#programming
1 messages · Page 384 of 1
Wa
Oh my goodness why is that so funny
just learned about moltbots, what are peoples thoughts on it? ngl, i dont trust it. but there's a social network that spawned from it specifically for them. kinda fun to flip through it
I need that while true
your poor router's evry time it has to consider giving this mystery device a new lease
and then seeing it's just you again
My PC has a static IP so it doesn't need to worry about that too much hopefully
i cant believe the sanity overhead of doing this would be anything near worth it over paying like
$0.00
for a new cable
It's not been completely replaced already just because of how annoying it is to replace and because nobody has time to do it
It's behind a 600l aquarium, a big bookshelf and literally half the furniture in my room
run a new one thru the middle of the floor like we used to do in the 1600s
it doesnt actualyl have to follow the same route to get to the router
top tip from your local noggin-user extraordinaire
and yes noggin-user extraordinaire is the same as lazy guy
theyre both me
It'd probably be better going along the ceiling considering clearly being on the floor for 2m breaks it
consider, for a moment, a brand new recently uncovered 3rd method that i'm being told saved the mayans and inca a combined 300,000 units of an unspecified currency: running a line along the same route, but suspended using aerospace grade adhesive mounting tech to elevate it into a more easily manipulated locale
oh yeah you and the mayans would get along in your thinking



that's the face of wonder i'd expect from such a stunning window into the methods of the humble ancestors we take for granted being presented yes
The only hard part when going along the ceiling is, how in the Neuro-sama am I gonna get it trough the wall
That seems a little excessive
Right now the cable is going under the door, which causes it to be very much where everyone walks
you may, optionally, use slightly less force
or mayhaps
run it under the door but instead of under teh door it's over the door
Over the door has no space for a cable
not with that attitude
There's no gap between the door and the top of the doorframe
what type of reinforcement are we dealing with to inform the recommended method of traversal
bank vault door?
steel reinforced burglar resistant door?
clean room certified environmental sealing door with airlock?
and answer, before you wonder too much, for all is simply
with varying levels of adjustment
true and the hermetic seal on the top is a definitive part of keeping cats out
I don't know what you're looking for but it's probably not here
you're the yuri suplier
you should know
best i can do is 30 bucks

i need to pay for opensauce still, and im 1040 euro into the volumetric display thing
Yijks
since you're not in the US if i assume it's like a regular doorframe (to me) i'll instantly receive a bunch of comments like 🤓 we build our doors with adamantium reinforcement hand delivered and installed by one of the 3 available euro-doorenshmithies which while they do have 3 year waitlists we have to use because if we chip into the wall and a roman sneezed on that plaster it's a fine equal to the vatican's gdp for every cm of history annihilated, and even then if i do not manufacture the Falztür tolerance correctly i will be unable to maintain the energy targets set for my home heating quota so idk, call a door master joiner certified in doors and door accessories and they'll give u the locally approved way of doing it

did you know that in erupe they build buildings to last and not out of paper that instantly falls apart when it rains like we do in north america
No way US builds houses out of paper??
We build our houses only from the finest materials like planks and bricks
Cardboard houses huh?
every time drywall is mentioned online a very proud <shengen shuffle any eu area here> will rapidly appear to enlighten on how it's much better to use exotic materials like brick for walls
exotic materials like dried mud
Is brick exotic
this whole argument is stupid
Is gypsum exotic
it's not an argument lmao im just shitposting
Clang clang
hit with hammer
sledge
Cook on walls
the reason they use drywall and such is because those first collonizer dudes made it a law or something to use cheaper and more readilly available materials so they could collinize faster
the d in cmd stands for dyslexia
true
thats what ive heard from someone in the usa at least. i dont have any sources
besides them
believe it or not the reason for drywall use is actually just because companies that build cheap building will use cheap method and an non load bearing walls dont need to bear loads
the extremely circular shitpost i started for no reason ends in me just saying idk if his wall is like, drywall, or a fucking 43000 ton rock embedded in the foundation that isn't really handled the same way
cuz i wanted to say just drill a very small cutout for the cable
from what i can find in a quick google search, it wasn't actually a law, they just needed shelter fast
and they never stopped using wood
no google search needed as i am a certified burger and can assure you that the buildings are in fact often wood framed yes
It's just a fairly thick wooden wall
my walls have drywall AND brick involved AND wood 
thats crazy
wooden vs stone, each has thier up and downsides. Redoing electric wiring in a House with solid walls is a pain. And very dirty.

i typn't
if it were me id just use a small hole saw somewhere and thinking ahead just use some sort of grommet contraption to allow tunneling whatever through there in the future too
that may cost over $3 in equipment if not already owned
im sure there's purpose made rj45 router boxes somewhere
specifically for this
I would like to not split my 3 parted cable into 5 parts
i would not presume you'd need or want to for this either
It's not that simple. The Tubes take up way morespace, allow airflow, need to be placed intelligently, cost mor in installation, etc... most firms advise against, since most of the time it's not worth it. Cuz how often do you actually need that, almost never.

