#programming
1 messages · Page 430 of 1
both would wash candly floss in water
This is NOT Unreal Engine 5.
This is Godot RTX!
Running at 60FPS+ on a base 4080.
Path tracing and DLSS ray reconstruction have come to Godot via Nvidia's fork
Not too shabby by the looks of it
Apparently the actual framerate is roughly 80-150fps in the editor in their (presumably unoptimsed) test scene
Mistral 4 series incoming 
so update on last night's SMB misadventures
60TB of files

audio recorded at a healthy 18mbit/s
Sounds great, what's the catch 
have to beta test the palantir reality altering vision system for htem
@real sierra amazon has never forgiven me for looking for a new bio for you
3 items are things i could care about here
more bio suggestions
moon
funyun is this audiofile enough
what were you recording, that's like the bandwidth of trueHD
blu ray checks out that's about the highest resolution you'll get for audio over hdmi at least for 2.1
technically i think it goes to like 37
but that'd be like 8 channels/24bit/192khz
anyways that's a lotta empty spectrum
that being said i am all for it because 
is this the program that was annihilating your system
My guess is fused pipeline, so all or nothing
https://www.youtube.com/watch?v=4ZlwTtgbgVA digital foundry has a video on it already 
Digital Foundry is now fully independent! Join the DF Supporter Program and support the team: https://bit.ly/3jEGjvx
Coming in Fall 2026, Nividia's DLSS 5 is a brand-new technology for RTX 50-series GPUs, bringing next-gen quality, photo-realistic lighting to any supported game. We went hands-on with four existing games running with DLSS 5 in p...
The catch is it looks like crap without rt enabled
it seems to be doing surprisingly well in starfield for example which doesn't do rt (there's still a lot of optimisations necessary to make this run on anything but a 5090 it seems
)
idk where you got bluray from
I think this was direct from a hyperdeck
nop, that one was me trying to concurrently list 40k folders on an smb share
p
d
b
e

"i see the b, the d, and the e, and i say to 'em "why?" oh. the p. "
I was wondering why people were typing individual letters.
worked it out.
e


Features: Power Voltage Converter Step-Up / Step-Down Transformer Universal AC Power Supply Conversion Rear Panel Voltage Selector Simple Electric Plug-in Design Built-in Fuse Protection USB Device Charge Port Includes European EU Power Adapter Plug Grounded Outlet: Accepts 3 or 2 Prong US / Euro...
always
11 (gmt)

YAYAYAYAYYAYAY
Guys, any advices where i could learn everything about Machine learning and etc
Seeking some guidance
Chatgpt and YouTube
Ehh
I guess if it works it works
However i would appreciate some ppl with experience lmfao
That's why am there
Chatgpt will give you the textbook, youtube will give you the people with experience
depends on your definition of "hack"
real hacking in the modern day is more or less jsut social engineering
I guess it works
Alr
Thats the boring kinda hacking, for sociopaths. The better kind is hacking devices to get homebrew which is very much active, or pen testing.
"for sociopaths"
thats just how its done
doesnt mean you have to like doing it
most dont

