice

[fast pagoda]

pytorch immediately exploded due to having an nvidia card and an amd card installed at the same time

[olive sable]

https://tenor.com/view/mili-fuzi-victim-neuro-evil-neuro-gif-14779149578470937570

[fast pagoda]

[olive sable]

🎸

[fast pagoda]

fuck i havent played gh in so long

[shy axle]

just use the crappy cpu that you found in the mysterious tech pile + hopes and dreams fr

[olive sable]

https://tenor.com/view/funny-does-he-know-gif-6358776805734031876

[civic flax]

Greetings I come here seeking help

[fast pagoda]

the CPUs in my server were $35 each

[olive sable]

mine are like 15 each iirc

[civic flax]

Long story short, I have a veadotube avatar that i want to use in OBS but it shows black

But even then, when I tried to use autoconfiguration wizard, it also failed

I checked if my GPUs were up do date and they are, so I dont know what am I doing wrong

[olive sable]

uh

"veadotube"

i see

[civic flax]

its a pngtube thing

[fast pagoda]

i’m getting a black/blank/pink/green screen on Windows!
that’s usually a graphics card issue. update your drivers, and DirectX also! refer to the install guide. you can also try to run the app as administrator!

https://veado.tube/docs/help/

veadotube on OBS gives me a blank screen!
lots of possible issues here!

if using Game Capture on Windows, try selecting SLI/Crossfire Capture Mode (Slow)!
try switching between Game Capture, Window Capture, Spout2 Capture, Syphon Capture… see which works!
if on Windows, try forcing both OBS and veadotube to run in Performance mode, in your system graphics settings.
in general, refer to the OBS guide, which goes through each capture mode.

[olive sable]

for obs there's this tutorial

veado.tube/docs/usage/obs/

[fast pagoda]

https://veado.tube/docs/usage/obs

[olive sable]

ye that

[civic flax]

thanks

[fast pagoda]

[civic flax]

i am just tired of having to deal with GPU related issues i hope it is not the gpu

Do you think the GPU is the primary reason why when you turn on your computer it sometimes show a glowing black screen?

[fast pagoda]

if my computer does not boot immediately i usually assume it is memory training

[civic flax]

Like, I see the logo and the spinning wheel after the Windows jingle, most of the time is a black screen

[fast pagoda]

does it ever boot after that

it's a black screen BEFORE the windows logo sequence?

it may just be configured not to show uefi/bios splash

[olive sable]

glowing black screen as in the backlight is on bu no display?
what OS are you using?

windows or linux?

[shy axle]

From scratch or fine tune?

[civic flax]

After

[civic flax]

Windows

[olive sable]

hmm

[civic flax]

I have been having this issue for like 10 months

Maybe less

[olive sable]

hmmmmm

[civic flax]

I have FR this laptop 3 times and the issue persists

[olive sable]

laptop

[civic flax]

https://tenor.com/view/crashing-out-gif-1614287010081694538

[olive sable]

you said you updated the gpu drivers right?

[civic flax]

Yes

Through device manager and Nvidia's little stupid app

[glad path]

undertale ost

[shy axle]

Somebody understood 🥹

[olive sable]

your laptop has an IGPU on the cpu

so i guess update cpu drivers

[civic flax]

i checked all of the stuff and is all updated

[floral hawk]

Qwen3 TTS is being a bitch to setup but I got my Hungarian homegirl to let me clone her voice

I had to run it through CUDA BC latency was hell and it was zero-shot generating voices per sentence. It's still not stable

[maiden geyser]

undertale references in 2k26, ok grandpa

[olive sable]

then i have no clue

[fast pagoda]

honestly at that point it's time to pack it in and swap to a different software for this purpose

[civic flax]

So no OBS

[fast pagoda]

it's probably not OBS if i had to guess
never heard of veadotube but it's definitely the easier sacrifice if i had to guess

im guessing 2x

[civic flax]

Yeah

[proven merlin]

Ensure both programs run on the same GPU

[proven merlin]

It might bug out if they arent, i Personally can bypass the black screen by using a different capture method

[proven merlin]

That one hurts, as i dislike Python, like C, but am fairly blind

I feel attacked

(but good one)

[silent cloak]

C destroys python anyday

[fast pagoda]

this is going great

[silent cloak]

https://tenor.com/view/rafael-nadal-rafa-microscope-tennis-tennisgifs-gif-20844960

[amber fractal]

That pan is killing me

[fast pagoda]