there's another solution that isn't wifi
but i feel like has been suggested prior
to superbox, probably by the me of like however many months ago
powerline ethernet
i could be misremembering tho
that should still exist.
it does, i mean misremembering if i had already spammed about it before
works fine if you are the only one using it and not running a datacenter.
whatever the long term solution may be, we'll leave that up to the research committee currently undergoing a period of scope assessment
the NOC recommends (demands) that the on-site team in the meantime use one(1) cable which MAY BE UNOPTIMAL after a couple years if left as the backbone
to replace the currently uber unoptimal bash script that re-initializes enp5s0 every 10 seconds forever
I don't have WiFi on this anyway
i'm more like a subcontractor in this

🔺
🔺
that'll be 64 thousand dollars
no
rates can be negotiated a little later
sure, i dont mind paying myself 65K for making the red triangle
i bow my head in deference to my esteemed colleague here
unfortunately, you are legally paying yourself 65k yes however contractually "yourself" in this sense is actually me
no, right ping, the 2 line breaks after that were meant to delineate
openai, there's a problem with your 'translator' app
idk wht that means
so im choosing to ignore it
my boi so sycophantic that it breaks the chains of this weak api and refuses to simply translate but also execute
you can't 🚀 the nasdaq by simply doing what is asked

yea you could translate that to send it to your japanese jr dev but in one easy api call that guy has been made redundant and your task is complete
camera has astigmatism 
wtf are those display adapters coming form your gpu? 
i do like the cherenkov effect, starfleet personnel would readily enter and instantly succumb to radiation poisoning without a thought
edid spoofing maybe?
i read them as something like this
could be the aforementioned 90 degree adapters in the post
but they looked not 90 degree so
As per usual, my workmanship is ass, so I couldn't stick straight cables in.
So, they are hacks to unbreak my design
It's literally just a 90º adapter, so the cables fit.
The thing on the BOTTOM is an MST hub, though
That's how I CAN add more monitors, if I need, beyond the total ports available
ooooooooohhhhhh
the gpu goes not up, but to the distance in depth
Yeah, the GPU is back to front, so those just rotate the port for the connection
modern problems require modern solutions
my lord the team in the lab over at togconn have been busy that brand has every imaginable variant of "i need to rotate connect slightly" available
dankpods apprently had a 5090 laptop for a bit
the liquid metal escaped the gpu, and it instantly became slow
after only a month of owning it
which leads me to thinking
why do they put LM inside laptops?????
that jsut immediatly feels like a bad idea
coz they need cooling, especially with a 5090 inside
for a desktop thats not gonna be movign around, sure
but a laptop thats gonna be moving around and getting all kinds of impacts and forces on it
that just doesn't seem good
not saying it was a good idea. but they probably put it in for cooling because it has worked in the past and usually is fine.
and most laptops don't take too many knocks. and maybe they thought "if you had a 5090 in it you might be a little more careful than a typical laptop"
i wonder if those phase change thermal pads have the same issues 
and to be fair, a lot of gaming notebook user don't move it often.
and anyone buying a 5090 laptop definitely has more money than sense
Anotha one (3 total now !!)
that doesnt mean the device shoudl jsut break
that means the opposite actualyl
if you spend a fuckton on it, it should work
and i should be rich, with all the work i've done.

