#programming
1 messages · Page 329 of 1

immn a collebcitble
mass produce
find jn srcarde

wah
mr slepy
memeeee sleepy
i think konii 2026 prerelease has a output token bug


i like moth
mmmoyh
NOT go away
@umbral thorn patch pls
nah it is been here for awhile, this release makes sure of it
oh elvy.
you pinged me too and i was thinking tf im gonna do?
popipopkoipo
moth (мотылёк) 
moth (моль) 
nOT note
mot
i like both
mottuh
equally
eat wool 
lamp
i wish i could
im just l8ke moth
envy
do you also eat wool 
moth, "night butterfly" or "mot"
mot is also slang for punching


that meme has been in my phone for 4 years. dont think i can find it
sun
i unfixabl
bwaa
ndiro yare yare
i guess we'll need to put it down then
ohhisee
ohhidintnsee

ttttttrrrobbbioo
oh no, wasnt fatal
i'm tapping out
now konii even worse injury
byebye
ok i may have overdone it a bit
#programming
Your message got 10 reactions

by
i have mnore vodka
improv 

intended for or likely to be understood by only a small number of people?
gatekeeping?
no not really it's like
imagine you invent a language to speak with your friends
that would be esoteric
an esoteric language
how would that be filtered tho?
what?
NOT abbreviate esoteric role play
why?
oh
hey
we were keeping it pg
except for the murder
is murder pg?
pg13 i guess
murder is like the most important part of programming
how else im gonna get that ram?
bank robbery
the way i do bank robberies you do need some
for legal reasons i havent robbed any banks yet
yet
Hey, how does neuro "see" the game in real time? Does it take screenshots and process them at an insanely fast rate or is it something else?
thats the way i assume it works. as vedal sometimes needs to wait for neuro to "take a look". tho its possible its just integrated into the main model instead of being a seperate model
I would assume that too if it was on a game where you can wait a long time before doing anything
what game are we talking about
But for a game like osu, that wouldn't really work
Osu
well osu is its own ai.
not at all related to neuro really
Wdym?
Is it not neuro that plays it?
the osu bot using "older" machine learning tech. its seperate from the neuro you know and see on stream or hear talk
Oh
👆
neuro only gets sent stuff happening in game that's significant (combo break, miss, fail, pass, song name, etc)
they do exchange some information back and forth. like neuro telling the bot to select a map and the bot sending back info about its current status or if something happened
Okay so let's say I want my ai to see my game and interact with it in real time
How should I start?
insert programming FAQ here
easiest way is to take a LLM that can do both vision and text generation. and then feed it a image every once in a while into its context
Alr
a harder but more reliable method is to make a bot from scratch and train it on the specific game like the osu bot works
thats how most "AI"s you see in videos and such play games.
like the trackmania one or the ones that learn to walk
Okay I'll try that
well for osu! specifically you can't really use an LLM to play it because it just can't react fast enough
slow down the game for it to work
That's unfortunate
AR -1000
but yeah a LLM can't play realtime games
The LLM vision systems I've seen are still terrible at figuring out where things are on screen
thats why the neuro integration recommends turn based games or giving neuro control over a higher level AI. like the driving in cyberpunk
How do I convince the llm that it isn't running on a Meta server but on my pc?
and thats also the reason neuro sucks at the quick time events 
It would basically just put " there are many circles on screen" in the context without any other reference point
you pc is now meta's property 
put it in its system prompt 
Is that in the model files itself or is it in a secret menu I haven't seen yet?
it depends on how you load the model. there is normally a way to set the system prompt. or also how you talk to the model
most models come with a default one thats probably what you are seeing
but yeah... AI stuff is a deep deep rabbit hole you can go down. its gotten better now with the AI bubble but... its still a complicated mostly DIY topic
"ollama run llama3.2:3b"
do /set system "here is the system prompt"
inside of the chat interface
Okay thanks
but also i would not really recommend using the CLI for this kind of stuff 
if you want a better UI to tweak stuff and maybe integrate tools with use something like LM Studio
thats what i switched to from ollama
I mainly use the GUI app
But I downloaded the model through cli
Apparently it's very slow compared to ollama
I don't know
i haven't seen a difference. and if you truly are after speed you can use something like vllm. or just llama.cpp directly.
I started with this ai shit less than 2 hours ago
Okay thanks
well... you got a long way then
ollama doesn't anymore i think
Yes I do
This also might be the reason I'm letting go of my precious rtx 3050
shit hit me like a truck just now wtf
oh right.. don't they like have their own version thats basically like 98% llama.cpp anyway or something? i remember a bit drama about that part. also that the gui is closed source etc.
Because the online ai unfortunately despises vpns
what
What are your pc's specs?
idk i remember something vaguely about that kind of stuff. but i also haven't used ollama in like a year at this point
Token usage: total=1,338,490 input=1,095,096 (+ 15,325,824 cached) output=243,394 (reasoning 207,552)
i got a 4080, ryzen 9 5950x and 64gb of ram
Now THAT's some usage
Wow
+15 million cached is concerning 
I have an rtx 3050 (6 gb ver.), ryzen 5 5500 and 20 gb of ram
15 million 
It's mostly MCP decompile output I think