 thy true's

Norall thou not thou thy thou before it shave, thee lover of
Which to'llBatch size or sequence length mismatch
Batch size or sequence length mismatch
Batch size or sequence length mismatch
Batch size or sequence length mismatch
Batch size or sequence length mismatch
Batch size or sequence length mismatch
Batch size or sequence length mismatch
Batch size or sequence length mismatch
Batch size or sequence length mismatch
Batch size or sequence length mismatch
Batch size or sequence length mismatch
Batch size or sequence length mismatch
``` shakespeare himself

[silent cloak]

Just random pytorch output

[fast pagoda]

this is zig

your favourite

[silent cloak]

Pytorch masterpiece

Making ur own library?

[fast pagoda]

i cloned some random guy's zig .12 repo for an implementation of https://github.com/openai/gpt-2/blob/master/src/model.py and have been porting it to .16

i havent replaced 99% of it like the diff implies it's just been formatted lol

nice svelte 1928 lines

this is the first time it has actually compiled tho'

turns out

lotta shit changed between .12 and .16

std.io.getStdOut() was used a lot and has obviously been slain so i replaced with std.debug.print, std.rand is gone for std.Random very important to rename that , readVarInt died with I/o changes so std.Io.Reader lacks it completely, had to swap a bunch of var to const due local vars no longer allowed if they dont mutate, and he had a @cImport("math.h") that was not needed because Zig can in fact do math without c's help

[silent cloak]

Kek

Why gpt2

[fast pagoda]

fast

[silent cloak]

Aint that shit pretty dated by now

Pretty sure its still running tensorflow

[fast pagoda]

it do be tensorflow

[silent cloak]

Dinosaur in machine learning years

The great torch migration was a big deal

[fast pagoda]

i just kept it because that's what it was and i figured port first and then deal w/ figuring out how nightmarish that part i s

yeah my first loras were TF

i miss the dashboard tbh

[silent cloak]

Tbh i hated TF

I was glad to jump to torch when i was able to

[fast pagoda]

i barely remember anything about it ngl

[silent cloak]

Keras and tensorflow are both relics now

[fast pagoda]

it was when i first was running any training of any kind

mostly in colab notebooks which skipped a lot of the pain probably

[silent cloak]

Those and deeplearning4j were the first libraries i used

Before that it was from scratch

[fast pagoda]

<scratch language joke here>

[silent cloak]

I didnt like how clunky they were until i didnt have much of a choice

[fast pagoda]

as jank as it still is it is kinda wild how much MORE dogshit these frameworks were a few years ago even

[amber fractal]

Scratch LLM

[silent cloak]

Its a billion years in the world of programming

Those things are in a museum

[fast pagoda]

i know tf and jax are still supported alongside pytorch in google cloud infra but idk what they get used for other than google's own stuff for jax

[silent cloak]

I never used jax

It was pure TF for me or TF with Keras2

Sometimes DL4J

Then just torch

I used other libraries too for other use cases but alot of the times i just wrote it all myself

I forgot what one I used for NLU though

[young plover]

lmao
Nintendo never sent "the email" to me personally and I'm keeping it that way.
I'm also not convinced it'll ever be possible unless some NSA backdoor is discovered. They locked the Switch 2 down tight.

[silent cloak]

Cap

It'll happen

Unless they come and take away the hardware or stream the console to the device

[fast pagoda]

we aint converging with this one kekw

[shy axle]

what size u training?

[tender river]

switch 2 hardware bricks itself if it detects tampering attempts

very annoying

[fast pagoda]

it's just gpt2 124M (cpu training) on tiny shakespeare

zig trainer

[shy axle]

W

[fast pagoda]

loss is throw a dart at a board

rn

[shy axle]

its diverging lol

[fast pagoda]

honestly that might describe it well if it really explodes

[silent cloak]

they put all their money and effort into preventing mods than making good games

but it will still be broken with time

as every console eventually is

[fast pagoda]

what exactly is the benefit to owning a switch 2

[frozen igloo]

https://tenor.com/view/pokopia-burning-stare-burn-funny-ditto-pokemon-burn-gif-8329318217662298298

[fast pagoda]

understood. Carry on

[jagged turtle]

$500 dlc to your physical hardware

[amber fractal]

My favorite creature

[fast pagoda]

compelling

[fast pagoda]

you forgor

[fast pagoda]

he's helping how he can

[fast pagoda]

gpu pytorch version LMAO

[floral hawk]

So guys, qwen TTS sounds good but it spikes the hell out of my CPU and the latency is hell, what am I missing here??

It's also literally crashing my animation controller. I'm not sure where I went wrong, but 15s of latency AND it's absolutely eating my CPU even tho it's GPU CUDA bound

[amber fractal]

Small edit to a sample

[olive sable]

i see a face

face pareidolia

[silent cloak]

cat

or joker

https://tenor.com/view/joker-batman-the-joker-why-so-serious-gif-14263650074740919861

[amber fractal]

There is no facial images in this dataset

It did this on its own

[floral hawk]

Well I found out why it was eating my CPU, but the latency is still very real

I can see online that it has 97ms to first packet so like what the actual fuck

[olive sable]

cat

girl

catgirl

[rigid snow]

[olive sable]

Idk if this shows it properly what im seeing

they have a scarf

[rigid snow]

[midnight sigil]

I can't see the joker

[rigid snow]

that's what i saw first, how

[olive sable]

i cant see it either

only catgirls

my mind it poisened

i blame quack

[shy axle]

[rigid snow]

look at this then the original maybe you'll see it

[olive sable]

oh

very faintly

[rigid snow]

it's super clear to me

[midnight sigil]

[rigid snow]

also this: run it on gpu, you need maybe 3gb of vram, i already tested it on cpu yesterday and it just does not work, from my testing i got 0.25x realtime on 8 threads xeon e5 2680 v2
on latency being 15s: what the fuck? i'm pretty sure you can configure streaming, here's the first google hit https://github.com/dffdeeq/Qwen3-TTS-streaming