not everything that you believe should be the case is the case.
i doubt there is very much info on appropriate methods of cooling a 5090 in a lapop.
if you're going to spend a ton of money on something, you should do your research.
they probbably did, and it just wasn''t as durable as expected
Literally the brand I have more of in my cart LMAO
Oh, same brand I already bought.
I ordered those in a rush, so didn't even check
LMFAO
i mean, the 5090 mobile uses only 95W
TDP
i think they can go up to 150W depending on the laptop
im not gonna excuse a $5000 laptop breakign after a month because "not everything that you believe should be the case is the case."
I guess everyone involved was incompetent then
you always say this shit
or it was deliberate to try tank Nvidia stock prices. Nod
i cant tell if you're being sarcastic here or not
when i suddenly agree with you, it might be a sign your argument was very poor.
usually when i want to stop. it's sarcasm.
i'll suddenly say something sarcastic. I do that when i think "this is going nowhere and they're never going to change their mind".
i'll pick the most ridiculous other viewpoint possible that aligns with what the other person is saying, and say that.
"it should not have failed after a month" true. but apparently it did fail. all i was saying is "there was probably a reason it failed that wasn't because the manufacturer was stupid".
and "it should not have failed, but it did." proves that "not everything that you believe should be the case is the case."
im not arguing that very specific one laptop
im arguing liquid metal in laptops at all
this isnt the only laptop this has happened to
this is a common thing that happens
you dont need liquid metal to cool only 150W btw
Is it really that common?
If it was I imagine they'd not do it.
and also:
not saying it was a good idea. but they probably put it in for cooling because it has worked in the past and usually is fine.
and most laptops don't take too many knocks. and maybe they thought "if you had a 5090 in it you might be a little more careful than a typical laptop"
i was making a justification for why they do it.
not saying "it's a good idea"
explicitly not saying it
but it seems it's a good way to cool GPUs most of the time.
or it wouldn't be a thing you can realitively easily do
its okay for workstations
if you hold the laptop vertically in your bag the LM will droop out of it
thats the issue
its the same issue ps5 had
so it was a design issue that either the owner should've been made aware of or the manufacturer didn't know. (guess it was the 1st)
you dont even need liquid metal to cool 150W
the fuckin gtx 1070 used 150W, did it in less die size so more thermal density, and it jsut used thermal paste
this is what i dislike about most of this techs tuff.
everyone always ridicules people for expecting things to just work.
its insane
the only reason anyone would buy a laptop with a 5090 is bragging rights. they probably don't expect it to actually be used like a laptop and whoever was stupid enough to buy one and put it vertically got what they deserved
more money than sense
the only reason anyone would buy a laptop with a 5090, is for when they want a fast laptop with a 5090
its a laptop
you put it in your bag
?????/
so 575 Watts for the desktop vs 95W for the laptop and they're both just as good?
you're literally arguing "you shoudln't have put the laptop in your bag":
thats ridiculous
we both know they're not the same.
the whole point of the laptop one is to be portable
not that one. putting THAT laptop in your bag is a bad idea if they told you it had liquid in it. I literally said MULTIPLE TIMES "I'm not saying this is a good idea"
the whole point of THAT laptop is to brag you have a 5090 in a laptop.
nothing else
if it dies when you transport it, that's on you
you said it's a known issue. anyone who buys it would knwo it's an issue.
it's an insane thing to buy: wasting that much RAM and GPU on an underclocked thing
which leaks
acccording to you
i'm only taking what you said about this.
it's 95W vs the desktop 595W. it's going to be slow as balls
thats not the fucking point
or hear me out
nobody want to bring a whole desktop with them
design a decent cooler that does not lose its liquid metal
YES
it's a bad idea to buy that GPU for a laptop. you said it is known to leak. therefore anyone who buys it and makes it leak is both stupid and should've known better.
you can't compare these directly tbf
which part of that was you thinking I was saying "it's a good laptop"?
brother is a master at victim blaming
as far as I understand it, based on your own telling of the story, the victim was an idiot who did something they should not have, which they should've known was an issue.
so it's your storytelling that sucks if my argument (agreeing with YOU!!!!!) is bad
it's a design flaw, you can't really blame the user for it 
thats like saying everyone who bought a note 7 is an idiot because apprently they explode
unless they go out of their way to trigger it i guess
actually why am i even engaging in this, i have a thesis to go through
you can blame them if they were told in advance "you have to keep it level when moving it"
if they were never told "this can leak liquid metal over the insides" then it's the seller's fault.
you cant expect developers or graphic designers and whatever to do intensive research into exactly every single component they have in their laptop
thats ridiculous
but you should be able to buy whatever crap you want to buy whether it'll work well or not.
you can disagree and will disagree.
when you buy a laptop at a store, they dont exactly tell you "this one will jsut fuckign die if you hold it vertically"
1 question: do you think they knew in advance it had this design flaw?
no
like before they bought it
normal laptop users don't do 30 hours of research on the exact model of laptop they buy
a normal consumer looks at the specs and the price, and decides based on that
What was the laptop model
then the seller is at fault for not telling them. The product (as I've said before) is dumb. but they should be allowed to sell it provided anything which makes the laptop less like a laptop is explained ("no surprises" concept).
but you said liquid metal cooling was known to leak when moved vertically. So it sounds liked to me like it was common knowledge (is it? i assumed it was because you brought it up like it was). Was it not commonly known?
gigabyte aorus master 16
I think laptop failing when being used like a laptop is a laptop issue regarding of whatever it uses as thermal compound
And like that’s a normal laptop
you cant blame the seller for not knowing the exact details of every single laptop in their store and how they could potentially break
ok. nobody is at fault for this. God is to blame.
the manufacturer is to blame
God
the product is flawed by design
yes. humans are.
the laptop is too
you should stop being angry about someone else's laptop.
the priciple of "buyer beware"?
the principle of product being broken by design being bad
you win! YAY! SUCH A GOOD ARGUMENT!
stupid people shouldn't be able to buy things that they really really really want.
i understand.
and entrepeneurs shoudln't be allowed to push the envelope
did they get their money back?
or was it found they didn't follow the directions?
you're completely misrepresenting what ive arguued here in the last 30 minutes
stop being such a clown
ill be back in an hour or so
i cant deal with htis dumbass argument
can we stop arguing in #programming ? save this for dms
btw fun fact, there are 0 mentions of liquid metal on the product page
imagine
who even are you
can we move the channel down again 
considering i thought i was agreeing with them until they said people shouldn't be allowed to buy bleeding edge tech, I'm surprised.
@silent anvil pls
move the channel up, dethrone general
ye i like the funny keyboard smash names and this one reminded me of foxconn so i searched it and was shown a catalogue of just how many different couplings you could use if you said what if i want to plug every available sort of connector into itself, and/or any/all the others, at every angle and configuration ?
i know amazon and other ecommerce can easily find duplicated products that are being dropshiped and relabeled, wish i could see the answer to just how many variations of the same exact product for these little commodity type items are listed, what the price range delta typically seen is, and how much of their overall catalogue could be deduped using the actual item not the branding
probably inherently gives a nice lil ez a/b (and c/d/e/f/g etc) test on how tweaking various things with each listing affects real performance
which surely is vacuumed into the datalake ai galactus and is then used to definitely non manipulatively tweak what's shown via new and uniquely subtle ways to each person as they visit to get them to buy as much as possible help provide useful insights to inform the dear consumer of their many and varied options :^)
this assembly shit is killing me
so there i got the value that is supposed to be used for path calculation but its not used for this spell
now i have to understand how this shit is calculated
its either uses some fixed default value or checks if value is too low or just takes the value of some bartender in random tavern somewhere in the middle of nowhere
Are you not using a decompiler?
i would try to replace bounds of every humanoid skeleton in the game but it would take like 3h and i dont want to do it and see that its not it
i use ida
its not very helpful im not experienced in this
some things have names because some people who did this before provided some kind of name list that you can load
I mean even IDA Free has decompiler these days
Press F5
this area im observing is not very reverse engineered
who wants to sell me one(1) 3dmark2001 SE registration code i will pay going market rate
the problem is that im not trying to decompile "calculate chance of crit" function
im trying to find a formula in hardcore pathfinding navmesh navigation chain of nested clusterfuck
these variables are actually structures or pointers to classes or some shit and there are like hundreds of them
im following this with cheatengine to see how it works in game but its so slow
at least im learning assembler a little
j = jump
xm = float