Btw is it worth learning social engineering, when i was way younger, i had a "dream" or a wish learn that shit someday
There are ways to exploit the wetware without social engineering...like password or username reuse or the like.
But exploiting bugs im software or systems is much more enjoyable.
Found an exploit in a system my employer was using where the upload folder had script execute permissions, so you could upload a php file and run it on the system.
uhm, so...
unless you're legally allowed to because of working at x company. dont
would X company be in rural india perchance
Social engineering is just a fancy way of saying "lying to people"
Kk
Thanks
I'll keep it in mind
people who know some form of dark psychology are just generally less enjoyable to be around
whether it’s practical or not
I hope so, im building one now.
and learning in general is quite fun
From the scratches?
Yep
Hmm, interesting
the problem is when they start to use their psychology tricks in their everyday lives. its kinda like manipulation in a way
That is exactly what it is.
When i bought mine pc with 5060ti 16gb, one of the first thing that i did
Is downloaded chatLLM, investigated the dolphin/basic AI models, hooked them with discord API's
And chatted with them by phone
It felt kinda entertaining
saw the bitrate displayscreenshot you sent and automatically categorized "larger than dvd" numbers as bluray lol
However idk what else that could be useful for
Any ideas lol?
hi rezak
Hi dude wsp
How r u doin?
I feel like funyun is the type of person to confidently be able to prove whether the chicken or the egg came first
Probably u're not wrong lmfao
but then just forget
You might have it backwards, you dont get an ai and figure out what to do with it...you use ai to solve a problem you have
The egg came first, laid by something very similar to a chicken but not one. Cus thats how evolution works.
believe it or not, that not-chicken? albert einstein
He was NOT a chicken so I see no flaw in this logic
Hmm kk
trying to stay ahead of the latest parts shortage and otherwise avoiding my real responsibilities but comfortable hbu
as long as im still an onion flavoured ring im happy
Trying to move on, after "heart-breaking" shit, then trying to find myself in ML
And grinding english for better ielts to probably study in Coventry University or smth simillar
your english does seem better than previously actually
Is it?
yeah a bit
I haven't practiced in a while
not that it was bad mind you
I was just watching yt eng content creators
Nothin' else
it was just clearly from someone who originally spoke a language which was more direct or something
I bet am still making same mistakes over 'nd over
you are filling in more of the minor things between words you probably would've skipped before
either tha
or
i am schizo
that's also very possible
Could be lmfao

Idk, time will tell.
we literally just read a study by someone in coventry university in school
Porque no los dos?
little column a little column b
non english 
Lmfaooo
Coventry University is now in 🇰🇿
they have like 1-2 buildings in Kazakhstan
Memes are the universal language
That's why am trying to getting mine ass there
I heard it's pretty goated
And very strict
Also they told me that i need at least IELTS 7.0 to even pass there
Y porque no es posible que empiecen lo mismo tiempo
my spanish is horrible ignore it
English please?
its not possible for them to start at the same time
Its ok, I dont speak Spanish
i manually type into chatgpt
google translate says I suck at spanish but wtv
"I use AI to type to chatGpt, Thoughts?"
Jk
The phrase "porque no los dos" is from a meme taken from a commercial for old el paso.
Sry idk how to react to that.
I think am outta of context
this is the chicken and egg convo
#baking strikes again
putting the voice modes up in a couple phones next to each other with like chatgpt/gemini/claude all talking to each other
@fast pagoda yo btw i forgot, how's ur birthday went tho?
If am not trippin' it was june 28 or smth
it is currently march
shiro mentality
Lmfao
i mean, you were very close on the date (3 days off), so even tho it's march, it went great
lol
Btw
The english "improvements" might been caused by talking to a YT "GKEY"
Valo NA content creator, we've had decent conversations past months or smth
im im amazed someone would remember anything about my birthday of all
25?
hai
Lemme check

And Shiro has 365 entries (366 some years)
then there is the person that has 1 entry every 4 years
at the age of 5 i was born without a face
The joy of changing Linux distro
dear god
Also the place I'm copying to is a mechanical hard drive
no.....
I'll tell you when it is done :3
200,000 already
Wait the estimated time is going up
ran out of cache
Makes sence
holy crap just realized Cathy from Until then
Also I don't need 99% of the files I'm copying but I don't wanna miss any of my game stuff
wha
awwww man its too late
Idk what you are talking about
just the-
Just what?
oh nvm i guess
You could've spent those 2 hours playing the best story game ever
I just realised the files in copying includes the entire until then source and the entire press kit
I don't need that stuff
Its only like 8 extra GB
next time do yourself a favour and make a separate home partition
How do I even do that??
As said I have the press kit and the source code (I had to find where I put it)
Oh yea this video contains flashing images
the idea is instead of creating two partitions, one for the boot and one for everything else you make 3 (a dedicated one for your home folder)
can also put use a different drive i guess
and then just mount that partition under /home 
means you can share it across installs
How would I configure it to use a different home folder?
I didn't know that was possible
it's just another folder, you can add an entry to /etc/fstab with it as the mount point
You can also change the home folder path with usermod if you wanted to mount it at a different location for some reason.