6gb for AI is yikes
I can sell it for the same price I can get an rtx 2080 ti with
running any fairly good and competent model will be difficult 
you could probably run MOE models tho
Don't worry
The 20 gb ram is there to handle overflows
i took that into account
where in gods green earth are those the same price?
Oh
i have 6, you can run a 12b crazy quantized
I mean they're both around 150 euros
with a tiny ctx window
MoE is decently good for your usecase
qwen3-next for example takes like 40gb of ram if you use the offloading mode in lm studio (tho im still waiting for the slider...)
If i read the prices correctly
Oh wtf
qwen3-next-80b is like 50gb on disk
Turns out I HEAVILY misread the price
i would honestly recommend maybe something like qwen3-4b-2507 pretty good model for the size and doesn't take much vram. you can probably run 32k context without offloading on 6gb
Or atleast got it from the wrong source
this fits in 6gb and is just lobotomized enough to make sense
other sites sell new, unopened gpu (20 series has been out of production for many years)
best deal for ai has been 3090 used for many years
you want used gpu for anything not in current production
only problem with 3090 is you need a good enough psu
Theyre all around 300 bucks
the least i saw 3090s for is 450 ish
I meant 3060 and 70
cheapest 2ndhand 3050 here is 180. cheapest 2ndhand 3090 is 500
and the 3090 is so much better than the 3050
surely the 3050 price is an outlier
Their ads have been getting worse every single day man
ublock origin extension:
does 3050 even make any sense
no
like for anything
it's always been shit
If only that existed on phone...
Ouch
Yes
brave browser app:
Osu
(comes with adblocker)
gpu for osu 
what
gpu for av1 
2080Ti only worth $180 at most imo
Aka Windscribe™'s POWERFUL custom DNS solution
Not only does this beast of a dns have a cool name
"POWERFUL" so it's just like tha adguard app or that other one
windscribe
yes if you got it as a gift
?
It also has dns level ad blocking!
i've literally never heard of them
i don't need you to read their ad to me
thats a name i havent heard in years
Its not their ad
Its a custom generated one
By my braincells
Yeah I bought their build a plan thing
It's not bad
thats how "cheapest" works tho. it is specifically less than the avarage
But not perfect either
did you hear about it from an ad or from a product endorsement
3050 only worth $80 imo
I honestly have no idea where I found out about it
i mean "outlier" in terms of bad value
I got an entire pc for 150 euros with an rtx 3050 inside
And 8 gb ddr4
- ryzen 3 2200g
Which i fortunately upgraded
clearly doesn't work if you're still seeing ads lol
damn
It wasnt on
scheduler pretty shit on it and its bandwidth bottlenecked so it only get to utilize 70% of its theoretical raw compute
that's fine
If I turn it on it'll just show me offers from london
actually i found a 3050 for 145 
still nearly useless gpu tho
Wtf
Why would they do that
idk
moeny
Is it possible to flash a custom vbios on the gpu?
that's why i said its pretty bad of a gpu especially when its sometimes overpriced
Because I'll gladly do that to enjoy this gpu to its fullest
it's not very powerful then lmao
proper solution will block all ads, not just localize them to somewhere else
I mean the whole point of a vpn is to spoof my ip
The thing itself is a vpn
The ad blocking stuff is the dns
That's included with the vpn
it clearly does not block ads
It does block ads
least confusing wording
ah
i've had facebook market work with locations that i am very much not at though....
3060 is pretty nice
how often do you move lmao
I don't
then it's a thing you set one time
Oh
idk man...........
Does it not change depending on your ip?
Man I wish I had a decent 3060 with not half memory and probably memory bandwidth
I CBA to pay for it tho
I don't use it enough
i always have a tab with it open and with the location set, it's a property stored in the url
i'd take that 960 ngl
if you don't like keeping tabs open, you can just bookmark www.facebook.com/marketplace/[YOUR CITY HERE]/search
should work
Alright
cheapest i could find:
3090 is 500
3080ti is 500
3080 is 300
3070ti is 200
3070 is 160
3060ti is 150
3060 is 165
3050 is 145
steep drop at the 3080
not even for parts wtf
that's a deal
unfortunately i'm not in latvia lmao
3080ti isn't reasonable its $350 at most
couldnt find it lower
I just remembered that my friend has like 2 rtx 2080 tis just laying around
but its close to 3090 performance and bandwidth, just with half vram
Are they a good upgrade over my rtx 3050?
should be
3070 for $160, i'd take that's the sweet spot
you're only missing out on av1 decoding
2080 seems like ~50% better than 3050 8gb version
iirc 3050 is 1660Ti level so its even slower than a 2060
its euro, so 188 usd i geuss
close enough
true
Quick question, are you from Belgium or the Netherlands?
in terms of performance 3070 is best, but it only has 8gb for some reason
belgium ye
Oh nice