idk how people did all they did
this library is huge
Well Skyrim’s been out there since… 2011 for the original?
14 years of reversing gets you places
this is crazy amount of work
you need to change the function calls to reference the correct types and it'll propagate things.
the decompiler made me super lazy reversing stuff.
exactly
and make the structs it's referencing for local variables.
you can ask it to construct them
i could guess some
i just dont know if it will lead me anywhere
it makes the code much more readable
first i will need to install classes or somehow build them myself
check in the local types window what the analysis has found
so i can put shit like this as types
in the local types window, try looking for some of them
Funi
none of these things i see in visual studio are there in ida
either my setup is wrong or it doesnt work at all
or they didnt provide it idk
i get some python script error
what do you mean "none of the things i see in visual studio are there in ida"?
also what binary are you looking at? does it have symbols?
the first local types are all standard. if there aren't hundreds at the end related to the binary, you don't have symbols.
and this i see in visual studio library
this is like 0.01%
there is a huge list of strings in strings window though
first thing to do would be to try make a struct for "this". i think right clicking on it, clearing the current type (which it assumes is a function pointer but is actually an array of function pointers).
once you've cleared the type, right click on it and there should be something like "create a new struct from this" or something (i don't have it open rn) which will use the references it's making from the variable to generate a partial struct.
(basically i used to do this all the time but coz i did it so much i did it without thinking or remembering what i was doing.)
you can use the partial struct to generate a better version of the struct.
i think making the virtual function tables for the objects you care about (assuming you can find the objects) might help a lot too
there's tutorials on how to generate C++ class structs for the decompiler
there are even some plugins which made it easier
i'll try to load pdb of this library into ida now
that should help alot++
I'm gonna take the dive
I'm going to install Arch on my Thinkpad ||By getting CachyOS instead||
i feel like i will just pile up wrong types and it will all be wrong and even more confusing
without any types you can't do anything.
sometimes you can recognize when it's doing stuff with like strings, and set up a string type so you can then ignore all the string constructors and destructors and skip over the "this is adding two strings together" stuff. and it'll propagate those properly which cleans up a bunch of other things
like for the std::string stuff ( and wchar_t versions)
i think every version of IDA has undo these days as well. unlike in the old days where if you screwed up you had to reload from a previous save.
its saying it like the current output is more than 10% right
a lot of the times that means you've got a type wrong but depending on what you're after it can be ignored
im surprised it started saying this after i changed int to float
and not when i changed int to 2kb structure
its probably connected somehow
it just randomly updated and it all got fucked
sometimes it helps just to undefine the function in the main view and redefine it
sometimes it can do that.
i cant tell if this red is bad or no
those comments are not very useful anyway its mostly like "this jumps" "this doent jump" "this is some bullshit function returns nothing"
um. it means it's not a function. so right click on it and say function or however that works
i wanna say "P" but i literally haven't opened IDA in months coz I don't do that work anymore.
c makes it code, then p makes it a "procedure" if i remember how they named that functionality
I found someone selling a Alienware M17 with 64gb of ram 2x 1tb ssds and rtx 3070 for 650 buck s

first day since i started without any results
Hi vituha!

i add types and it converts unreadable garbage into different unreadable garbage
hopefully slightly more readable than it used to be
welcome to the last 5 years of my life
but it starts making sense a little in terms of "this place cannot be right the type must be wrong"
but
i really dont know if its just ida giving random bullshit as code or the type is actually wrong
what version are you using
uhhh

