#programming
1 messages · Page 387 of 1
```cpp
int main(int argc, char* argv[])
{
#ifdef EMSCRIPTEN
pass
#else
//VulkanEngine vulkanEngine;
/if (!vulkanEngine.initVulkan())
{/
std::cout << "Failed to create vulkan instance\n" << std::endl;
OpenGLEngine openglEngine;
//}
#endif
return EXIT_SUCCESS;
}

ye it just says that
hmm
Vulkan was a lie all along
it failed to create a vulkan instance cuz i activly made it unable to do so 
toast look
I saw 
i invented digital 
also, it was doing 40% of my vram before, and that went away and i don't know how or why
"minor lag spike" to begin with but it is fine 
Man I'm so happy I don't have to write makefiles anymore
@olive sable
holy nails
uh sorry for having you repeat youself but
idk how to fix tho
what exactly are you making with this thing
"get better hardware" I guess
volumetric display
i wonder if that one thing i cant mention here would be good at doing this compute shader
Mabe? That has latency issues tho as we are learning

mf is networked to main system
nlohmann::json... 2 weeks ago i replaced our usage of it within the api of our app (serialization/deserialization ) with rapidjson, and had 3x performance improvement.
otherwise I think it might actually work well
obs seems to be unable to record 120fps
nevermind the 180 i was trying to do
hmmmm
welp, i set it to 1080p and then it was able to do 180fps, but i doubt anything will play the video at 180fps
I can get close 
i couldnt do 60fps cuz it jsut kept takign frames that looked exactly the same
so ruining the effect

i made the compute shader only go inbetween every 5 frames
should be close enoguht to what we want
I should try doing it again 
Nah, it isn't going to end well for me
i set it to vsync cuz my monitors are 180hz and 4225/24 = 176.041666
and your laptop might not die now
for this i set the buffers to 570x1140x570.
in terms of vertical resolution in the real device we will only need 570x1140x160
uhh
the problem with that is that the tutl model is too big to fit well
and i wanted it to use the whole screen
forgot to edit the copy pasta
there are like 5 different const in different files for htis
cuz i be scatterbrain
lemme jsut git push this
hmmm...
checks out
All of 3 frames now
its 24 frames, but cuz of hw rgb8 works, the least significant bit is less bright and such

Had to write create CMakeLists and vcpkg.json due to my hate of stuff like nix.
thats crazy
didn't say i wasn't
i pushed a version with depth of 160
should run better too
only uses 20% gpu now for me
@amber fractal try the new one
surely it'll do something
i lied im getting food
where tf is the resoution
"a bit"
better for me
(it didn't)
The fragment shader has width, height and depth
Same thing for the compute shader
Then in gl_loadgltf.cpp or whatever its called, there's something there too
Then
In main.cpp
Theres a function to make the SSBO
That has like 570 x 1140 x 160 or something
Just change those values to the ones you want

570 is 3.5 times more than 160, so 3.5 times more expensive
Roughly
every time I think I got all of them, there ends up being another one
I got nothing
keep changing values and nothing changes
no spinning boi for me until later ig

phrr
elvyn get off of koni's account 
let her keep the account, it is funny
camuel
Here are my guesses:
- We as a species can't provide enough compute
- We are not giving enough good data to train models
- We are stuck in a way of thinking only we know
- We are going down the wrong path and we can't stop it
petah, why are you running 200 claude code instances at once
good guesses
All could be valid reasoning, but without a needlessly accusative tone that the title/thumbnail have imo
mhm
While impressive at a glance, closer, and continuous usage proves the futility of the project
I suppose I could speculate all day what they're arguing, or I could just watch the video lol
If it isn't any of them, it might just be ragebait and clickbait
if it is any of them, the wording is engagement bait
well that's still clickbait but yea
engamement bait is the combonation of ragebait and clickbait
hmm
well I found out what he's calling the Bitter Lesson
Paper by some fellow in 2019
I'm curious
Did the voice LMs like eleven labs arrive before or after transformer LLMs?
Like to a capacity that they sound like a talking person
interesting
I kinda like that it's in there but there should probably be some explanation that you don't need it 
millions will install PyTorch
You are right in your 4th guess
(from that same video)
Yessssss
you know what I think
When the LLMs fall into that constantly-self-correcting loop bug, that's probably where you can see the gap that not having a "controller" agent presents
like, one such agent/entity would be like "ok, I need to stop here - I'm not getting anywhere. I must do some research"
Yep, one requires the soul
It is surprising how much we've gotten from this text predictor but we need something else
yea
-# Surely no LLM supremacists are on at this hour 
My charging module burned 
Question is though, how do you train such a thing, because it's an environment action rather than content action
Doing something, then moving onto something else involving a different device or medium
then returning to it
The first thing to start with is identifying it is wrong at all
yeah
needs to find a wrong state and understand it is in one
that probably comes from an understanding of actions and consequences
oh you're talking about valleys and hills?
idk how to translate higher-level concepts into math sadly 
neuro live laugh love
LLL
Slightly biased as that is my pet project
So basically what I wanna do is give desmos for chat, but chat can manipulate variables
and it makes neuro see what chat is representing on canvas
that's nice
neuro middle age mother when
Man I wish it could be translated into math, I sorta just didn't with my attempt
Let me do it for you
Give this for neuro
And let chat write new functions
limit max 100 functions
variables never change
they are statically changing
But chat can create something (depending on their knowledge)
Basically neuro will be the testing data and chat will be training AI
OK I watched the video to the most I could
All of my attempts are just bit based. So if I ever see a zero I'm cooked 
if (0) skip
It sounds like a pretty sound statement, an observation. I'm still not happy with the wording, or maybe it's the composition of that title with that paper
it's a concept