[floral hawk]

Yeah I had a similar reaction to the latency, even with streaming it was like 6s

I shifted and I'm going for Kyutai Pocket TTS, Qwen legitimately just isnt fuckin worth

[rigid snow]

streaming tokens in or streaming audio out? because i'm thinking the latter, idk if the first one is possible with this model'

[floral hawk]

I was also running full on GPU but piper voice modulation (built for cpu) was running over Qwen and causing issues.

Audio out

For twitch its just not practical.

[rigid snow]

i suppose you weren't running the gguf quant

[floral hawk]

I tried both it and the FP16

[rigid snow]

damn

[floral hawk]

I think it was user error, or incompatibility at some point in my codebase

Zero is built very fucking convulutedly so it's entirely possible I overlooked something somewhere but I don't think so

Nonetheless, I don't wanna keep fw it

[rigid snow]

probably not user error

[floral hawk]

I mean I was even doing full GPU inference

[rigid snow]

[floral hawk]

Yer

Somewhere, somehow, I probably fucked something up tbh

Or maybe not

I trust your judgement.

I'm gonna do Kyutai tho. 200ms latency and voice cloning is actually wild, and running fully on ram is even moreso

Then I can just shift my vision over to GPU.

[rigid snow]

en only, no training code

[floral hawk]

Big sad, yeah

That was my main pro for Qwen

Zero is meant to be JP bilingual

She initially was

But over evolving use case it got canned. i was hopeful to bring it back

[rigid snow]

oops not en only but still no jp

[floral hawk]

どうしようもない

(Nothing can be done/That's life)

[opaque wharf]

https://tenor.com/view/ada-indonesia-coy-adaindonesiacoy-indonesia-ada-indonesia-coy-gif-7180128435367941739

[fast pagoda]

right side console is python//pytorch running the same trainin vs cpu on left w/ scuffed zig trainer lmao

[fast pagoda]

almost just pasted an ssh key here

by accident

[stark needle]

[olive sable]

[stark needle]

actually

is that shit a fish or a plant

[marsh coral]

i think its closer to a fish than a plant

or smth more ancient

[floral hawk]

Pocket TTS goated

[marsh coral]

[stark needle]

what

this shit is predatory????

i just know clownfish protect them and get protection back

or osmething

[manic fossil]

Everyone has to eat

technically plants predate on stars

[glass flower]

they eat crabs

[olive sable]

Triangle3D

[opaque wharf]

I usually pasted some command too but never keys lol

[sage crag]

aminal

NOT fish

cnidarian

related to

jellyfish

coral

and hydra

[glass flower]

they are hydras?

[sage crag]

❌

related to hydra

[fast pagoda]

https://tenor.com/view/monster-bob-blob-stare-gif-1160213367645308177

[rigid snow]

what is this

vercel

[glass flower]

poor mans rust

[rigid snow]

LMAO yeah why do this when rust exists

[glass flower]

they renamed structs to shape... there is no reason...

[sage crag]

structure

shape

[rigid snow]

where are you getting this from, did you look into it or is there a digest somewhere i can look at

[sage crag]

morph Wow {}

compose Wow {}

[glass flower]

random twitter message that popped up after i read the original tweet

[rigid snow]

docs?

ah

[sage crag]

geom Wow {}

"zero docs"

[rigid snow]

skills are docs

[sage crag]

easy search

"zero language docs"

"zero language ai docs"

"zero language new ai docs"

[rigid snow]

are you the agent

[sage crag]

"zero language new agentic ai docs"

[sage crag]

"zero language json new agentic ai docs"

it not come up

[rigid snow]

no itd be like "zero language new docs 2025 hello world example how to print"

[glass flower]

wait... their filename suffix is a 0

[sage crag]

thinking for 10 minute

ugly

[glass flower]

hello_world.0 is a filename

[rigid snow]

ye

[glass flower]

https://github.com/vercel-labs/zero/tree/main/examples

[sage crag]

me make language

helloworld.yuri

[glass flower]

they renamed Option to choice too...

[rigid snow]

fuckin choice are we deadass

HEY

[glass flower]

i found it first

[sage crag]

string String

src code

rust without

memory safety

[rare bramble]

all the examples are like 5 lines long

[sage crag]

borrow checker disabled

[glass flower]

i didn't think they would manage to make rust uglier... but they did it

[sage crag]

[N]T waow

[warped narwhal]

What is this language meant to be used for?

[opaque wharf]

To be fair, rust is not ugly, just verbose

[sage crag]

better typesystem const expr than rust

[rigid snow]

raises

raise

[sage crag]

struct Wow<T, const N: usize> where [T; N]: (), {
items: [T; N],
}

[civic flax]

How do I do that?

[sage crag]

make machine with

only 1 gpu

[rigid snow]

genuinely why use this instead of rust

[glass flower]

idk

[sage crag]

agentic json waow

ai compiler

0 = rust + ai

True!

[rigid snow]

rust == -ai

[warped narwhal]

Is this an example? The code doesn't set len if you init with values

[sage crag]

[opaque wharf]

E = mc²+AI

[sage crag]

ye

[glass flower]

its from here: https://zerolang.ai/reference#functions

[sage crag]

0 = rust + ai
E = mc² + ai

mc² - E = rust

[rigid snow]

buket

[sage crag]

zero lang dot ay eye

[rigid snow]

tehre is a borrow checker

they. brorowed itr

:badone

[sage crag]

:mlnt

[rigid snow]

The current native checker tracks simple lexical borrow conflicts, rejects assignment while a value is borrowed, and rejects returning references to local bindings. Borrows lower to direct address expressions; there is no borrow registry or runtime alias metadata.
well

big word

me

[sage crag]

basically

[jagged turtle]

[sage crag]

cant assign to value that has references

cant take reference to local value

doesnt support

lifetimes

or

interior mutability

so essentially

useless, not real borrow checker

[topaz tendon]

My first triple-A gaming experience on Linux
(game doesn't launch)

[sage crag]

you picked like

the most recent major game that doesnt work

due to

NOT drm reason

[topaz tendon]

A game's a game

Guess dualbooting remains a necessity

[tender river]

can vibe coders please all switch to zerolang

i dont want my eyes to accidentally stumble upon any of their output

[glass flower]