I noticed your username that's why I asked
3080 10gb is worse in terms of vram per compute
3080 is like way faster than 3070
if we focus on vram then 3060 would be best i think
but only extra 2gb vram lmao
yeah
honestly, best vram per compute would be K80
then P40
but you don't want "best vram per compute"
are you mayhaps from the low countries since you ask this?
I am from the center of the country
Yes
i know you're programmers but surely there;s someone here with a not awful taste in design
i remember i said like yesterday instrument sans is a nice font and everyone whoichose to comment said it was bad
and i said context matters
i think this is nice and has character and personality to it unlike all uses of this font like veryone does with vibecoding and shit it;s the new cool font
even the us govt "gold card visa" website uses this font which is cool it looks nice at least
but it's become very overused
i think i cooked
don't look a tsong card design too close i literally just did those just so i have some context
this isn't like extraordinary but i think it looonks nice
@signal beacon which version of 3050 you have
The shitty one
6 gb

hmmm
What are the drawbacks of a quantized model?
mltncandy
3060 6gb is unbalenced because it is in a laptop, no other reason 
ass?

doesn't look that bad there
But it would look bad depending on where it is
rip
And that is the exact reason why you should use the Samsung Sans font!
3050 6gb is literally just 1060 or slower
the WHAT?
samsung has a font
call the police
you know what i meant by the
"gold card visa"
go slook there that's how it's used everyuwhere now
hype it's literally overnight, maybe the font dropped on google fonts and THAT's why it's now everywherrt? anyways go look
that's how it's used everywhere
Get back here mr.sung
says you

i have absolutely no clue whay a gold card visa is
If i had image perms, id show y'all how beautiful it is
i thought visa wa sa pasport
us president card
wht
gold card that acts as living permit
Very expensive visa bypass card
ye
the brother himself has a special colour?
needs to match him being made of money 