a bell!
the only real downside is if you're using a single drive, every partition has to have a set size so it's a bit less flexible
But I have 1,214,019 files to copy until I get to install fodora
and i guess you have to do the partitioning manually
I only have a single drive but I'll probably do it anyways since its not like my home folder will get that big (my current one is only 1.4 million files)
I think tor is currently having a mini heart attack because discord isn't working too well
you can mostly avoid this by using btrfs subvolumes or, if you don't like being tied to btrfs, LVM
but makes the setup even more complicated 
Turns out I downloaded the wrong version of fodora so I'm glad I got bored and noticed
it wouldn't help to the point that the transfer would be what one could consider fast but i'd use rclone because it does multithreaded xfers, if you have a lot of small files
another thing is putting everything in a 0 compression .tar of some kind which makes it into more of a single file transfer which is much faster
NVIDIA announced new slop on twitter
Oh yeah
I saw
People are not happy
live slop filter 
It looks interesting, I wonder how it'll look on release
NoVideo
There's not even info on what hardware is going to support it. Maybe just new gen, maybe not
I do think it has potential for a very specific type of game
The games from like 2005-2015 which tried to be unreal engine 5 hyper realistic without unreal engine 5 hyper realism really being possible
I do think for those in particular it could be poggies
Considering that's the case for the base game too I'm alright with that 
Blackwell only since they're the only ones with the neural rendering hardware
I mean it seems to be a trained model like rest of dlss, no reason for it not to support different styles of lighting generation, etc. depending on a game
It's funny how plain the base game looks in comparison
Would be nice to have a path tracing comparison
me when no ampere (i bought the cheapest laptop i could find): 

ps6 cancelled, take your ps5 pro plus
speaking of the ps5, they announced PSSR 2 apparently, no clue how good that is
Wtf is pssr
Ps5 pro max plus with a $100 ac adapter
Communist playstation???
Will it have 3050 inside 
it's a software update for the ps5 pro 
hey, at least chatGPT seems to respect the robots.txt now, it never used to
And people call these models yappers
I bet it still spat out 1000 lines about itself
It didn't
Of course most of the top scientists that made and released these amazing models were all fired or quit the other day 
they did?