based on protondb it should just work.. https://www.protondb.com/app/2483190

[sage crag]

mipmip

vibe coders all switch to

[topaz tendon]

Yeah that's the weird thing, it just doesn't

[sage crag]

mojo

[topaz tendon]

Tried several of the env variables to try and change it but nope

[glass flower]

i hope you mean launch options

[sage crag]

me make ying language

[rigid snow]

game env vars

[tender river]

main.🦊

[sage crag]

helloworld.🦊

[rigid snow]

MAKE

[topaz tendon]

No, env variables, like PROTON_ENABLE_WAYLAND, VKD3D_CONFIG etc

[sage crag]

🦊 allowed

fop allowed

fox allowed

ying allowed

[glass flower]

but you use them through the launch options in steam right? like copy that block and then put it into your steam launch options.
but also what proton version you using?

[sage crag]

huli allowed

[topaz tendon]

Yeah that's where I put them. I tried Proton-GE, proton-cachyos-slr and Proton Experimental, same thing on all of them

[glass flower]

this is really odd since its steam deck verified

[topaz tendon]

I guess Linus isn't the only one with the Linux curse

[sage crag]

just hu sufficient

[glass flower]

try following this: https://steamcommunity.com/app/2483190/discussions/0/839502231284745285/

like read the comments and see what works

[sage crag]

compiler will reject hushou

[topaz tendon]

I've seen that thread, tried all of that

I did my googling before complaining

[glass flower]

well... idk then

[tender river]

lsp that registers all 2-letter acronyms as the possible extensions for the language

[sage crag]

AA

AB

AC

[topaz tendon]

At least my main games do work better on Linux, so it's not all that bad. Enough to make Linux my main OS instead of Windows

though I hoped I would only need Windows for VR

[sage crag]

abandonded archive
active block
assassins creed

cf

cold fish

[glass flower]

i mean... the game obviously officially works on linux.. so it has to be a issue with your setup and not directly the game

[sage crag]

wrr

[topaz tendon]

Clearly, but hell if I know what is it if other people also shrug their hands trying to help me out

[glass flower]

i did notice that you ran it with umu? atleast that what it says in the screenshot you sent you didn't pirate it right?

[topaz tendon]

Nope

dunno what umu is, saw it in the log tho

[sage crag]

huhu

xiao huli

[glass flower]

wdym you don't know what umu is... umu shouldn't just launch with steam. its a way to launch non-steam games through a steam emulation layer iirc.
do you have it on gamepass?

[sage crag]

xiao huniang

[topaz tendon]

No, I have it on Steam, but since Steam and its launch options didn't do anything to get it to run, I've been also trying to run it via Lutris

That's what the screenshot is from

[glass flower]

i see...

[sage crag]

wrr

[glass flower]

what exactly happens when you launch it through steam? just nothing?

[fast pagoda]

pogga

[topaz tendon]

Different things.
Most of the time, absolutely nothing.
Sometimes I'll see the splash image for a few seconds.
And like one or two times I even got a black screen, as if the game actually tried to open for once.

[glass flower]

weird...

[fast pagoda]

that game isnt even out yet

[glass flower]

what gpu do you have?

[topaz tendon]

It's in early preorder access

[topaz tendon]

4080S

[fast pagoda]

sounds like a great recipe for not working immediately with every linux setup

[glass flower]

nvidia open or the proprietary drivers?

[opaque wharf]

What compositor did you use? Because I know I've experienced similar things with KDE and when using gamescope, it just works

[fast pagoda]

probably doesnt even work with every windows setup

[topaz tendon]

The proprietary, the most recent driver

[glass flower]

gamescope and nvidia are a very bad combo (sometimes)

[topaz tendon]

It is KDE, I didn't try gamescope but I can give it a try

[opaque wharf]

Yeah, sadly it isn't

[glass flower]

but that means you didn't try all the launch commands because all the ones i saw used gamescope

[jagged turtle]

.no

[opaque wharf]

But then again, I do know that from experience because when I play Sun Haven, it just changes the button from Play to Stop then back to Play

[topaz tendon]

I focused on the env variables lol, I thought people were only using gamescope out of preference unrelated to the game not launching

Wasn't aware gamescope can solve such issues

[opaque wharf]

When I use gamescope the game runs just fine

[glass flower]

it might. it might not... idk

[fast pagoda]

isn't proprietary done

580 being the last proprietary driver

maybe update to the current 595

trying to run it with an old driver wont be great prob

[glass flower]

what

[topaz tendon]

It is 595.something, lemme double check

[glass flower]

proprietary is still alive and well...

[topaz tendon]

Yeah, 595.71.05

[glass flower]

they just changed how its packaged

and moved the old ones to AUR

[topaz tendon]

Yeah it's like proprietary with open kernel modules or something?

Or the other way around

[fast pagoda]

nvidia package is nvidia-open now though so if you're on just nvidia it'll be open ig

last nvidia driver i installed was for volta so im stuck on 580 anyways idk, im probably conflating the open becoming default to proprietary ending completely

but i dont think you need proprietary anymore unless you are on a below turing card

[glass flower]

but its still the proprietary ones just named differently no? otherwise they would just be in the kernel

[topaz tendon]

  Device-1: NVIDIA AD103 [GeForce RTX 4080 SUPER] vendor: Gigabyte
    driver: nvidia v: 595.71.05 alternate: nouveau,nvidia_drm
    non-free: 550-580.xx+ status: current (as of 2025-11) arch: Lovelace
    code: AD1xx process: TSMC n4 (5nm) built: 2022+ pcie: gen: 4
    speed: 16 GT/s lanes: 16 ports: active: DP-2,DP-3 empty: DP-1,HDMI-A-1
    bus-ID: 01:00.0 chip-ID: 10de:2702 class-ID: 0300

[fast pagoda]

so yea non-free the proprietary module only goes up to 580.xx

after that it's open only

[topaz tendon]

These aren't fully open though, are they?

[fast pagoda]

but i dont think it functionally is any different to install like there isnt another package ig

[topaz tendon]