i actually think that website looks kinda nice
but still overused font
and subject is
ye
unless they changed it since
i haven’t looked since it came out
ye still uses the font
They could be both
If they work then they work
honeslty im more focussed on the "fuck u skrillex you think ur andy warhol but ur not!! ❤️ "
6gb modules exist
LISTEN TO IT MIXTAPE IT"S SO GOOD
send link
Hey guys is mixing different ram kits with each other really that big of a deal if speed isn't a big deal?
its kind aa big deal
at the very least match the clock
Okay what are the disadvantages?
ddr4 is quite tolerant with that but since you're on am4, could be a big deal
though if you don't liuke bass music maybe it's not for you? idk listen anyway
They're all set to 2133
timing and spd mismatch
Because one of them has 2133 max
sometimes can't even boot
it will run everything at whatever is the worst of the 2 sticks
Are you implying that it may or may not boot?
There are 4 sticks
best to find identical dimm
3 from corsair
can you sned the model numbers maybe?
if the timings are somewhat close it could work
1 from samsung
My pc is off rn
if the 3 are the same set use the 3
So 2 of them are identical
oh
1 is different
use the 2
Yes
dude we don't know wtf is black or red lol
the 2 in slot 2 and 4
specs matter more
I did that
okay
Yeah the point is that it's 2 and 1 and 1
shouldn't kill anything even if you try the mixed ram setup
just refuse to boot with it
yeah it shouldn't kill anything
2 sticks is bound to be better than 1 stick with better specs. unless the 2 are heavily outdated
can remove the offending stick if it doesn't work
if you're lucky it could work
give it a try
oh
did it boot?
Yes
also what are the sticks
With absolutely no issues
Ddr4
capacity, speed, timings
2x4gb 1x8gb 1x4gb
uh
These won't even run at 3000
enable xmp
and they all run at 2133 together
POST fails if I go over 2666
seems fine, if a bit cursed
Yes
aight you know what
slow ass ram
keep that
that one was funny
i wonder why it wont post at 3000?
maybe needs 2T?
i am a firm believer in "ram speed does not matter"
what cpu is this again?
2133 is fine
Ryzen 5 5500
2666 is so fucking slow for even 2020
- shoudl be same memory controller as 5950x no?
5th gen is new enough it shouldn't mattee yeee
if the timing is tight its fine
no
It's just not very fast
350

5500 is not vermeer 
Its cezanne
eh real world perf 5500 is almost the same as 5600
according to my very trustworthy source as google ai sumamry. cezanne should even handle higher clocks than vermeer
I had to bios update and ages update for it to work
My memories are failing me, are the asrock boards garbage on AM4 or AM5
It should
fucking clanker misinfo
i mean memory clocks
Should I just buy an Intel arc b something instead of an rtx
They have like 16 gb of vram
where did I even hear this from, looks like non issue surely
even for gaming b580 is decently priced among brand new cards
and for ai chances are you DO
What would make me want to use/need cuda?
cezanne seem to be pretty good with memory overclock
a lot 
would try actually i have an am4 board and samsung b die 2x8gb kit
that kit is worth like crazy nowdays
might wanna try hitting 4933
suppsoedly the chiplet design or whatever causes vermeer to be slightly slower?
how in the frick is it $250 in this ecomony
used is higher priced
because intel is based af in the gpu department
B580 used is not it.
only could find a full pc for 2.2K with a i7 265kf and 2x16 6000 cl36 ram
overpriced to shit
overpriced as shit
just get the 13950hx for cheap
$220 nowdays at most
well, engi sample
but i always had a really great times overclocking these engi sample
honestly
embed failure gif
i could send you a 13900 es if you're willing to pay for it + shipping
dont buy brand new
2dehands
Yeah i know I was just looking
But what the actual fuck bro
its about $210 rn
400 bucks for that?
the new gpu prices from official vendors are always so silly

they treat that shit like retro game cartridges or something
13900 es is top 1 best SFF cpu build imo
for value
Theyre reselling the reseller's offer
"this isn't in production anymore, that means the supply is limited and we can charge more for it!!!"
i think thats an upgrade over my 5950x
24 cores over my 16
i don't understand why they do it
ehhh
8 p + 16 e cores lol
Because 💰 🤑 💸
but surely nobody buys it at those prices...
Unfortunately in economics, businesses don't give a fuck about you and they only care about profit
I mean i understand why but it's still unfortunate
the p cores are 30% more supposedly
the problem is getting the other parts
You'd be surprised
Honestly out of all of the :new-leaf: experinces this week, this has been one of the better ones
only on k models :(
but 13900 es is hella power efficient
ah
if tuned right
you can run it on sff cooler
at 200W
and it'll still just hover at most 88C
the ihs is good
Translation: Sold 182 times
price changed over time
lmao 182 people got scammed
Yeah but still
but yeah 3050 is a scam anyways
What if this didn't change price dynamically
e-waste card
182 idiots then
Yes
even 5050 gets more respect than 3050
but in my home town 3050 is the default pre-built card. often paired with 13700F
I completely forgot about that card
How much does it cost
shittiest pairing
$200
i wanna throw up
for 50 series you buy new
Z790 mobo is 200., so then id be half a thousand in for not that much performance over my current pc
if i get it as a whoel new pc and use this as a server then id be paying like 1300 + kidney for ram