@grok summerise this for me
does anyone know if there is a public api for the lava lamp color i can use for a projekt of mine
Summarisation is just the absence of usefulness
-# sometimes
there is not, your best bet is using computer vision to extract the colour 
I think it is but naturally it'll sometimes end up obstructed so there's that
There is not much stopping you from decompiling the android apk and making a wrapper to forward the received data to a flask api
Idk if it’d try and force you to pair to an actual lamp
Actually
I’ll try it out later
default settings tend to start a whole story by itself if you say something a little too unspecific
The unsloth ggufs are known to be busted for Qwen 3.5
of course "hi" is an extreme example, but it sometimes decides to go on a 3000 token self-gotcha "but wait" internal rants
the unsloth ones work fine for me, just had to tune probabilites and theyre fast and smart
Tbh this behaviour isn't overly surprising though that aside
The length of the reasoning process is based on confidence after all
Id bet my organs its only for new gen
with a system prompt and probability adjustments it does really well, ive gotten it quite focused
havent tried the official releases because they didnt have ggufs out when i set it up
I use bartowskis GGUFs for Qwen 3.5 given the reports of issues with Unsloths quants
i'll check em out then
I will say the one unfortunate thing about these models is that the sampler parameters which net you the most precise answers also cause the occasional looping issue
Its not a huge deal to just regen when that happens though
Duality of qwen (first is the looping issue)
can adjust penalties for that as well, though it hasnt looped for me with my settings and default lax penalties
I keep repeat penalty off since that's their recommended preset for precise answers
The very occasional looping is a small tradeoff for higher quality answers imo
quality depends on what youre asking for, outside of code i think adding some penalties is beneficial, because if it goes off the rails itll shift the probabilities too much and tell your garbage anyway
@mighty thorn can i dm you so we can maybe figure it out together
but i guess you can keep regening until it gives you the right thing
hahah yep
I try later
But I’ll dm you
So yes
okey thanks
i basically only really use it as a jumping pad for research elsewhere, finding obscure shit or solving one off trivialities, its not useful for anything serious to me
Problem is sometimes it needs to refine through iteration and penalties throw that process off
to me llms are just lossy lookup tables of what's on the internet
yuh, cuz thats what they really are
its not really refining anything, the reasoning is just producing arbitrary values that shift the probability distribution towards a particular area within the network, which can sometimes be closer to what you typed in
it shouldnt be too short nor too long, but the content within the reasoning is basically meaningless
See you'd think that, but the current consensus is that the "aha, hang on a minute" moments actually do represent a shift in the outcome
Ehhh
It often gets It stuck into loops
Its just feeding itself junk in the hopes of producing an accurate answer
it doesnt really, its an artifact of the weird ass ways they train reasoning on these things
Why waste tokens when u would get the same quality anyways
Good enough prompting and wording will provide the same result
"aha" is it getting more stuck within the probability space it already got to, its just the way they trained the reasoning makes it more prone to do so
New opus has that issue
the quantization mangling doesnt help of course
its a hack to improve probabilities for shit prompts, which works for most stupid questions people give these things
There was a paper that came out recently which specifically dispelled this assumption
I wish I had it on hand right about now
I mean its essentially just reprompting itself
half the llm papers are written by people who think they invented god, i dont trust them
More specifically this is the myth it disspelled
Well that's because all the real researchers moved on
Its still pointless if u prompt it right
Plenty of papers show that the chain of thought can just be bullshit by the model
(As its not thinking its predicting)
So why waste the money on the tokens
yeah sure, theres plenty of other papers saying the reasoning phase is dogshit and not representative of anything, its all very specific to the proprietary training processes anyway
theyre probability blackboxes and everyone claims the opposite things every other week, i dont care that much
Here we go, "Despite this, inflection points (e.g., backtracking, 'aha' moments) occur almost exclusively in responses where probes show large belief shifts, suggesting these behaviors track genuine uncertainty rather than learned "reasoning theater.""
the industry right now is interested in convincing investors to give more money
You can get bullshit overly long CoTs where an early exit nets the correct answer
But that doesn't mean the reasoning is inherently all bullshit brute forcing
I mean its all bruteforcing by nature
But the question for each model is how worth it it is
Like with opus currently its REALLY bad
And can drone on for long periods
I think sonnet did a better job but it would still have its hiccups
the anthropomorphic language used there tells me the paper is nonsense, it likely just bounces out because the endless "oh wait" meaningless tokens take over the meaningful tokens and shoot the probability into a random direction
theres no belief shift or whatever nonsense
Its anthropomorphic language but not ascientific, its just abstraction
they dont think
Fills its own context until the probabilities shift a ton
using anthrophomorphic language for a statistical black box is as unscientific as it gets and i dont understand why people accept this
This is fair to say though, the probability shift all those tokens create is an inherent part of the process
So why waste all the money on essentially padding tokens?
Anthropomorphic language isn't inherently useless or unscientific
Context is key
I think as long as u provide all the data upfront ur odds will be at their best
for this tech its unscientific
there is a vested interest in making them appear "smart" to extract money from people
Legends use "machine learning" for a reason
the majority of the "research" papers reflect it
Its click bait for research papers honestly
"belief shift" is nonsense, call it what it is, a generation probability shift, which is boring as hell and obvious
Because in low confidence or inaccurate scenarios reasoning chains give the model a greater chance to course correct and shift into a state conducive to a higher quality answer
I doubt there is a proper cost to result comparison out there
Yes like with most computers time will give better results with many types of calculations
its not necessarily a waste of tokens if done well, the pregen phase should pull in adjacent data/probabilities which can result in the main gen being closer to what you needed
enrichment basically, theres no reason it cant be done inline, but looks better if you dont see it of course
I think Gemini does it the best because it doesn't use it for long
since enrichment can technically be arbitrary garbage, as long as it makes the probabilities better
Models like the current opus generally do it the worst, massively overdoing what should be a simple generation for a small prompt
if you ask it about say Python code, enrichment could dump tokens related to programming in general, which wont make any sense but would get you a net better outcome as it would pull stuff in from further out
Oh btw @split steppe apparently unsloths March 5th GGUFs did fix the bugs
So if you downloaded your quant later than that you're fine
ive been using the latest ggufs so should be good then
Its just an easy way to explain an abstract concept
I mean thats why i think mcp is a better tool
Its only an issue if you have AI psychosis and genuinely buy that the models are sentient
The issue is that its being used to push snake oil
we should be careful about accepting intentional antrophomorphization of mathematical concepts we already have clear definitions for
because its not just "making it easier to read", it later is used in courts, when people are convinced it really do think
Marketing vs technology
most people using this think it actually thinks for real, programmers included
it matters how we talk about it
As much as I hate the whole online "ermm actually" its kind of needed for this tech given where it is
you literally cant buy ram and ssds right now because people think these things can think
Its essentially the programmed "fake it till you make it"
I learned about it back in 2015 to use for game anticheats
Which is still probably one of my favorite uses for it
I think i just used an LVQ
putting lipstick on the mess that is the web data
Kek
I will say I've been suspicious of Unsloth of late after someone sent me this screenshot
Yandev tier
well anthropic did burn an entire library of books actually iirc
lmao
thats all data science code actually
matlab fans
you have to remember these people arent programmers, you gotta remember they are academics
If pytorch or tensorflow got abandoned they would drop to their knees
its literally just fancy data analysis
Do people still use tensorflow
Or is it all pytorch now
I remember the shift around 2017-2018
i did a model with my data science friend some years ago for a product we tried to make, its all disposable code only serving to put data into a model and then get data out of the model, there is nothing there to make nice
Gpt2 was around for quite a while
It wasnt until 3 that the general public started hearing about it
I think i ran gpt 2 locally back when it was released
people use whatever gets paid by <vendor> to make the proprietary gpu/tpu shit work
A decent chunk did hear about GPT 2, but they only knew it as AI Dungeon
Meh that was a bit later
i followed it since gpt 1
But yeah I remember that did get some devs into it
People were fine-tuning a lot
(Though often relied on Google colab)
ai dungeon was nostalgic.
technically a little earlier for the industry in general, i was watching where it goes since google deepdream
remember when nlp uses rnn