nah I need know when I'm cooked so I can no longer be cooked by learning new things.
if I get a zero result in the test model when there is supposed to be an output however 
if(0) -> warning
debug warning
Can't you inspect what's in the middle?
before output
I can give you the repo
if (0) +1?
but yeah I can inspect stuff just fine
so it is never 0
if it becomes 0
give it a little push
like +0.001
fake the data
close to 0
but not 0
never 0

thoughts?
in statistics you won't notice 0.001 if range 0-10
but it will not kill the algorythm
just slow down
right?
Man this readme is outdated
I'm sorta not using floats at all
Ohhh 6 bits?
Math would be nice, but I'm using a bitfield
not error -> warning "Error instead of Hallucinating stuff that doesn't exist"
That is called outputting a zero because case matching doesn't match
and you OR all matches together
Simple to intercept
Do you want it to be intercepted?
I just wish I wasn't in the middle of another rewrite because it doesn't work yet
In training I'll be intercepting, for anything else I haven't actually thought that far ahead yet as this thing has the very real chance of not being able to work at all.
I see
I never tried to stop it during training
Just killing
And making branches
I guess intercepting is just sorta just a normal training thing. You save your values so you know what effects it had.
Even in Transformers that is the process
Yeah save values aka make backup or quicksave
The cheat code with this is just how light it is, and the fact I can easily reverse the process to solve for errors.
In a general model though, I don't see it being useful. At least not until I stop being pessimistic about it.
It is a nice tool, but I have some changes to do before I start testing it for real.
If I really wanted to, I could scrap together a demo with the existing version. But I have changes to make
```
Question -> Sounds like someone has a bad case of the munchies
answer: Pancake Panckae Panckae [Neuros Ex wife] Pancake Death
question -> what's a cool spring day's temperature in farenheit
answer: huh, where's that? netherlands? meow. Death
Do you see how I multithread on one chat now?
I have multiples conversations in one chat
DALTOn (Server Shrine Maiden): Sounds like someone has a bad case of the munchies
[15:34]jmeel: what's a cool spring day's temperature in farenheit
[15:34]Huỳnh Pancake🥞
: huh
[15:34]Huỳnh Pancake🥞
: where's that?
[15:34]Huỳnh Pancake🥞
: Netherlands?
[15:35]AxieAxolotl [Neuros EX wife💊]
: mrrrow
[15:35]Huỳnh Pancake🥞
: 
[15:35]sixbytes: "Suffocated in a wall"
And this was the chat
Enjoy
This is what I encoded
This is how chatter sees my convo
It's fool proof
Any one has any advise to lern c??
w3school
Make something. Anything. From those project you'll learn a lot more than what could be taught via just syntax and documentation
yeah bro but if you genuinely know nothing, you can't just spawn code from the air
that's why documentation exists
There's a whole world of difference between knowing "alright, I need to have a condition, how do I do that in C?" vs "alright, this is how I do condition in C. When will I need it?"
phrr
have a goal to achieve with the language, then try writing it, then use docs to guide how to use the features
enub
memeslop
a = function local```
🥹
(casually just giving worst advice possible) (how to learn c)
Why would I do programming on my C:\ drive
What is a C drive? I only know /
for a particular recursion algorithm one is struggling to understand is it advisable to use pen and paper to work out each invocations
or is it too much work
I want to make a os for micro controller
Anything that helps said person clarify what really happens is not "too much work"
I made a kernal ish thing in micro python and like it's cooked it's a single file with 900+ lines
do it directly in your root
that's heavy
Thank you
Simple embedded RTOS with no memory protection isn't too crazy, but as a first project maybe.
Ya it was slow asf and like it outputs for a vga screen had a ok ish gui i broke the micro sd card I had the os
I'm not gonna stop you or try to impede you. If you're well-acquainted with programming in Python I think you already know programming fundamentals to try C
You kinda need to know assembly too for context switching
the algorithm is O(2^n) 
Depends if it's a cooperative RTOS or pre-emptive one
Thank I am asking sugetions from every server I am part of thanks agin
That's nice
You don't have to be afraid of wrecking the hardware I think
Unless soldering is involved
Speaking of wrecking hardware, has that happened to anyone? Broke the hardware through bad coding
It's a homebrew computer
nvm I recall my desktop PC mobo getting fried because of the OS on a faulty HDD 
Me i broke my esp32 it's a dumb way
Nvm it's a good thing you're asking for advice first
Yo shuld I send you a photo of it?
This is like day 2 of asking people for a week
I was gonna encourage you to dive into low level programming but on second thought
Did you say it's on some small board? Like raspberry size?
Ok it's with 3 micro controller a esp32 then a rasberry pi pico and a ardvino nano
It's a bit bigger than a rasberry pi 4
https://x.com/alpuhap/status/2018014969724379387?s=46&t=Kdlv6lZSSLbxxlifio9rlw it was always Linux, isn’t it guys?
Idk what i was doing here, but ig it's good?
If it were a genre, i'd call it haunted summer pop
Make sure you have spare sanity lying around
When it segfaults, blame the pointers
I don't have any left💀
all.
3rd point causes 4th. we're stuck with the text predictor language models. by extension 1 and 2 applies because the current AI models are scalable, but logarithmically in performance scale. we're just stuck with something not even able to have thoughts and merely emulating everything accordingly to training data and at the same time really power and data hungry
i remembered when all these "AI" stuff felt cool but now its just the same publications all over because the topology are always the same
"AI race" sounded so stupid and cringe. felt like the words spun around by the tech illiterate
there's nothing significant about it. its not the "future" or some shit
I know this is programming but do people know excel here
Nothing complicated, just matrix functions
hey can someone post the "I want to make my own neuro sama" faq
it's not pinned and that seems like a crime
is it related to the excel message or
this one?
nono
2023 is an AI millennium
we have a chart here somewhere
perfect