idk
just get a B760m
you don't need good phases for 13900 es
80 bucks
its locked, and i doubt you could figure out how to unlock it
locked to 5.2ghz
1.2v

at most its only drawing 220W
any 8+1+1 60A could do
just get the one with vrm heatsink
the 13900 would most likely be drawing at average 20W too lmao
i think its just not worth it rn unless i need a 2nd pc
depending on your tuning
you can make it run at only 5W
while basic browsing and discord
and youtube
i hope you're listening i listen to over hafl already
its a very efficient chip
i was waiting on a link
its the exact same chip i used to argue against the "arm is more efficient" tech bros
ah
which is fun i think
cant wait for my 13900 es phone in 2 years
(iphone)
might wanna put that in my todo list
i9phone
i could probably build a brick phone ngl
pico itx or whatever thats called
powered by 1C lipo
idk what are you doing with it
Hosting it locally
lcd screen, (igpu?), 13900 es, noctua nh-d15, 15Kmah battery
And fine tuning it
perfect combo
Messing with it
then inference, it'll work
needs cuda most likely, nvidia
Doesn't amd have it's own equivalent?
Alr
yeah it's about how much you're willing to pay to not deal with spending time working out how to do something with your specific setup
instead of the more popular option
22K mAh
for lipo
dj smokey recommended that i listen to the album from start to finish. so i guess i have to
i like how the songs have seamless transitions
uh huh
I mean i could find a way to optimize the llms even more but that would take a very long time since I have absolutely no knowledge when it comes to coding
feels like the kind of songs i shoudl play on the big sub, but its 5am and im on headphones
also i'd just get those sff cooler heatsink and strap smaller 40mm fans on the side

website says "give this command in shell, see it run"
nvidia card in machine
you do it, it runs
amd card in machine
you discuss the output and the errors with 6 different discord servers and 3 different llms for 2 days until you figure out how why you can't have an easy workaround
Oh
Wow
i live alone. envy
i listening on speakerds
"no knowledge when it comes to coding" and "mainstream support is not there yet" don't mix well
Galaxy Collapse by Kurokotei followed by a random Camellia song chosen by your local AI
Yeah i see
the moment i graduate i will move out of my parents house cuz i got that social security 
not unless you're actually intending to stress yourself out for its own sake, but if you just wanna get it done and spend your efforts on something else then yeah
hey if it's LLMs you wanna work with, you don't need C++
all the higher level stuff you work with is in python

Okay
you may need other languages if you're diving deep but frankly you don't need to unless like, that is your actual job
i like the part where it goes bzzzzz br br br rb bzzzzzzzzzzzz br br br br
if you just want to train and use DL models, it's all python
How long does it take on average to learn the basics of python?
you're most likely writing the wrapper at most
a week?
python pretty easy to learn
can be done in python
a few days if you're dedicated?
Oh wow
im afraid i must tell you, that is C++
That's very fast compared to other languages
like, the amount of python you need to know to do deep learning is actually not that intense
you won't be doing wizard shit
No i know std::cout is c++
oh
I meant the language needed for llm
ye i was thinking all the actual inference or underlying code not python
you can do just fine with like maybe pytorch
python uses C libraries and stuff for it no?
if you understand basic programming structures (ifs, loops, functions, classes), can download and import libraries, and do some basic Input/Output
you can understand an AI training and inference script
Wait that's it?
yes
WAIT
the whole point of machine learning is that the machine is doing the learning
what did you think it was
you're not going in there and coding the weights on your own
that shit is all on the training
That actually made me remember a lot of stuff I learnt a long time ago
ye?
its math and shit
you do some fancy highschool math
From the android kotlin docs
a training script for an AI model can be like, 50 lines of python or less
I had no idea what it was honestly