Maybe it's more complicated and I just don't get it yet lmao

[fast pagoda]

it's nvidia it's stupid

[topaz tendon]

I thought nouveau is the fully open one but it has poor gaming compatibility

[stark needle]

<insert nvidia copypasta>

[fast pagoda]

driver 590, arch in the aur pkgs replaced nvidia with nvidia-open, nvidia-dkms with nvidia-open-dkms, and nvidia-lts with nvidia-lts-open

[floral hawk]

This mf stole my name

Absurd

[fast pagoda]

so yea anything post 580 is open

[stark needle]

@zero is this true

✅

[fast pagoda]

omg it's u

[floral hawk]

It would be so cool if I could have her in here

She has discord integration already

[fast pagoda]

[floral hawk]

But having her in public servers besides my own is something ill prob never do

Just not really smart.

[fast pagoda]

she'd get you b& here anyways

[stark needle]

youd need admin permissions anyway

[fast pagoda]

well

she'd get banned prob

[floral hawk]

Yeah ik

I thought other people had their AI in here

[stark needle]

only if its a self bot, otherwise an admin would have to allow it

[floral hawk]

Not really something I want to do, just curious

[fast pagoda]

not generally without permission, if they dont have bot tag and are a bot it's a self bot which is TOS which isnt allowed due to community server

[floral hawk]

Tru tru

[stark needle]

@fast pagoda btw the dflash didnt work cause my model is fp8 quantized

[fast pagoda]

F

[stark needle]

i will still try mtp

god i wanna get this 1 million context working so bad

[floral hawk]

I'm working on the final polish for V1 of her runtime then it's just twitch integration and her first full release is final

[stark needle]

the GDN layers use 80x less compute

[floral hawk]

The minecraft brain definitely needs work tho, she's really not that smart in there

[fast pagoda]

:omemga

[stark needle]

i still need to figure out how to do actual dynamic thinking

wait the dflash model was also using kv cache

what

[fast pagoda]

it involves kv tho right it just blasts hidden features into the kv

DFlash adopts a fundamentally different strategy. We treat the fused target context feature as persistent contextual information and directly inject it into the Key and Value projections of every draft model layer. The projected features are stored in the draft model’s KV cache and reused across drafting iterations. This design provides strong and consistent conditioning throughout the draft model, enabling acceptance length to scale effectively with the number of draft layers https://arxiv.org/html/2602.06036v1

so it'd have some sort of kv

[stark needle]

holy frick

dflash was trash

i have 1.7M available context now

[fast pagoda]

mtp?

[stark needle]

ye

1.7 acceptance

[fast pagoda]

how speed

[stark needle]

wait im running out of mem

uh

lemme lower this a bit

[fast pagoda]

1.7m and a dream

[jagged turtle]

:omemga

[rigid snow]

i'm getting <0.4s latency with https://github.com/andimarafioti/faster-qwen3-tts (not quant, ~5gb vram) streaming. you have to be doing something wrong

[floral hawk]

That's what I figured. I switched to Pocket TTS and got practically the same effect and quality for no GPU use so idrc now

[rigid snow]

[floral hawk]

Yee good times all around

Can't wait for a stable V1, I'm so ready to not be coding 7 hours a day

I've been putting dayjob hours into Zero for over a month now

Alongside my full time

And learning japanese. I'm losing my mind (and my sleep schedule)

[stark needle]

according to this table

supposed to be ~42t/s

or osme shit

[fast pagoda]

now let's see paul allen's Qwen3.5-397B-A17B

[stark needle]

cant run

at 1M context

[fast pagoda]

well then what's even the point really

[shy axle]

what on earth could you possibly need that much for...

[rigid snow]

[stark needle]

discord bot

[shy axle]

thats seems... excessive

but alr

[stark needle]

im running a 750000 token prefill rn

[shy axle]

:

[glass flower]

you are doing.... what

[stark needle]

making llm process 750k tokens input

all at once

[glass flower]

i get that... but... why

[stark needle]

testing performance

[shy axle]

dam

[jagged turtle]

"hello everyone welcome to my benchmark today I'm going to be feeding 750k tokens at once to a model hope you enjoy"

[stark needle]

this was performance at 100k token prefill

Avg prompt throughput: 10000.9 tokens/s

i can prob optimize this tho

[shy axle]

what sort of god pc do you have

[stark needle]

2x gb10 connected with 200gbe infiniband

[novel flare]

Hi, any tips for understanding better js dom? It is difficult for me to make a logical bridge between the element that I want it to do things and the code I'm writing. Should I think from a specific perspective?

[ember hound]

just think it as a box, u tag an elem and just throw it in, if u want to retrieve a specific one u reuse the same tag

[shy axle]

[amber portal]

hmmm

[shy axle]

hey since when did you come here

[amber portal]

since now

[shy axle]

but why

[amber portal]

why not

[shy axle]

fair enough

[amber portal]

good boy

[shy axle]

OK BUDDY

[amber portal]

hehe~

boddy

[shy axle]

no embed perms here buddy

cant use ur message logger

[amber portal]

are you sure

[shy axle]

wait what

[amber portal]

who said i was going to use it

[shy axle]

you usually do...

[amber portal]

yeah but thats in a private server in public servers uhh

i think its against the rules?

[novel flare]

hi wlc

[shy axle]

hi

[shy axle]

get on jjs

[amber portal]

why

[amber portal]

helo

[novel flare]

hi but it was for him xd

[shy axle]

"Erm this is the programming channel, move this to general"

insert more yap here

[amber portal]

good point

[rare bridge]

at least you're familiar with the rules already

[rigid snow]

update: it works. for prompting tts emotion - insane. great. added latency for tool calls is kinda ouch but i like it nevertheless. need a description -> facial expr system next

[trim valve]

its so hard to tell if my pc is running slower because of matlab or if discord is just incredibly slow to use

[jagged turtle]

both are not mutually exclusive

[trim valve]

true

[ember hound]

maybe clean some dust?