They were really cheap to train atleast
I remember it used one for a swear detector
It was good for things that didnt need to "respond"
I need to dig up my own anticheat codes
Some of them were fully built from the ground up
(Mostly because it was for a big network)
nn based anticheat sounds interesting but i wouldnt trust it and surely thats gotta be expensive
Better than current slop
Spyware ring 0
i think the devs are just lazy and outsource it
yuh
A client can lie but a server with an anticheat will flip them the bird
write your game server first, maximally processing there and the client should just render state. state inspection stuff like wallhacks are harder to beat, but you can for sure infer cheating from basic statistics most of the time
assuming fps and such
most other games i dont think need an anticheat to that level anyway
Client side cheats will be harder to deal with by nature
Visual hacks and such
But anything else is game
Riot vanguard is one of the shadiest pieces of software people willingly allow onto their systems, it's insane
As someone who has built anticheats professionally for 5 years i consider that shit actual malware
Its just pointless
im just waiting for solarwinds 2.0 but this time its gamers
vanguard, denuvo, eac, here we go
Don't get me started on those
They just dont work
And open a door for hackers to rat you
They also fuck with performance and device support along with the fact that they dont take long to crack
we should let people votekick, this is why i like valve more even if its full of cheaters, people who want to play just kick the cheater and we're good
Some people pirate games they already own just to get denuvo removed
vac is trash but it does sometimes also detect cheaters
Vac is old
Vacnet is the tech I work with tho
Idk how much they actively update it
never actually looked into it much, i just know it barely works nowadays
They probably need to make a source 2 version
you can sometimes spam ~100 reports to vacban someone lol
The issue with generalized anticheats like VAC and others is that they heavily rely on how they are implemented
Alot of the implementation on older games is probably more dated than actual VAC
Making it hard for it to properly do its job
mmm, interesting bit there
Alot of anticheat work is saying "what is every possible way a person can do anything" so u can block everything out of that bound
generally i myself dont see a point in general anticheats at all, id rather put in the effort in making server as authoritative as possible and then having good community tools
Thats what Minecraft does
Gmod has server anticheats too
The issue is alot of these AAA games dont give users a server file to run their own
not sure thats needed to do it well
Unless they give us packet level control then it is
aaa servers are also real complicated pieces of work sometimes
They are usually really badly made tbh
Alot of those games u can have hackers essentially take them over completely
Or worse in cod games where they can RCE you
maybe, but also imagine needing to handle 100k DAU in a realtime game, they gotta have complex setups for some of these
The orchestration tools are fine im just talking the actual net code for individual servers and clients
which is often tied with the infra stuff 😛 im sure theres aws deps and such just assumed in there
They need atleast some server side validation though
minecraft in comparison is dead simple to aaa stuff
Well Mc has its own net code issues (though not as bad these days)
but then again hypixel would prolly like a word
they also handle 50k dau or something at their peak
I know alot of their infrastructure due to some old friends
They use docker for some stuff but their general systems are ran off of salt
wdym by salt
If I recall they only use docker for the vanilla private servers
ah so smth in line of terraform or ansible
Then everything else is salt
mmm actually maybe more like kubernetes it seems, or bit of everything
interesting, gotta read more on it later
From RE//verse 2026, watch James McGowan and Bas Zweers' talk: Deobfuscation of a Real World Binary Obfuscator
Talk Slides: https://github.com/AnalogCyberNuke/RE-Verse-2026-Slides/blob/main/Reverse26.pdf
We analyze and deobfuscate a real-world commercial binary obfuscator used to protect several pieces of important software including Windows K...
back.engineering always has some cool stuff
Some deobfuscation work on Riot Vanguard’s kernel mode component
Given these folks consider it fine I’m on the side of considering it OK too
Which vendor did you work at?
whew
20 PFLOPS FP4 
748gb mem 
it's got an interconnect but technically it's 252 gb hbm on the gpu and 496 lpddr5x on the cpu