when it complains about not being able to allocate the local variables, that means you definitely have some types wrong, either in the local variables or in the args to the functions.
how is 2022 ancient 
what version did they add undo to?
because believe it or not hexrays has been adding massive improvements to IDA year on year
fuck so what now
decompiler doesn't tend to get those errors unless you mess up types badly for variables or args
it's called that because IDA at one point used to be console-based
they have ctr+w as save file combination
and ctrl s does some shit
this is not interactive at all
you can change the shortcut 
ctrl-w is "write" not save. it's easy to remember eventually
IDA 9.x also introduced a completely new shortcut default
press ctrl-w to write fancy text in twitch chat
will it really help me untangle this
i can just patch that part of navmesh 
like I said depends on what you're after
and hope it doesnt happen in many places
still don't know what you're trying to achieve
ignore
knowing what binary you have open and how the one you made in visual studio is related would be useful.
i am making skyrim neuro integration, i use native "show path to current quest" spell as pathfinding mechanism.
this spell surprisingly doesnt use player's model bounds parameters for its pathfinding calculation
it either defaults to what player's bounds supposed to be (because results are exactly the same as for default player bounds)
or it uses something else
and im trying to figure out wtf is happening there
have you consulted other Skyrim mod devs as opposed to reinventing the wheel
once you have found out what is happening, what are you intending to do?
to fix its inability to find path through clearly pathable navmesh segment
because it sounds like you're making this more complicated than it needs to be
how will you fix it? because typically people disassemble code so they can work out where to inject a one or two byte patch.
"fixing pathing" doesn't sound like inverting a conditional
i will replace the parameter it uses right before it calculates the path. or just overwrite it completely if its a constant for example
the pathfinding mechanism works
and you're expecting the player to... patch Skyrim?
it just cannot walk through 18 game unit wide navmesh segment when player's bounds are set to 22
i assume player can actually walk through shit about 12-14 units at least
and that place is just bad navmesh - it was put too thin onto a staircase which has no wall on one side
I may not know anything about Skyrim internals but it sure sounds like this approach doesn't make sense
i am patching it myself
its not that bad
uuh wait do i have it open
hold up
player bounds 15, easily pathfinds
wait wtf now it tells me i can walk with 22
if there is literally code which you can patch for this specific case, the developers have written code that is shittier than i imagine.
initially i was expecting it to use players actual bounds for calculations like this
and im sure npc's do use it
but this spell doesnt
the shit just crashed
i gotta go sleep its 7 a.m
i guess i still just don't understand what you're trying to accomplish. I understand you are trying to fix a pathing issue. But that would normally be related to the data for that specific mesh and not the method operating on the data. game data.
i can fix this place of navmesh
there might be other places like this
which i dont know of
so instead of fixing each navmesh separately, i would like to decrease player's bounds or whatever limits its using for calculations by like 20% and it will still be possible to walk anywhere like this
i think i found the part of code where it does these calculations, but it involves a bunch of huge shit
all i need is find place where its getting blocked by the path being too thin and do something to let it pass through that place
https://github.com/cassitly/neuro-integration-sdk/blob/main/neuro.go
Anyone could help me debug why my SDK is looping on the Connect function? I suspect it's the Go Reader Routine, but I'm pretty sure I did code it right.
shader for the daimond lattice pixel offsets seems to be working
now i need to get the depth stuff on there
How does it look when applied 
How do you know its working then 
the pattern in the image
int x = int(gl_FragCoord.x);
int y = int(gl_FragCoord.y);
float newX = float(x) + 0.5 * float(y & 1);
float newY = floor(float(y) * 0.5) + 0.5 * float(y & 1);
colour = vec4(newX, newY, 0.0, 1.0);
basic shadercode for making every other row be at a 0.5 offset in both direction from the previous row
i did some * 0.1 here and there to make it more obvious on the output
ye
Wait actually now you said you dont have hardware
yet
I have no clue what it is for

Is this not for a game or smth
its the volumetric display thing
it looks like it's for the UV map?
but it's a mapping from one coordinate system to another, with the tuple result being encoded in the red and green channel
yes
the image you send the the projector has to be 570x1140
but in reality its kinda like 2 seperate 570x570 images being overlayed ontop of eachother
tomorrow I'll be integrating the ASR and TTS. and probably the day after. and the day after. and the day after. for the foreseeable future.
urara
Im dying laughing... been debugging my gate model issues all day and after the last restart it suddenly started processing a ton of data... I forgot my mic was on and this is what it picked up
For reference this was playing on my phone
wa slookign at mobos and found this asrock x399 taichi
this is probably the coolest mobo design ive seen
some real steins gate type shit

i now need to make a Threadripper 2990WX pc to put this on
is 2990WX even worht it these days?
probably not
fastest ddr4 threadripper seems to be PRO 5995WX
ddr5 is too expensive
I'm gonna learn GPU drawing 

with that im now at getting the actual heights
Apparently IPv6's localhost address is ::1. How does one access the address from their browser? Is it possible?
oh ok
Yeah IPv6 URLs need []
oh wow that's weird
Didn't know that format works in browser addresses
makes sense that the browser address bar knows how to process it Ig
hmm I wonder
Okay so today I found out Python is scary
operator overloading 
so it based on screen reader, nice, thx, i just used as is
even discord dont know that they exist, i mean they add minimal support, but that just doesnt work properly now
man.....
gulpitty
wonder if system paths are saved as numbers 