[trim valve]

nah it comes and goes, so who knows

[nocturne olive]

I have found the real reason Wayland is unusable for me
Wayland does not allow choosing what GPU applications and games run on using the convenient set of environment variables

For whatever reason

[stark needle]

erm

[tender river]

average cs grad

[rare bridge]

none of them win, the claude instance they had secretly participating got the job

[opaque wharf]

Neuro and evil got a job

[rare bridge]

vedal docks their pay tho

[unreal notch]

Anyone here

Like a expert on html or website making?

I got a homie here like with Alot of questions

[fast pagoda]

$0-$10 they must pay the shellac ATM a deposit

[nocturne olive]

The person below you in question

[rigid snow]

afunyun html expert

[opaque wharf]

Can confirm 100%. He is HTML expert

[nocturne olive]

Created NeuroSynth website

[stark needle]

a

fun

yun

[sage crag]

child

[ember hound]

only 100? man this is tutorial level

[opaque wharf]

@fast pagoda what benefit did you get as konii's adoptive child?

[stark needle]

🇦 ➡️ 🇫 🇺 🇳
↕️
🇾 🇺 🇳

[civic flax]

You think Starforge would be a great investment?

[opaque wharf]

As in, investing in the company?

[jagged turtle]

why tf is my git folder taking up 26gb of storage

[fast pagoda]

once every 6.7 years, konii bequeaths upon me one nugget of wisdom regarding what yuri to watch next

[fast pagoda]

you committed binaries or models or something thicc

[jagged turtle]

actually now that you mention it it's probably compiled binary files

[opaque wharf]

That seems like a good deal

[civic flax]

buying a pc

[opaque wharf]

Are you comfortable opening up electronics in general?

[trim valve]

now you get the joy of learning how to unfuck a git repo

[jagged turtle]

?

[civic flax]

absolutely not, i am having way too many issues with my pc, i tried using obs and the cheapest and less stressing methods for vtubing and that did not work

[jagged turtle]

oh actually it's probably also a shit ton of python wheels

[opaque wharf]

rm -rf ./repo && git clone <repo>

[civic flax]

the most upsetting thing is that this laptop is brand new

[trim valve]

oh wait i imagined a .git folder

[civic flax]

so i am running out of options, i updated all of my graphics cards, did everything to the book, and nothing works

[trim valve]

I thought you were comitting all the binary files lol

[civic flax]

this just feels unfair

[jagged turtle]

I'm not stupid enough to commit build artifacts

[opaque wharf]

Then I think not. Buying pre made will not solve your software issue

Their after sales will only get you so far

[civic flax]

this is what it see

i keep seeing that it is a gpu issue, and since this is a laptop, it is a dead end

[trim valve]

try changing the capture backend

[civic flax]

how do i do that

[trim valve]

in source properties

[civic flax]

how do i get there

[trim valve]

double click on the source named "leyend"

[civic flax]

okay

[trim valve]

idr the exact setting name, im not at my pc

but you should have a dropdown between a few different capture methods

[civic flax]

i did the allow transparency since it is recommended everywhere

[trim valve]

scroll down a bit in those settings

[trim valve]

oh wait try window capture instead of game capture

tl;dr delete that source and make a new one using window capture instead

[civic flax]

you mean display capture?

[opaque wharf]

No, window

[civic flax]

wait i found it

it was down

[trim valve]

https://obsproject.com/kb/window-capture-sources

[civic flax]

alright, what would be the next step?
This is the most progress i have ever done so far

[trim valve]

is that what it should be displaying?

[civic flax]

yeah

[trim valve]

then just click ok

[civic flax]

what about the green screen?

[trim valve]

you should be able to get rid of it with a filter

[opaque wharf]

You filter it with different menu later

[midnight sigil]

color key is what you're looking for

[trim valve]

open the filters for your source

then add a colour key

[civic flax]

filters...

let me see

BOT

[opaque wharf]

Oh nyo

@rare bridge

[trim valve]

change the colour type to "custom colour", set the colour to the one the app is outputting

[opaque wharf]

@fair cipher

[sage crag]

brand recognition

[civic flax]

is this in OBS or VT

[trim valve]

you should also be able to lower similarity and smoothness as low as they go

[trim valve]

obs filter settings

kinda like this

[civic flax]

okay okay

[midnight sigil]

Acri theme user

[trim valve]

oh obs theme

[fast pagoda]

why am i so weak to these

[opaque wharf]

Wtf is even that bro

[sage crag]

thats grandma

[fast pagoda]

an unavoidable click black hole

[opaque wharf]

Turbo granny? The urban legend?

[civic flax]

Oh sweet it worked

Im gonna puke of stress

[fast pagoda]

shiiiiiiiiiiiiit

[trim valve]

[sage crag]

they say that the first thought that comes into your head should typically not be shared

but

[trim valve]

slightly surprised that you're only getting 29fps out of a max of 30

[sage crag]

whatever that is, it looks like a jaundiced ugly feraligatr

[opaque wharf]

See? Buying a new PC wouldn't help

[civic flax]

👀

[sage crag]

[fast pagoda]

10 months of effort

[civic flax]

dont be mean to him, he is doing his best

[sage crag]

i will be mean

[civic flax]

one of the swarm members helped me on doing a 2.0 model with better colors

[stark needle]

this u?

[sage crag]

it took a similar amount of effort for you to be created

[fast pagoda]

[civic flax]

i owe so much to the swarm because i got a laptop from them, I think it was R who helped me, and now this

I was so losing hope

[fast pagoda]

10 months damn

[sage crag]

ye

lazy

[stark needle]

wtf based

[opaque wharf]

Go on, use this

[fast pagoda]

sorry for your loss

[stark needle]

swarm

[sage crag]

youre adopted, it not my problem

[civic flax]

Actually a month

[sage crag]

[fast pagoda]

you were just staring in a window or

[sage crag]

adopted

[fast pagoda]

[civic flax]

i am starting, now i need to figure out how to make the other things to work

also, i do have an important question