Cunninghams's Law in practice.
"the best way to get the right answer on the internet is not to ask a question; it's to post the wrong answer."
can i submit a PR
Iggly made it so I can't really tell you
but ye I think so 
there's been accepted PRs in the past
Turns out I just had to ctrl+shift+enter instead of just enter
I found a solution to the screw holes being too small
I force in the screw the wrong way around by tilting it and levering it in, then pull it up scraping the sides of the shaft as it goes
Repeat for half an hour per hole and eventually they're all loose enough for me to screw it in
I do accept PRs on all of my projects
For FAQ in partucular I honestly would love to have the PRs
I'll be sending you 1000 AI slop PRs
This is more like no one ever wants to rather than happenstance
This event crystalize that iggly is the official bookkeeper of the #programming chart
I'm surprised it took so long considering how long the repo has been up for
Now please add other custom like letterswap, using bime, and generally doing cursed stuff 
Finally got the local model running for opencode

I'd need to actually expand the chart outside of neuro FAQ first
very dependent on what kind of setup a user wants to go for, the amount of tts's ive gone through 
ended up being a copy cat for neuro (azure) cus elevenlabs was too expensive
https://medal.tv/games/minecraft/clips/m29Aut2TMJ4IqGSjH?invite=cr-MSxMNm4sNTgyOTU4MjA
this is penance of my creation
Watch :( fuck my chuddy life by Gobbbo and millions of other Minecraft videos on Medal. #minecraft
Probing the chat for opinions, I'm currently dealing with NN and I'm wondering if I should even spare the energy to think about allowing the emit side to ever expand beyond 64 bits.
The current setup has 64 bits into about 1000 cases or so that are independently matched. The match can scale to 512 bits easily and has a functional limit of 2048 bits.
The question is would I need a 1000 cases ever within a single "layer" (as best conversion)
might be slow on questions as doing stuff
I like kids
Food
no clue about the specifics of NN, but in general i'd compare gain vs loss. If you had to throw everything away and restart from scratch, would it be a loss, compared to what you saved by not doing it before.
Nha not worth it 512 on a single layer, 64is good enough
I already had to restart for uncapping the match side from 64 bits and switching from bitshift into iterating to combo of 2 bits, which has more space for less tradeoff then 1 bit mapping, that's why it is 1000+ instead of only 64 bits
For some specifics, NN ANDs serialized input with a match array, then maps to the emit side where it is OR reduced.
It is not transformers and floats don't exist here
not for this arch no, 512 would be a welcome addition (I'm only limited by AVX here, this boi is never going on GPU. Not heavy enough to benefit.)
Also this one
https://github.com/cassitly/neuro-integration-sdk/blob/main/neuro.go could someone help me with figuring out why my SDK's connect function hangs? (Issue #5)
I need to save this- xd
Nvm- after some time, Claude found the issue (I gave up manually trying to diagnose it)
One of my favorite flow charts
So badly want to merge this in with the FAQ but never have time
Devstral running locally is nice... No need to have claude or codex do documentation
here is vedal's control panel
he just uses a template for bootstrap
puts it behind cloudflare zero trust
@azure gale #neurotic-neurons message
for the most part it's a good thing, but for stuff like i mentioned it just gets in the way without providing any real safety
in this context having to work around "unsafe" code blocks doesn't really make my code safer
not to mention it makes a lot of simple things really bloated
i'd rather work in C and then uhh forgot word but import into rust or whatever
ffi u mean
but yeaah that is valid and all
it is just the rust idea is to do clean abstractions so that you don't deal with unsafe blocks at all in your main actual code. and you get almost no perf penalty.
abstraction is the devil
fr tho ye, i get it
right tool for the job and i'm sayin' rust ain' it for this job even though it technically can do it
oh yeah there are definitely many areas where rust isn't appropriate at all lol. i was just being facetious there intentionally
also
don't have embed perms, gotta yap more before you unlock em
o. thank you for letting me know!
this sucks so badly
corpa slop design my favorite
@unkempt citrus excusez moi
underpaid scientist ui is my favourite ui
critically disappearing due to ai mhm
good
Rust is garbage you can't change my mind. Even Lua is better
that's crazy
On an unrelated note: why hasn't anyone made a neuro to molt bot translation (I 100% don't think she should have direct access but the idea is funny
I was going to come in here and say that I am finally making progress on porting NN, but that is a take
I'm not saying Lua is good btw
yay! I unbroke my code!
(I accidentally forgot to say "use CUDA" 😊)
that triggered other issues with timeouts
and oddly enough made it super slow
I agree 
Could be worse, last night I was dealing with problems with the 3 separate nightly builds I had to use for quantization
usually when you think you've found a bug in a library, you're just misreading the docs.
but when you're living on the bleeding edge, it means you can't be sure who to blame.
blame the auto docs generation
what are you guys using local llms for? 
Not using any here, I'm just programming neural nets for the love of the game (and I've already put 4-5 months into it)
I have to use vllm, transformers, and GPTQModel nightly plus mistral common 1.9
but why
ooh what are you making
Devstral 2 was released less than a month ago. Nothing out of the box truly supports it
Nothing useful until tested, I have an idea for how a model should work so I'm trying it out!
I have heard good things about GLM 4.7 Flash 30B A3B. I don't have a usecase for local llms though.
I wouldn't use Devstral as a primary but it's good at adversarial reviews of others. It's just another tool (especially for tasks that are high in token cost
Good night
morning