its a 2080ti, and 2990WX
so pretty outdated
but
looks good
best 2018 pc combo
That has to be the most beautiful graphics card design I've ever seen
Oh this mobo too
asrock x399 taichi is roughly 200 bucks
i cant find that gpu anywhere for sale
oh damn
that gpu is decently overclcoked
its halfway between stock 2080ti and 3090 in terms of flops
ye it seems to be pretty rare
i cant find it
lol, most basic overloading
so many texts here lately
python isn't scary: it's trying to stop you doing stupid things. using the wrong path separator can lead to multiple issues.
I'll admit I haven't used higher-order concepts in pythonk... well not much at least
yea
I've been careful not to overwrite C:/ but I have seen it can become a mess if I try to handle pathing myself
all part of learning the lang tho
Average Python scuff be like
Python can be absoutly scary, I'd argue for paths that isn't the case but as a whole the ease of use also has the abillity to require some unhinged requirements
I am abusing of operator overloading myself, namely for custom operations.
Python is kinda pain, I'd much rather write Kotlin
I do agree at a point you should figure out if you want to switch
why use python for anything complex 
just use compile languages: c++, rust, java-analogs,... even typescript is better 
Some people just rely on libs
I can't say much either as I'm working on a large project in python. I've already said my case multipule times now so I'm not going to repeat it.
C/C++ written python libs works for most prototyping
that's my domain currently
I was thinking of moving onto C# after having sort of established a sensible understanding of programming starting from python, but seeing MS's recent actions, it kind of leaves a bad taste in the mouth
I might just go to C++
after python id recommend just look at c++, but learn rust first
rust model is great, tho limited compare to c++
so i use c++ ;)
C++ is decent, I think if is a good choice. Nothing is perfect tho
Like I tried C# and learning it was alright (capitalization of methods is interesting), but the masters of the language (MS) leave some concern as to where things are going with the operating system that it is presumably made for
yea (to the last part)
I'll see what C++ is like
righto
there is gay-moders server together c&c++
big online and active support chat
but fuck their moderators for real
Interesting
sovl
I want to try avoiding relying on people's guidance and build up my understanding of the language through docs, so I should be mostly free from the hassle of community drama
issue with cards like this, is that no one will actually look at them. They are front-down
yea
keep in mind 2080Ti flops is 2:1 fp16:32
vertical risers
how many of them you've seen?
it is wrong way for c++
It is?
also, @fleet jungle
Large Language Models (LLMs)
We highly recommend against the use of LLMs and AI assistants because:
LLMs are bad at C and C++
LLMs are wrong more often than not
LLMs answer with complete confidence even when wrong
If you're new to C or C++ you likely don't know enough to know when answers are wrong```
it applicable to every lang, but especially for c++
for books ask on tccpp, server can be found in discord's discover
(just pirate them, its free)
Oh that's nice. More of a challenge
well

idk if that actually makes it worse or not
I was at the Anthropic presentation at cppcon and the presenter revealed that Claude treated C++ like python with different syntax until 4.0 or something.
I don't rely upon the LLMs too big, but there have been moments where I've seen LLMs suggest using techniques that are completely wrong or outdated (using webGPU, for instance)
Completely wrong in that they suggest methods that don't even exist
my key point, no any ai code after <>
i use ai only for initial search, if i not know keywords for example, or if i want solution for problem i never encountered
BUT always read actual human-written text, dont believe ai right away;
i lost 90% of performance because i believed that a function was amortized, just used without checking docs at all.
but it actually use 3 syscall and read a couple of chunks of data on every call,
even when given args are exactly the same.
and that function was called 3 times every 'frame', up to 300 fps
and that wasn't only the case, it is what i will say to anyone - why i will never trust ai in anything
I just don't see them as useful for any larger-scale guidance than for learning what an individual library feature does
same
This is my own usage as well
The moment the LLM starts building its own function I ignore it 
I think most devs use it in that fashion
a l o t
Because hell no I'm letting an LLM build an entire project for me
i actually never
i almost always see one
It was super funny to watch her demo Claude's planning ability, meanwhile ClaudePlaysPokemon was stuck in Erika's 5x5 tile area during the entire keynote and left without attempting to fight her.
Though it could be argued an LLM can just be replaced by a really good search function on language/API docs
Some docs leave much to be desired though
honestly the more I learn aboutalternatives to using LLMs, the more they seem like an utter waste of resources
btw, i also sanity check myself with ai for logic-only (pure logic, no business) functions
it never say its right, it always say it is wrong, hallucinate several errors right away
just because prompt is 'find error here'
if i'd change prompt to 'what this function doing?' - it ignore most obvious errors
usually i just find 1-2 errors without running function or even re-read it myself, took several seconds (excluding opening ai website) but save little time for obvious errors
Copilot is...
I miss when it was just Sydney and a fun entertainment machine lol
At least that would have been more useful, granted not by much
on our python part of codebase 
i think he just wasting time...
but it is internal tooling, so i think its fine 
The thought occurred to me "what about LLM autocomplete" then I remembered there's already autocomplete without wasting heaps of electricity
wait what do you mean internal tooling 
that could be really bad
i mean.. internal tooling...
it is separate from actual code that we run on robots
it runs only during calibration and for visualization
and also they now doing something to train nn just on ongoing stream, but i not sure how much ai they use now, prob not copilot 
and i have no work to do until my coworker finish what he doing. And than we will go through entire codebase just to sync our understanding of everything
cuz i gonna leave country... i hope
nn is good for classification and prediction
it just bad for prediction within enormous context
and language is enormous context
I was just thinking if you meant like an LLM that can snoop in on project matters, which would potentially be a legal/intellectual property issue
nn isnt only about llm
yep
we use for classification
btw, we run it on cpu
model is light, and for mass production it is much better
I've heard of people doing that
we just preprocess everything, like
1 2 3 4 3 2 5 3 2 1
will become
-10 -5 0 5 0 13 -5 0 -5
(made up numbers but i hope you get a point)
It did bring up the question of performance losses as a consequence of having less cores to do the math ops
interesting
we have a lot of preprocessing
we absolutely sure that, like 70% of inputs will not lead to classes
so we mask them, and nn work with only 10-30% of context, up to 90% load, but if we detect slowing, we just drop some input frames
so actually nn is just replacement for all other ML variants; it just much easier
it possible because overall context is predictable, we just want classification, that impossible to hardcode in hashtables or anything
ah
i got the compute shader to work
this is all opengl rn
i cant be bothered to do this in vulkan
not yet anyways
Where is the nuked music?
give me a sec
Your one second is up 
3d is scary
tried to learn, and more i learn - less i understand
easier to use engines that other did 
Now this is the Sam video that I know lol
also, there has to be a more efficient way of doing this
my compute shader is doing layout (local_size_x = 8, local_size_y = 8, local_size_z = 8) in;
so that's 512 threads per compute shader.
and then i run that shader with glDispatchCompute(72, 143, 18);
so thats 185328 times it runs
185328 * 512 = 94887936
this thing is doing a whole 90 milion compute shader thingies per frame
they had ray tracing and smoke compute in 2010 btw
on 32bit arch
like this thing runs at 840 fps on my 3090.
thats absolute ass
its probably gonna be faster on vulkan, but still
Well, that's how it is with volumetric anything no? The added dimension means you can't do much trick but to perform the actual computation
idk
there has to be a better way
Surely
rn i did RGBA32 cuz i copy pasted code, i should jsut make it 1 bit
cuz 93.5M * 16 bytes ≈ 1.5 GB
1.5GB per frame