i tried using auto configuration wizard and it failed, it told me to check on my internet connection but the internet connection works fine

[opaque wharf]

OBS auto config?

[civic flax]

and that is why i said that it must be a gpu issue since everything is updated

[civic flax]

yes, in obs

[sage crag]

config wizard

[opaque wharf]

The wizard is not very smart

Needs to go to school again

[trim valve]

if you click tools on the top bar it should let you rerun the wizard

[sage crag]

id say, trust that its a network issue

[civic flax]

that is the strange part, i updated everything, and the issue persists

[sage crag]

[civic flax]

is it necessary to use autoconfig wiz or not

[trim valve]

I barely use it

[sage crag]

[civic flax]

like, do you really need it

[sage crag]

[trim valve]

but only because I don't stream to youtube

[opaque wharf]

Nope

The wizard is not the brightest

[sage crag]

its convenient for selecting a bitrate for a streaming service

but

[trim valve]

you can do all the config in settings

[civic flax]

i see i see

[sage crag]

convenient doesnt mean good

for example

my phone is very convenient

however

it speaks only lies

[civic flax]

almost like neuro

[sage crag]

haunted

[opaque wharf]

Is that on mobile browser?

[sage crag]

ye

[opaque wharf]

I am more impressed they even thought about making the site responsive in the first place

[sage crag]

[stark needle]

bro @olive sable

new xiaoxin pad

comes with openclaw slop

[sage crag]

pluu

[civic flax]

Also, is it fine if bits of the green screen are visible or is it normal

[opaque wharf]

It's normal for irl stuff. But it can be fixed for digital stuff

[civic flax]

in vtubing, like, i shrinked him down and i see random green bits

[trim valve]

oops

you can increase the sliders I told you to turn down a bit

I forgot that rotating would happen

[civic flax]

how do i do that

[trim valve]

reopen the filters for your source

[trim valve]

drag the similarity and/or smoothness sliders to the right a tiny bit

[unreal notch]

Blud atp just make a call center lmao

[unreal notch]

This fr?

[nocturne olive]

This person

[stark needle]

:omemga

[sick owl]

https://fixupx.com/Teknium/status/2055698637800509692

Teknium 🪽 (@Teknium)

↩ (@steipete)
I’m sorry that you’re this desperate that you will take such an unscientific benchmark instead of any established one.
︀︀
︀︀Also qwen local is one of the most random length models there is with all its looping. And we smoke you all on quality benchmarks on every open model. Here’s wildclawbench by internlm, same speed on open models, much better results.

**💬 48 🔁 31 ❤️ 989 👁️ 33.8K **

FixupX · May 16, 2026, 5:15 PM

Holy clapback

[nocturne olive]

[rigid snow]

LMAOAOAOSOSOSIDIEO DHHSHSJSJSJSHSJS DJIEUXHDISJ??????!?

holy fucking delusion

[bitter phoenix]

Did you see his post where he spent over 1m in a month with nothing to show

'I must code but never ship'

1m just in code, damn

[rigid snow]

openclaw is genuinely the worst performing piece of software i’ve had the displeasure of touching ever

[bitter phoenix]

It's actually really bad, even the usage command print outs take time

[rigid snow]

open crawl

[stark needle]

Have y'all tried openclaw

I still havent

In theory my gb10 Supports the Nvidia nemoclaw thing

[rigid snow]

don’t

try hermes or pi or whatever else

[stark needle]

Im just gonna write my own harness

[rigid snow]

or that

[bitter phoenix]

I've played with it, I would say its most useful for automated things. Like checking the news for you or going over your stores reviews every week and emailing the important ones to you, if you don't have daily things an AI can help with it isn't worth your time to setup

[trim valve]

fucking windows

uninstalled WiFi drivers

and my trackpad drivers

[bitter phoenix]

I'm checking out the Aurora optimizer, on the left is Muon and its prefered batch/gradient accumulation numbers, on the right is Aurora using the same everything

[trim valve]

even better

it's not automatically fixed stuff this time

[kind nimbus]

There is also now Nvidia NemoClaw i think

[sage crag]

https://tenor.com/view/evil-vtuber-3d-head-turn-concerned-gif-15495186653811856266

wrr

[minor sail]

Is there a decent coding-optimized llm I can fit into 16GB vram? Most models I have tried are either so slow my ide runs into timeouts, or so dumb if I ask them to scan my code for todo comments they ask me where to fucking look, even though context is provided and/or get things wrong like trying to add vitest to a rust project.

[sage crag]

terraria NEW

[minor sail]

mind you, I was asking on a project using a new/customized rust react-ssr backend

[stark needle]

isnt loss of 9 suspiciously bad

for llms

as in "not converging"

[bitter phoenix]

These are new models pretraining

[stark needle]

still its suppsoed to drop to 3.8 immediately

[bitter phoenix]

They're tiny. It depends on the size of the model what it starts at

I'm comparing optimizers right now

[stark needle]

tiny model supposed to converge faster

scaling law moment

[bitter phoenix]

Onto a bad score, I would assume. It might be that the tokenizer for this one is too big, I actually tried out a 128k tokenizer and that started at 11 somehow

[trim valve]

istg

[bitter phoenix]

I have a new run for Aurora. I guess it's just unstable

[trim valve]

Im gonna have this laptop on Linux by the end of the day if it keeps pulling shit like this

[stark needle]

ive had gemma 24m converge to 2.4 loss

24m parameters

[bitter phoenix]

That's wild

I've had 100k models converge to 4

[stark needle]

ur doing sth wrong then cause

[bitter phoenix]

with heavy gradient accumulation

[stark needle]

3.5 is easy

u can do nothing and get loss of 3.5

[bitter phoenix]

I'm doing lots of random things

[stark needle]

whats ur lr scheduler

and LR

[bitter phoenix]

I actually didn't change that I think from whatever it was last
LEARNING_RATE = 0.0003
MIN_LR = 0.000003

The graphing slope and angle should be the same whether its doing good or really good

[stark needle]

gemma 2 120m on 70k training examples (which is basically nothing)

[minor sail]

does anyone know if intel gpus are decent for AI interference purposes?