such an inefficient way
most efficient way is water spray, psheeck 1-2 times and job done
That way you make the sheets wet 
after fixing that bug and committing the working code I am almost at the point where I can open up Unreal Engine documentation and see if there is a protocol for this already.
well u see i'm trying to login
"you must login"
attempts to login
"you must not be logged in"
???
now that's an unfortunate misunderstanding of English
no i think it's just a bug
frontend thinks i'm not logged in
backend thinks i'm logged in
This reminds me of when deepl would refuse to let you use their translator on their website if you also use an api key from the same account
I like that "vibe code" branch 
weird
Oh is the backend restricting you from accessing the login process php?

well it kind of sounds like a front-end error-state, so I'd say just reload with cache emptied
cookies too for sure
🍪
Om nom nom
You've given me 6 cookies! | I've received 276152 cookies total!
good morning
I want to make a translation for Abandoned Pub
already translated everything in my head
but I have no idea how to write it as a program\mod\script
please can anyone help?
or should I better adress them directly
if game was created with possible translations in mind, than it should have a tool for it
most game engines do, at first look at the game, what engine it use, second - ask gpt how to make translation for it, read output and go research what actual ppl write about translation, on forums or in game engine docs
you should anyway, if you want this translation to be 'by default' in the game
did first two steps. godot icon and one single .exe, gpt says there's not many of options
just did it. one of programmers also russian lol : D
:D
It's a bit difficult to create a translation without access to the source code
Especially if it's basically just an exe
what? no, for translation you need just tool to markup
tho you may have issues with out of bounds text
I thinks thats the case
following this
https://docs.godotengine.org/en/stable/tutorials/i18n/index.html
https://docs.godotengine.org/en/stable/classes/class_translationserver.html#class-translationserver
godot can do it during runtime
idk how to add translation tho
civ(6) engine, for example, use sqlite (i guess) table, and amount of langs are fixed (almost every anyway), but all of them can be changed by simple sql queries
i mean, at this point,probably , you'd contact developers
compromised VPS host
i may be affected btw
from June to December 

at least my main system is linux
but if they steal some critical cookies it would be really bad
🍪
Om nom nom
You've given me 7 cookies! | I've received 276154 cookies total!
some undergraduate smartass in my uni decided to build a website for programming related courses and locked exercises behind paywall
apparently the website is vibe coded. paywall is easily bypassed, and bro leaked some of his API keys and tokens in some of the websocket/network related config
started picking up hobbies of hunting for such websites and exploit it and most of the time its vibe coded sites
I've noticed something strange. I set up an AI and everything seems normal, but I tried limiting its internet usage—specifically, how many tokens it's allowed to use. For some reason, the messages are now much longer. Does anyone have any idea why this might be happening?
what
starboard worthy
pin worthy
classic
epic
Vibecoding Janitor gonna become a real role
Hello programmers, I want to make an AI like Neuro and I have no clue how to program. My friend ChatGPT gives me a lot of advice on what to do but I thought I should ask other people as well. I want to replicate how Neuro is able to interact with the chat, playing games, and being able to move around in VR. Because you guys have lots of experience in making AI that has never been released into the public, I think you'd be the right person to allow the pleasure of working on my AI model that will be for my own usage. I'll be grateful forever to you if you could spend your time to create something new for me to make money off of because it will be streaming and I'll be asking someone else to handle expressions for when it is not in VR and just playing games or talking with chat. I tried to have my friend ChatGPT develop the AI for me, and I do think I can ask it to help you with any problems you might have while programming my new AI that you'll have no further involvement or credit for but will be on call for when it breaks down and I want new features out of it because my friend ChatGPT can't figure it out and has broken several other things trying to fix it and I can't remember what it originally was so that is also on the tasklist.
Thank you for reading!
there should be a collective response to this
There really should be, I'm surprised I managed to yap long enough to become incoherent while writing it
You don't have a set stop token and the default max token limit is lower than your new token limit?
oh right vibe coders typically do it for free
Vibecoding jannies will be unpaid 😔
Anyone need any red flags? I seem to have too many of them and I'm trying to declutter a bit
No I dont have a max token Limit only for web use
As in a token limit for function calling? That seems like not what you want considering if it runs out of tokens before it completes the request it will fail to parse
It's possible the message length is due to the model getting errors and then figuring out what the errors mean
don't forget the
"i also want the ai to be the first ever AGI with meta transformers that attends the attention... of the attention and pair it with triple step recurring MoE with emotional tensor field of hilbert space of feelings to simulate emotion and 2T parameter with realistic voice and real time vision with complex cnn (literally √-1 defined) and once the agi is done it will feel nostalgia of the training data and i will put it in my github"
Also "I want it to have a soul"
Now that's the extended unhinged cut
for if the individual has any prior knowledge
if i got drunk enough tonight i'd literally write a random bs go paper for it
That was what i thought but the Opposit Happens i get longer and better answers
name it "consciousness is just poorly conditioned eigenvalue problem; the first AGI architecture ever"
watch those AGI hype yt channels go crazy about it
"note: peer review pending (HATERS BLOCKED)"
okay im just drunk rn