"squishy clip" is an aptly named part of the songs. i wonder why it sounds like a spring?
audio nerd please explain
Would it be a good idea to make it remember a specific code or command that would toggle training mode?
NEW EMOTE
rich people in your area: i want hand-tuned models hand-tuned by tuning slaves
it's not so much training "mode" so much as like

and you'll have somewhere where you define the model architecture (or you can literally just download an existing one)
and then you define the training data (literally point to a folder and a .csv with labels or whatever depending on what kind of model you're training)
I shall make TRM have common sense
and then it's quite literally model.train()
tldr you don't need to be a vehicle engineer to drive a car
unless you want to be
like a week idk
it doesnt take very long
like don't get me wrong, you can get deep into the weeds if you want to, but most of that is done for you when you do "import torch"

Well I'm gonna need some docs for this lol
- pip install whatever the frick you want
- go to a file
- import whatever the frick you want
- you're like basically done
Im looking for a simple example on github
I'd link my own but that's automatic dox on me lol
torch just works better on nvidia or smthn?
an epoch is like an iteration of training; it'll run whatever number of epochs you set
there are some parameters like which optimizer you use, learning rate, etc.
but those honestly don't matter as much as your training data
you'll see the DataLoaders there which just define and handle loading in the training data and labels
a library library
or maybe a basic tutorial will be easier: https://docs.pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html

Unfortunately
that also shows what defining model architecture is like
that's how you define model architecture
basically you're defining what the layers are
but what everyone does is they use an architecture someone else came up with
i used to let people use my gpu for training, but since then ive been sent ai generated training code which i had to help fix so i stopped doing that
a training script is usually simple enough that a competent LLM can generate it without issues (usually)

ofc the problem with AI coding is when it isn't the usual so
anyways point is, if you have basic python and motivation, AI work is doable
don't feel like you need arcane super hacker programmer knowledge
you also jsut kinda learn by doing it
A part of me wants to see how well opus does at implementing diffusion but I don't think I care enough to check
so do it
I've had Opus do well when I just do something like "I want a training script to train a classifier model. I want the architecture to be ResNet. It will take an RGB image and a .csv file as input. the CSV file will label whether the image is a hot dog or not a hot dog"
Bed too comfy for ml silliness 
oh, and if you see the word "tensor" a lot, don't get intimidated. it has its own meaning in math but in ML it pretty much just means an array of whatever dimensions
resnet is simple enough and probably all over the internet so makes sense I guess
like a 2D array of numbers, a 5D array of numbers, etc.
yeah, Resnet, UNet, the popular simple ones
usually no problems
FPS dropped by 1000 because of the addition of SpriteMap. 
Make it a vector
Wait, different map huh




because it's a copy
i know right i wanna drop my stuff like that too
but it's not sincere enough
that's why i don't
like i have one
wdym with sincere?
also the genre i make doesn'[t reallu appreciate sincere
putting my emotions in
like
you know
if you did anything aritstic
you'd know
it's a way to eexpress
not just create foir the sake of it
if we come down ot basics
to relieve yourselves of emotionb
and if you get strong one
you know what
check dm i feel i;m smashed enoght
#programming is where people talk about everything
google says i have
"Alexithymia"
Is there a guide on drawing pixel games?
guide on drawing assets in pixel games
dumb question in hindsight but i misread the original question
nix config is love, nix config is life, nix config must be cleaned
the config must grow
Samsung belgium just advertised to me how i can use gemini banana nano to generate my dreams.
Why they advertising other people's products??? You make phones samsung
Never not laughing at calling an image model banana
samsung and google are friends
lmao
https://gemini.google.com/app everything that has internet browser
:bih:
Vision built into a model can be nice but isn't great. Usually it's no more than one layer of the LLM in size. Plus it explains turning on and off vision, most likely a tool call that he can enable
Rag is a bit simplistic but generally memory is one of the most difficult parts. For TTS, XTTS_v2 is one other option and piper if going for English. For api open Ai compatible endpoints plus a hell of a lot of others as needed.
Haven't touched vision yet so can't say too much
For all: no but it involves a strong liver and lots of alcohol/determination
Depends on the method and level of quantization, between slight variation from unquantitized and lobotomy.
I think RAG is a very fair assumption though, especially given how he's treated their memories
like being able to just bring them up and read and modify them casually
Well yes but there is more nuance to it, the "upgrades" to memory are likely not just hardware but something like embedding management and vector DB matching
RAG is an incredibly wide net
well yeah, but that's why "some kind of RAG" is a very safe bet
when people ask I usually just say that, but any more detail than that would be hard to figure out from what we've seen
I from when he does talk about back end stuff it really does seem like he's not trying to gatekeep so much as phrase things in a way that people who are genuinely interested and capable of doing something in the same general concept would understand while not wasting time teaching people to make slop. I honestly think there is enough info to make something in the same general vein so long as you understand you are probably going to find yourself regretting your decisions yet continuing anyway