cpu mem bandwidth is 396 GB/s while the hbm3e is 7.1TB/s
Infrequenrly accessed weights storage 
DLSSlop cuz wtf is ts dawggggg???
So like 2.5-5 PFLOPS of actually usable compute
Working on a hunger games project.
Working on map generation.
I can't figure out how to make this river swing right or left. Only goes left.
(Terrain Type | Terrain Height).
?tag
huh
gonna be 10pf dense fp4 so yeah about 2.5pf
that's such a massive tensor array
V100 is only 125TF fp16
312TF on A100
990TF on H100
well i mean its decent
but i personally refuses comparing precision that is barely even usable on its own
Oh boi, you know you need something real specific if the line is this long.
i traced osage chan into the your ai slop bores me doodle thing
using a macro
my macro
very peak
seems like style issue
just use const temp variables
if python isnt that bad it can optimize a bit
I need to check for very specific conditions for terrain I'm generating.
just use temp const variables
most genetic practice
for cpp it is literally 0 cost
I have no idea what those are.
:gah:
const int i = 123*13*123/123123123123123412414;
const int ii = 2;
const int iii = i+ii;
if(iii){}
simple as it
Each tile is a dictionary though.
even at this quality i see that there is operators and/or
there is way to shorten this line
not mater
Ok I first segregated all the ors into seperate elif's to make scrollign easier.
Actually.
I should go to sleep.
It's 1:23 AM.
what
fric kwrong channel
budget all went to the components 
this is for now my circuit for switching between battery powered and plugged in for my small briefcase pc
dpdt to switch between battery and AC PSU source for the DC PSU
triggered when there's voltage from AC PSU means its plugged in
all budget to cad

then same dpdt also turns on the other 2 spst relays to start the charging. reason i don't directly 12v them from the buck is so that it waits for the DC PSU source to switch first before starting to charge the battery ig
relays random microsecond delays are ass but it is what it is
friend deleted the entire production repo for our project including backups
we're so fucked
code review is in 2 days
is your friend by any chance a clanker
named claude
natural stupidity mhm
diablotek leaked patent
spring failed the sql db is gone
we managed to get backup back but its completely blown to smithereens
postman keeps failing
imma fucking explode
i really wanna jump off a cliff
this is killing me
Can I use whipped cream instead of thermal paste, pwease? 
how the helly does one delete that and all backups without it getting stopped
this is what killed my 5900x