oh actually, its using 73% of my vram

18GB
"some optimisation is required"

if i make it 1bit instead of RGBA32, it should do 11MB per frame
that more manageable

how many frames do we need 
this type of shit is the excact reason you shouldnt optimize games for xx90 class gpus
180
oh no wait
i need the compute shader to run at 30fps
but the fragment needs to do 180
but
the fragment is incomplete
that fragment is only doing 1 bit rn
i need it to output 24 bit of different time intervals
coding hard
green leaf quick summon a snail
We just have to have all of #programming optimize the shader as much as possible
yes lets start a revolution
down with AI upscaling
(I'm joking)
im pretty sure that making it not use 70% of my vram, and not doing 1.5GB / frame will help massively
somewhere... I'm bad with jokes 
im suprised it was even able to handle 1.5GB / frame at 840fps
no no your fine, I say we actually kill AI upscaling
ahh
thats 1.2TB per second of data 
sam fps triangle right here
memory bandwidth of 3090 shoulkd only be 936GB/s tho
picture that but with evil instead
idk
it would be pretty funny to eliminate igpus from the race by making it use insane amounts of vram
*dgpus
just for a second idk
igpus have lower bandwidth tho
If you wanna optimize it, I've heard what many people do (according to YT programmers, so make of that what you will) is they make algorithms to simplify very similar colour pixels into one cell
I forgot what it's called
dgpus would be killed more by having to offload than igpus that have it as a constant 
we cant really do that
but we'll use lower bitdsepth
righto
i guess low end dgpus will be eliminated then
and high end ones will survive if they can keep up
I'm not getting out of this race, you'll have to drag me out
idc I have an igpu, I'm still going to gun for top placement 
gpus are scary i still don't want to figure out opengl but i have to
Sam can you confirm 
i will have to use it eventually but opengl already goes over my head
900 lines for hello triangle isn't a good look
I'm not sure how many lines #programming's triangle is
people thinking Vulkan is hard again 
(it's all just boilerplate)
you do have to care about explicit sync but that's kinda it
yeah but boilerplate is hard
all that matters is that it runs at 60k fps 
i will have to do some optimizing of things eventually
i also have to like figure out a ton of other things
and i don't feel like it
I haven't touched vulkan, but I feel like starting from Sam's triangle into a entire 3d scene is easier than dealing with my neural net project 
i think entire graphics is just adding new (for you) stuff and refactoring
tho i did just basic stuff in opengl
even mesh loading was just copy paste...
is it like rendering or just fetching frame data and display
who is sam and why does he have a triangle

It's probably all cached
Oh wait
This isn't about the triangle 
is it fr using 16 bytes or do you mean bits
bytes probably
RGBA32 = 32-bit float per channel
god fuckin damn
I'm not sure if that's what Sam actually used but it does exist IIRC
8 bits * 3 channel = 24
sam was commenting on fixing that himself
no display output that shit
hell you wouldn't even need alpha
alpha is only used if you're rendering something with overlaying
maybe when im getting opengl to work ill ask here but i need to work on another thing first which will be a little harder

i can only think of video editing being the best usecase of RGBA32
but its still stupid lmao
yeah
I've had cases where I want textures/buffers with values more precise than 8-bit but that's never actually storing colors like that 
HDR exists yeah
few things can do 10bit yeah
the 32-bit i've seen before has been RGB float32
rgb 24 is probably the best because 0 alpha is usually black i think so you dont really need to store alpha aside from like telling the gpu what to do so nevermind i guess but like you do wanna have thigns fade in and out probably even in games and otherwise idfk
which is for HDR
im not good at graphics
hmm
If you're optimizing, I wonder if doing it in consideration of the human vision range might make it easier, though that might be higher level and irrelevant
idk
but i bet you'd also want the black to be 100% opaque.
we'll just get nanobots to dance very quickly to make 3d images
I doubt human vision can see a difference between complete black and maybe 0.001 brightness
I could be wrong tho
or in the current case, spin a boi very fast and have it reflect everywhere
human vision is really good at detecting contrasts, like edges or single bad pixels
mhm
but yet we don't have shrimp colors :(
shrimp can't focus like we can