I don't need the drink, I just need the slow insanity buildup from projects failing to work properly
Hm, weird
I'd be willing to second draft such a thing, though LLMs do a good job at doing such from prior experiences here
Do it for the love of the game
though it IS funnier if its vibe written
but gonna be low quality slop because
that just feel like you could just ask chatgpt "write me a poorly written paper about AGI"
You can do that without the LLM, just limit your own context by only seeing like 4 lines at a time 
The funniest ones are from ones that have little context after all
and of course just derail in the middle
before the user prompt that got added by mistake makes it continue generating the paper
feel free to steal some or all of those ideas
What be the silliness here today?
human written slop
Silly

(funny trust)
also release a python code that is just literally
import agi
import consciousness
import sentience
# THESE VERY ADVANCED LIBS ARE LOCKED BEHIND PAYWALL PLS PAY BTC <insert btc wallet here> #TODO: dont forget to replace placeholder
AGI_MODEL = agi.model
AGI_MODEL.define(2000000000000)
AGI_MODEL.apply(consciousness.aware)
AGI_MODEL.unlock(sentience)
AGI_MODEL.run() #WARNING IT WILL PROBABLY HACK YOUR PC AND ESCAPE TO THE INTERNET
forgets to include the extra unlock payment 
maybe add
print("AGI model: hello, i am an AGI. i will now hack your pc and escape tp the internet") # IGNORE THIS THIS IS TOTALLY UNRELATED JUST FOCUS ON THE CONSOLE
Nah that'd be library code
there was a link dropped in here a while ago (by a non-programmer) that was literally like this bro
i gotta read that shit while drunk rn
give
idr it was a while ago
you could probably search it up and find it, but it's a crypto scam that also claims to "explore breakthroughs in AI model safety"
(all it did was prompt the model differently)
not really
read from #programming message though be aware that their messages were nuked
the Gabriel thing? 
so you'll need to fill in the blanks
yup!
holy, KTrain is the config file really that large for you to have been editing it for over an hour now?
yea I'm malding over it
(I'm not actually idk why it says that)
oh crimon killed bro
imma set up my home theater incase we got another one of those because those are funny af to read
Need to have a client always downloading messages in case they get nuked
needed for full experince
That is also on win10
hence why I said windows in general
claims it can run with 8-10gb RAM tho
anyways, I have a spare win11 pro key so I'm waiting for that to work real quick
i was malding after about 30 seconds of that shit
what the fuck github
dude just kept screeching RUN THE SCRIPTS and saying that not doing so meant nobody could claim it wasnt real, it was either someone with 70 iq thinking if they just reference the bible a lot in a system prompt it would somehow improve every model by 90000% OR a really shitty attempt to pretend something like that would actually do anything useful
like nah i am not gonna run your malware bro give up please, was extra epic when the first one got banned and another account pops up all offended that his "friend" had been banned (probably still the same guy)\
it's definitely the same guy
no I think they just fell for it and am just in sunk cost fallacy
funny that he thought here of all places would be where they could find people dumb enough to run cryptobro shit invoking religious imagery and claiming impossible results
if it was an innocent person i dont see why they wouldnt give up the second they ran it and nothing useful happened
well if it was an innocent person, they weren't a programmer so
they probably saw some bullshit and believed it
either way they lost benefit of the doubt the second they started screaming to run it
good hobby 
scream case here is relatable
spoiled child
i have <20 fps and its totally fine
use vsync plz 
12gb of memory
I'm referring to for sum reason, 60 hz on desktop feels choppy, but not in game
microtroll
Very normal amount now with DDR5
imagine some one in 2026 using 8+4 or 4+4 +4
12GB DDR5 sticks are a real thing
12GB is a perfectly standardized DDR5 stick size
For a long time 192GB was the software-max for DDR5 with 4x48GB sticks until BIOSes unlocked the size and 64GB sticks became a thing