[bitter phoenix]

And that's a pretrained model?

[stark needle]

wait lemme find the 24m

[stark needle]

this is from scratch pretrain

[bitter phoenix]

Ohh you're just using the same arch

[stark needle]

lemme find the exact params of the 24M params run

[nocturne olive]

All that matters is the VRAM

[sick owl]

If you mean inference then they're great value but you will find faffing around with the software a bit more annoying than Nvidia or AMD

[bitter phoenix]

Oh also actually, as in this is a tiny model; for testing speed, this model has a context window of 20 LMAO

[sick owl]

I'd still say the money to VRAM ratio makes them worth it if there isn't a competitively priced AMD offering where you are

[stark needle]

what is the llm supposed to learn with ctx of 20

[minor sail]

https://www.newegg.com/intel-arc-pro-b70-32gb-graphics-card/p/N82E16814883008 32GB for 1k is tempting, but I don't think I can mix-and-match with my 5070ti, could I

[stark needle]

at least do 512

or 384

[bitter phoenix]

The most common single word with the highest score

luck maxxing really

Higher context windows increase training time

[nocturne olive]

You probably can mix them

[sick owl]

I will day you can get the radeon pro 9700 with 32 gigs for around that price

Which would be faster and better supported

[bitter phoenix]

I could prototype on colab but there's brand new code, instead that's for longer term tests

[wispy pike]

so, this might be a bit off topic, but I'm surprised that growpramming from schlock mercenary became not only somewhat real (via trained AI models) in a few years after the concept was introduced in the comic, but also one of the most relevant parts of the field I chose to study

[sick owl]

If you mean VRAM sharing then no

[minor sail]

oh :(

[sick owl]

NVlink is proprietary and Nvidia aint gonna encourage you to buy competitors products

Its a shame

[stark needle]

@bitter phoenix try this

{
  "_sliding_window_pattern": 6,
  "architectures": [
    "Gemma3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "attn_logit_softcapping": null,
  "bos_token_id": 2,
  "cache_implementation": "hybrid",
  "eos_token_id": 1,
  "final_logit_softcapping": 30.0,
  "head_dim": 256,
  "hidden_activation": "gelu_pytorch_tanh",
  "hidden_size": 512,
  "initializer_range": 0.02,
  "intermediate_size": 3072,
  "layer_types": [
    "sliding_attention",
    "sliding_attention",
    "sliding_attention",
    "full_attention"
  ],
  "max_position_embeddings": 131072,
  "model_type": "gemma3_text",
  "num_attention_heads": 4,
  "num_hidden_layers": 4,
  "num_key_value_heads": 1,
  "pad_token_id": 0,
  "query_pre_attn_scalar": 256,
  "rms_norm_eps": 1e-06,
  "rope_local_base_freq": 10000.0,
  "rope_scaling": {
    "factor": 8.0,
    "rope_type": "linear"
  },
  "rope_theta": 1000000.0,
  "sliding_window": 512,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.53.3",
  "unsloth_version": "2025.7.8",
  "use_cache": false,
  "vocab_size": 262144
}

24m

(excluding embeddings)

[bitter phoenix]

Okay thanks, currently I'm recording data to compare optimizers. But I'll try out this out as soon as I'm done

[stark needle]

what gpu do u have

[bitter phoenix]

CPU only, I use my gpu for inference or it'll overheat

[stark needle]

i see

[wispy pike]

I'll get a liquid cooling system for my PC before putting AI on my computer ngl

seems dumb to only use a cpu to do this stuff

[bitter phoenix]

The only benifit is higher model sizes for inference/training if you have higher Dram than Vram, but worse latancy and everything else

[sick owl]

Claude usage limits

Someone needs to hurry up and surpass Anthropic already

[trim valve]

wifi back

still no trackpad

laptop officially much less useful until fixed

[fast pagoda]

they just fuckin rented colossus 1 and said they'd lift those

are they seriously out of capacity already

https://www.tomshardware.com/tech-industry/artificial-intelligence/musks-colossus-1-ai-supercomputers-inefficient-mixed-architecture-design-couldnt-be-used-to-train-grok-so-anthropics-using-it-for-inference-instead-musk-readies-unified-blackwell-only-colossus-2-for-frontier-training-and-potential-ipo

[sick owl]

LMAO

I knew they'd lifted Musks compute but the fact it couldn't hack it for training Grok is peak comedy

[kind nimbus]

meh i just went back from linux 7.0 to 6.18.31 LTS.
There seem to be some serious regressions in the i915 driver. My laptop screen just stays black and does not wake up anymore after idle on 7.0

[fast pagoda]

he used the rrat strat and just yeeted anything in there

[fast pagoda]

but what about experimental guacamole functionality

[bitter phoenix]

I've been using LTS kernels forever and haven't seen a single panic since

Zen is a nothingburger

[kind nimbus]

bro i was on intel xe for a while initially before i found out that i should rather use i915 for my hardware

ngl that shit cost me so much headaches. My laptop was hard freezing completely locking up the entire kernel randomly

i troubleshooted every piece of hardware before i finally found out that its the shitty xe driver causing that.

[nocturne olive]

NeuroSynth is here https://discord.com/channels/574720535888396288/1505584795856535592

[minor sail]

ugh, now I'm considering a motherboard upgrade to maybe buy two arc b70s to have 64GB of vram, but that'd probably mean an upgrade from dd4 to 5 for sysram and getting the same amount as I have now would be almost as expensive as the two cards

[fast pagoda]

on pcpartpicker rn my ram alone is so absurd it's just sad

[minor sail]

;_;

[bitter phoenix]

my selected ram is more expensive than the cpu

[fast pagoda]

maybe ill get lucky and someone will just come take me out

money has no meaning and i dont even know what i actually paid because of it

i find myself looking at like a $4000 gpu and im like

hmmm

not great not terrible

and then i have to remember how stupid that is

[flint dome]

just find a sugar partner atp or gamble

[fast pagoda]

i definitely paid less than 50% of what it costs now

[trim valve]

built in firefox backups are such an evil impl

the entire backup is stored in a html file