When I first started getting seriously into this stuff I almost pulled a Vedal and bought Gpus that almost didn't fit in the case I had at the time
im still baffled with how some people make the case clearance the biggest concern lmao
Wait I just watched a clip from today and i have to say discord update fucked voice.
triple gpu except the third one is on top of the cpu cooler
its a mess, but sure as hell does it look cool and somewhat satisfying
I can say that pycord dev build works if you manually add a keep alive packet running
well it's like "how they built the pyramids". People think the mystery is wondering how it was possible but it's really more we don't know some details on the exact methods used
we have plenty of plausible methods that could've worked
it's just hard to tell the exact one used
I spent literally 6 hours yesterday debugging pycord
Decided to re-add my 3090 to my PC for the lols and was almost foiled by PCIe slot spacing
Me trying to cram 2 3090tis and an A4000 in one case
In my previous rig in a Corsair 5000D, the 3090 was basically touching the front radiator
Wait is that a 7000D
Damn them blowers are thicc
The 480mm is 125mm thick including fans
Is this better?
Token usage: total=938,132 input=736,076 (+ 9,028,480 cached) output=202,056 (reasoning 160,384)
maybe
Btw i really just want to know if i could re-sell the pieces that arent fucked up
Nothing really specific
This is a better place for tech support. So answer a few questions;
- What is the PC, is it a laptop Nam eand brand.
- Whats the specs.
- Are ou able to get into Windows?
1- its a pre-buuld tabletop computer from pichau
2- i have no fucking idea😭
3- it doesnt even finish starting up atp (and it was turning off by itself before)
You want to daignose the parts to make sure they aren't broken. If you resell the GPu and the GPU is broke th eperson will want thei rmoney back
Fr
If it turning off buy itself than the PSU is fucked, if you rpeace the power supply than t wil more than likely work again
I mean the tech said it was that our cooler wasnt cooling anymore
And i do live in a very hot country
Although i kinda forced the pc to work by using a fan
And it did work (for some time)
Now i ferl like something has melted inside it or smth like that💔
Also what is a PSU
Dayum
Power Supply Unit, its what delivers power to all the computer parts. Its th emost common fail poit on a computer. But if you ran it with the CPU cooler that didn't work that its probably jsut fucked, I'd recommend buying a new PC and trying to salvage what you cna from that machine to resel
I did it for 12 hour sonce, my house was warm
Yeah that was the original idea
Well, eitherway i wont have a pc untill christmas next year💔
My yiutuber carrer is fr dead now
*youtuber
Wouldn't a PC with a broken CPU cooler just throttle and shut down on its own before any damage?
At which point it's a cheap and easy fix
I forced it to work by using an external fan
I was fr forcing it past its limits
Yeah they ghetto rigged it. It could still work with a new cooler.
Are you saying it's still working right now with an external fan?
Was*
It doesnt even turn on anymore
But a slap a Hyper 212 in it
And it was getting insanelly slow so
It’s like 20 bucks and can mount on any socket
Idk i really feel like it isnt worth it
kitchen just a bit warm
Uhh not in brazil unfortunatelly
wasn't full blast thankfully
A stock cooler still shouldn't be much even in brazil



konii powered by choco



why would it