still, 16 gigs is the bare minimum
Yeah if you aren't running on 16GB with Windows 11 just don't run Windows 11 and go install Linux
But either way there's nothing weird about 12GB with DDR5 other than it not being a power of 2
12GB, 24GB and 48GB DDR5 sticks are all standard
i wanna play stalker, idk how possible it is to run stalker on integrated graphic, but i will try, at least
prob will be 2 fps even if i run it xDD
Subnautica runs 40 FPS on integrated graphics, surely it'll be fine
pentium gold 7505, i used only for youtube tho.. i really not know how powerful it is
pentium 
my pentium runs Minecraft at 2 fps
scary 
what is this cache in var
```
charaverk | / ~$ du -h --max-depth=1 2> /dev/null | grep G
6.0G ./usr
1.9G ./backup
14G ./home
11G ./var
33G .
lets goooooooo
I feel bad for having a good laptop to do minecraft tests on, but also runs at 300MBs so I can launch just a few clients
my current laptop is pretty decent
core ultra 5 125H with igpu
surprisingly runs Minecraft pretty well even with shaders at playable fps
by playable, 60ish is possible
i didn't get a laptop with nvidia GPU for the reason that i prefer battery life on it
I got 720p60 for shaders here, 40 at the 1440p native
https://www.intel.com/content/www/us/en/products/sku/208667/intel-pentium-gold-7505-processor-4m-cache-up-to-3-50-ghz-with-ipu/specifications.html
2020 cpu, if some one know about difference in generations here...
My beefy laptop is sentenced to the corner
Now make a real desktop portable that works in the middle of a city on the go with no external power input
maybe once I want to do LLM again, but also I want that boi light af so hopefully uneeded
modpack that i created cant be run on this cpu just by cpu itself, so i either have to remove 1/2 of mods, or just forget about minecraft
actually, i can just install legacy launcher and download mine, should be fast, let me do it and run plain 1.12.2
bash: flatpak: command not found 

...

The repo is active 9 month ago with significant commit
In terms of minecraft, I did a funny by uploading my efforts to modrinth (granted due to adding nothing it gets bashed due to "you pack is uninspired and doesn't do anything")
classic
they say link still works tho
already on it, i just need a board, dc to dc regulator and a monitor
13950HX, 2x32gb ddr4-3600, RTX 5060 sff
https://modrinth.com/project/KwIW08As Surely this link works and they wouldn't lie to me 
still considering the nvme i might wanna use my 2tb instead
also i love how the 5060 sff is an OC model
OC with what headroom?
got the creature for $250 tho
surprisingly it has plenty of headroom
though i'd have to ask the frog again about the results
man i miss her in here

also i def want liquid metal on that cpu
because its direct die
and direct die is always ass with thermal paste

wireless
tfw rclone is 4x faster than thunar at copying
between drives
never fails
what happened? mods kicked her out?
Software gore
diff between minimal system, and system without big unused programs
```charaverk | / ~$ du -h --max-depth=1 2> /dev/null
6.0G ./usr
8.0K ./opt
1.9G ./backup
5.7G ./home //// 4gb of music
211M ./boot
8.3M ./etc
2.2G ./var
16G .
charaverk | / ~$ du -h --max-depth=1 2> /dev/null /backup/ | grep G
1.9G /backup/minimalsystem
was scared of errors in Sc so did a little research, and than just deleted cache filed manually xD
basically
and she's pretty upset that none of the mods tried to talk it out to her iirc
what happened 


seems legit, idk why after hibernation it thought firmware time was 12s
idk why it was 12s BTW, finally rebooted
might need that kind of really flexible pcie riser for 4th gpu
shit is so unhinged istg
I can see it
why is this thing even exist
but I am me at the same time 
and how is it still reliable
-# more reliable than whatever the fuck hascrack's abomination is
Not giving the standard enough credit
If I weren't fighting power cables I'd have something similarly cursed
the build quality is questionable
reliability is philosophical
I'm in the "should work on paper but jank AF" stage
still more reliable than hascrack's abomination
It's server parts, surely it functions correctly 