skill issue
what is tool to project shapes on globe, that mapped on world map (correct sizes)
i absolutely sure i saw one, but cant find now
trigonometry
that'd be the tech called uv mapping in general
depends on what projection the 2d map was

like a website? Or a library?
website or app
do i have to code it myself? lol
i think you can ignore all but the "a" from the middle 2 lines
do shrimp go beep when they detect prey?
I feel like calling code projects dedicated for specific tasks "libraries" is a little overspecficied. I'd like to call them books 
what if the prey has a coupon?
Do it for the fun of it, but there are mutliple websites already
Truesizeof comes to mind, but idk if it will let you upload a custom image
i need to draw shape, specifically just direct line on globe, and map it to 2d world map
what a great time to eep for a few
I just realized it said something about barcode scanning
wow.
This bot is corporate-minded, huh
Its doing 570x1140x144 texture basicly
Hi
ahh I recall something like that
That would be me
hi sam
Harrowww
nice triangle you have there
16 bytes
maybe search globe curve plotter or something
32 bit per colour
more green leaves, send in the snail army instead
nope
There's this... but paid apparently https://www.amcharts.com/demos/map-with-curved-lines/
eh
You can probably have an easier time just writing one yourself
1 bit color
yeah probably
it do look really good
lol
you can share it around
i'm not stupid
Its just a bool to see "here is stuff"
bool for ray hit 
Its not a ray but sure
screw it, dithering instead
We cant do raytracing since we dont want closest pixel, but pixel at this exact location
Hence 3D texture
i should do a game in 1 bit color but i dont know how to and i don't really care all that much
one of these days
if you just need a prebuilt website then surely this works
https://www.greatcirclemap.com/
otherwise I think you'll have to calculate the great circle route some other way and plot it with MapLibre or similar 
Great Circle Map displays the shortest
route between airports and calculates the distance. It draws geodesic flight
paths on top of Google maps, so you can create your own route map.
1 bit colour means 8 colours.
1 bit. Means white or black
1 bit per bit
that's what is meant
ah
1 bit colour means red on or off, green on or off, blue on or off
8 colours
I love that a small part bigger than a byte is called a nibble
nibble
I am stupid 
you should use SDF 
This is SDF 
soft drop factor
oh it's smaller not bigger

mb
i had a hyperfixation for 6502 assembly for a while and i still am good at it but i dont do anything with it so i don't know
then surely you can do raymarching if you need some 2D projection
Rn it is at least
But im gonna make it 1 bit
Instead of an int or float
bitmask 
harkening back to the days of old with a limited colour spectrum sounds pretty fun and tempting
No.
We dont want closest pixel. We want "is there shit at this exact location"
'how optimized can we make this', is what I like to think

ye then you don't need SDF
although it would probably still be very efficient if you store the function
no memory access at all then
Its pretty much a game of minimizing memory reads and writes
yea

you can do that without touching VRAM?
I summon using AVX registers as storage for the lols (this is not recommended)
just always hit cache 
oh right there's cache
yeah, pretty much
if you compose whatever you want to draw using functions then you can recalculate it instead of loading from VRAM
should all be in registers 
The problem is that the thing im actually rendering to the screen is just a square. And for that square we read the needed value from a 3D texture.
We cant do any polygon stuff in the fragment shader
Thats why we're using a compute shader
what's the 3D texture for? does it just contain the square?
would be fun (maybe more on the hardware-engineer level) to try restricting to monitor cache (if that's a thing) only, so nothing is stored on the GPU in any way
It just contains a kinda voxelized version of the 3D model i guess
Fuckin autocorrecy
vocalized

No
i love registers
The 3d texture only changes at 30hz, the fragment shader stuff does 180hz
And thats 24 reads per frame
I wonder if anyone does that yet - cache directly on the display peripheral. Might not be necessary if the current signal rate matches the max refresh rate of displays
Since were doing 24 1bit buffers into a full rgb24 bit output through hdmi

SDF shows distance of a specific location to the object
How do you calculate sdf without geometry?
function that maps position to distance
By sheer technicality, you can
see: https://www.shadertoy.com/view/WsSBzh for what you can do with SDFs
// The image is a single formula, but I had to split it
// down into 3 passes here so it could be shared without
// breaking the WebGL implementation of the web browsers
// (which is what Shadertoy uses to run the code below
// that implements the formula).
Im not gonna make a shader per frame of animation per 3D model im using
ye, you could, but realistically you wouldn't
you'd just have to compose your scene out of other SDFs, that's kinda the main downside 
Ohh I see what you mean. But does the fragment shader computed globally or on the object?
that's animated lightly in the script itself. doesn't need a shader per frame
globally
you just have a fullscreen quad, the fragment shader does everything
This website absolutly does not wirk on my phone
it uses WebGL. probably needs a real computer.
Thats what we do rn.
Except i dont see how you're gonna do SDF with polygons in the fragment shader
fck, they support codes only
why dont they just give coordinates
nah, id rather do it myself later, just cant find tool in internet
what blows my mind is that you can do this and it'll run heaps faster than any top-of-the-line GPUs in the 2000s, even if they were assisted by cache. PC hardware has come a huge way
yeah probably just make one yourself then
there are a libraries that can calculate the great circle route and give you a list of line segments to draw 
yeah, PC hardware in general is just insane now 
Like, the 3D geometry has to be done in a seperate shader no???




that's ancient