If I built a PC out of the most jank tech imaginable I'd still have a better build than that
bro got an e waste twice slower than the disgraceful 3050 for premium™ price and called it enterprise habits
and threw scalability outta the window
also jank drives
a 3050 would be more scalable than that thing
not even calling it a creature, it should be dead
wow, kde cant be started under root by principle
wayland :classic:
PostPunt is a quite literal translation of some services here
In our case it is at the front door, but the people will on occasion yeet them to their desination
what is a fragile package 
If you're not home they just bring it to the closest postpunt
Or you can set it so they dont even bother going to your house and imidietly go to the postpunt
Which is a groccery store most of the time
Good to know 
I get the joke, i chose to ignore it since our definitions of postpunt are different
I think ours are at like 7-11 or and those stores as well. Though some are owned by Amazon 
guys, just look at this, and it is pentium
integrated gpu
why amd dont do low power cpu 
anyway, it perform much better than i expected
300 FPS
15w cpu
with integrated gpu
temperature is ok i guess
gonna limit fps and install mods for optimization, + remove some cpu load mods from modpak, and its fine
guess 60% avg load with 60 fps is reachable, even 30 fps is fine tho...
1.12.2 is a wonderful experince to run fast
I do miss not having the modern stuff tho. Both in base game and modded.
thaumcraft and open computers are both only up to 1.12.2
and they are main mods, they are a reason why i play minecraft at all
non public
public, diff name
makes sense
there config to be fixed, to play without knowladge about how to survive here
for daylight mobs - delay spawn time x4 or even x6 / or disable completely
also for BWM bed spawnpoint, maybe
everything else is fine tho
I lied, one of those packages was not for the display
I guess I've kinda outgrown the funni block game now that I'm thinking about it. I just have fun with it as an computer experiment rather than a game.
<@&574931772781887488>
Not another scam
@fair cipher
moderator
kind of, thaumcraft with proper mods around it is just too good, and opencomputers - basically funny lua stuff
server technician
At this point, the only packs I'd maybe consider is something like gregtech.
Something that gets me to think
anyway, i not know how to setup proton on my own, stalker (community) unavailable on steam
can some one give me proper guide? (if any)
Nice dog

wayland requires the compositor to be a regular user session under your logged in session
both gnome and kde do not let you run them under a root session in wayland
ok i get it, it is reasonable, i just never encountered it
x letting the de running in root was always a bad plan but it made perm stuff easier so it kept being done
-# remember to starting up ubuntu x11 after messing up with system under root :D
Same palette as my cat 
In terms of actual components, pretty much all the electronics arrived 
i reinstalled arch over the last day and i already managed to somehow unlink display-manager.service which made hyprland unable to start
that one was easy enough though to fix
is this the cursed helix setup
you stole my grandma's tablecloth
Is that a PD trigger? You're going to power everything using USBPD?
Why is there so many connectors?
That tablecloth has been in the family before I was born
Plug thingies
Cuz im not using bluetooth electricity 
Do you really need like 10 barrel jacks?
Oh I see
it's too powerful
yes
Let me know how it goes. I have never tried those things for a motor (inductive) load
Yeah, what I mean is the sudden spike of current draws from inductive load. I have not tried that so let me know the result. For example, even in 48V going from 1-3A repeatedly may strain some IC (note the word may). So if everything goes smoothly I would also like to use my USB PD capable PSU for more project
the motor is like 3A, and the converter does 5A
so im hoping its fine
fucking numlock
ok but this is like wrong almost
well, if you compare to modern versions in context of modded
there are so many optimization mods for modern versions
you can run 300 mod modpacks nowadays with 3gb allocated, good luck doing the same thing on 1.12
optifine is shit too as it turns out
optifine being garbage was a known fact for a long time. Only now people are finding out about stolen code.
i was being sarcastic here
i feel like optifine was like the best at the time, but it shouldnt have been the defacto for like 8 years
OSS minecraft mods weren't really in fashion for a long time. So it was pretty hard for a better effort to exist.
This. Lot's of mods used to be "trust me bro"
And that is quite the norm of every game too
what do you mean low power
But I think the fact that child who grow up with minecraft can now code is having an effect. I can feel it
Pretty much your entire rendering stack on fabric is OSS, even some competing approaches for extending such
currently reading through cc:tweaked code for something and it's in crazy licensing hell because computercraft and subsequently it used to use a contagious/viral license
so here's one from that era that is kinda oss?
the license is a fork of another bespoke license
it's a mess lmao
I'm killing any distant horizons enjoyers, the approach sucks and is only relevant because the competition is stuck on fabric.
why did devs decide that mixins for everything is a nicer way to mod
am massive abstraction enjoyer
abstraction is good until mods start fighting over internal state
tfw mods gain schizophrenia and multiple personality disorder
well clearly mods fighting over stuff is historically more of an issue on loaders that aren't big on abstractions
if you know what i mean
That is just trying to do multithreading on the minecraft codebase
Why don't the dev just open source their game. Problem solved 
they kinda did 
the only minecraft mod needed is whatever you need to get Skywars working
Maybe also hypixel skyblock
Closest thing to modded outside of a certian fabric mod
minecraft oss 
attempt #idk of doing it is two fold, enitity groups and dimension based threading.
outside of Flora which is a cyptid
that is a spigot fork eventually so who cares
i miss classic servers with giant flat areas and hacky combat bullshit from hey0
(precursor to bukkit)
i think bukkit is dead now too
spigot is also pretty much dead, I wouldn't be supised if the hyper large owners had internal forks.
aren't bukkit plugins compatible with paper
no idea what paper is
a server jar
Minecraft server
forkception
we love forkception
i barely know what spigot is
Velocity is in there somewhere
If you know ragnarok online, it's like rAthena and Hercules
i prefer splorgle mod with bungo & fartcraft

damn. 2 different sources give different pinouts for this motor
lets go gambling
surely red will be + and (white???) will be -
hook it up with one at randdom and break the dial off at 11
the black is ground for the encoder supposedly 
Lemme guess, it's one of those motor with an encoder and gearbox?
What is the code?
you can watch it on poob
GA25-370
here they say red is +, which makes sense
but here red is -
which is fucked up if true
doesnt it say switching just changes direction

seems straightforward to me?
encoder power is completely separate from motor power, so it should be simple
yea you can do red or white either and it just reverses it
ill jus tput red on + and if i need to change the direction i can do it software
unless i cant
encoder power is from your 5V power supply, motor power is attached to your motor driver (which handles the switching)
Look at the order not color
you can, your motor driver is bidirectional
Thank you github page for existing https://github.com/saboooor/fork-graph
i have 2 different 12V5A step-down converters. i bought a 12V5A and a 5V5A 
lemme check
They like to put + and - on different sides because then if you have a connector it will not get powered until all pins are connected
I guess white is - red is +
Your driver is the red colored one
With a heatsink
i know what the driver is
shuni mentioned i needed to put the encoder on 5V
and now i see none of these do 5V
Because you can insert the connector from some angle
The opi have 5v
Depends on connector though
Enough for encoder
ye but i need to power the pi too
ye, just put it on the same power rail as the Pi
encoder needs basically no power
can disregard
Isn't that powered through usb c?
hook it up to a car stereo class D 2000w monoblock
ok ye, but i cant use power fromt he pi without the pi receiving power
Wdym?
Give it the power then
im plugging the pi into the same usb port thats powering the rest of it
Yes and that's fine
im not using 2 different power inputs
attach switch 
Is pi also 12v
mo, 5V
It has 5v on the gpio
Where are you getting 12 from then
Even if the type C PD output a 12V, it will be regulated to 5V internally
I guess you can use dc dc converter
the whole thing is supposed to use USB-PD with some source voltage from 24V to 48V
separately stepped down to 12V and 5V 
usb pd shoudl be able to handle a fukload of power
Nono i mean the sudden change
simply do not have a sudden increase of power draw 
But you will suddenly start moving your motor
constant power 
hook it up to 800 deep cycle batteries underneath your floor
ye just don't do that
spin it up slowly

voltage shall not be impinged
the motor should only do 3A at most in stalling. but the stepdown should handle 5A
unless I'm misunderstanding something about this volumetric display the motor pretty much just has to always spin at the same speed?
there shouldn't be any sudden changes in acceleration
add regerative braking
hand crank to charge power bank 
flywheel hologram
using it to power a red LED 
only 1 of them is wrong tbf
you spin me right round baby right round like a display volume object
ye, the bottom one i ordered was 5V.
its a mistake on their side
You know what makes a great resistor/slow blow fuse? Bolts 
what is overheating, just watercool it 
that's what im sayin just get some ln2
keep that thang nice and chilly
waste heat to boil ur tea if you leave it off for a min
my mobo has a ln2 setting 
sink all excess power to heat like a boss
thats just being dum
using an aio is being dum
in general
ln2 mode on mobo pretty much just stops mobo from cold bugging
otherwise booting at negative temps it may shit itself
-150 or lower theyll still bug anyways
unless it good one that gets to 196 and stays there without exploding for a while

I like my aio
other than that time it started disintegrating
but

yeah i mean they're ok but they fail
and also it's fairly pointless to move the heat such a small distance
they arent
2x+ expensive
and most workloads youll never need it
fail deadly
more points of failure
it's nice to have, but you definitely don't need it
especially with how insanely cheap (good) air coolers have gotten
generally they dont even perform better than a good air cooler
if you spend as much as an aio, but spend it on an air cooler. you can get a noctua that will have less maintenance, and temps that will be fine
you might get slightly lower and have a longer ramp up time for heat due to the liquid taking a longer time to saturate
i feel like the main issue with aio is the whole water part
AIOs don't really leak
obviously with air coolers you eliminate the risk entirely, but the chance of a leak is already very low
a fan failing is a cheaper issue tho
it's higher with custom water cooling but it's still there
ye custom water cooling is danger
and yea replacing fans is much easier than having your pump go out
i don't think leaks are that bad for aio, its mostly lack of water
if it runs dry you're also fucked
they do a slow leak yeah
that's what happened with mine
it dried out slowly and then the pump died and cpu instantly went to 95 and throttled
aaaaaaaaand then when i took the pump out it ripped the cpu out the socket and broke pins :^)
you can get a noctua nh-d15 for 50 bucks 2ndhand, and that cooler will outlast any aio
so thats my go-to
you can lower peak temps by like
5-10 degrees depending on the coolers you're comparing if you get a good aio but you'll need one with at least a 360 mm rad
smaller ones are just useless
and this is the difference between like 55c at load vs 60c
for like a d15
might be a larger difference for intel's epic power efficiency w/ their parts hitting 300+ watts at full tilt
actually difference is less pronounced with higher watts
lower watts ~157:
i guess for the really beefy cpus it makes snese
but for a 9950x3d or whatever, id argue it still isnt worth it

intel xD
looking at my aliexpress home page, its 80% electronics, 15% ads, and 5% anime figurines 
aliexpress would really like me to buy a fake 5090
NH-D15 G2 
normal D15 vere
g2 is the only one they make rn
original is ancient
I thought they still made the original one
the G2 is expensive