the sugar fused the ihs and the cooler together

what?
he's just built really stupid
we found an earlier build 
only spring is failing now
the db and postman are back
say patato if you're real
Patato

is someone willing to compile dotnets coreclr to arm64-v8a for me 😅🥺🥺🥺 pls #programming i need this

any help is appreciated 
it’s for my universal unity modloader and i CANNOT find prebuilt coreclr binaries for the life of me
and i’m on windows
so uh
cant compile it
cross-compiling isn't too hard, you can just compil it yourself if need be
not possible on windows, and i use a distro on wsl which is incompatible with the build pipeline they have
i’ll see if i can compile it on my laptop which is linux but i doubt it’ll go well
i tried every route
my own friends won’t even help me 
why are you asking for compiled binaries on discord
if you got errors, how the hell am i supposed to do it?
im not gonna spend an hour sending builds, and then getting "there are errors" so i have to recompile it
well humans are not perfect so i might have messed some things up but there’s a good chance someone else could do it right… cus yk… it’s in the readme
last route
dotnet doesn’t provide official binaries for arm64-v8a so i’m kinda at a dead end if i cant compile myself
look at the errors and figure out how to fix them
shouldn't even be considered a route
so just install another WSL distro?
they can coexist
CAN THEY?? am i really this stupid

i never knew wtf
alright well
im just gonna try on good ol’ laptop and if that doesn’t work i’ll try on wsp
wsl
sorry for my stupidity 
WSL is just the engine. but it's not obvious, i guess.
i installed ubuntu 3 days ago
best choice ever
ram usage at 1% i never seen it in my windows hahah
well… i am aware that wsl is the engine i just didn’t know that windows allowed you to run multiple distros at once 
that's why the wsl command has a list command
i don’t really use wsl so i wouldn’t know, i usually use my laptop
that "ram usage 1%" has to be the desktop with nothing open
time to arch 
and 128gb of ram probably
When NixOS?

i feel the need to question why half the server stickers for the biggest nixos discord server is about furries
I have no clue lmao
OS for furry 
Hey guys, can someone help me with a problem (or I'm going crazy) please?
I'm developing a WPF app. I've encountered a problem where when I update a property, it only displays after I reopen the window. I've tried various options, but nothing has helped. If anyone has any suggestions on what to do about this, please message me
wdym with "reopen the window"?
you want the window to update while minimized?
minimized WPF apps have their rendering engine paused
I'm going to record a video now to show what's going wrong, wait a second pls
it's like they forgot how to seal the packets at a right angle
I have a problem with internet, I'll send it later💀
Bruuuu wtffff??
They aint retro...
i think that means you're old now.
Thats like saying the PS4 is as old as a teenager...
Shit aint old
do 13 year olds count as teenagers
the ds i can understand, but ps2 is old as fuck
yes
oh, the original ds is form 2004
i thought it was newer
Fuckkkk PS4's that old?
you thinking of 3ds?
But im barely an adult myself...
2013, damn
Was it because I grew up with both a 360 and a PS4 since i was like, idk... a toddler?
tech advanced a lot in from 1990 - 2010, but i feel like not a lot has changed since then
I dunno about the Wii U but the PS3 and 360 are
Yeah that's the one that feels off
eventho the wiiu is classified as 8th gen, i still feel like its more a 9th gen console
or is it 7 and 8?
i cant rememebr the numbers
oh no im mistaken
its because both the wiiu and the switch are 8th gen
they make these gens so confusing
I've never heard of those generations
Even tho the PS4 is 13 yo, how and why is it still getting new games in the big 26 dawggg?
??? 
Not to mention how weak the switch 1 was, why is it still getting new games a decade later??
cuz its functionally pretty much the same as the ps5, just weaker
if a game you make will run on it, and putting it on the ps4's store is close to free, why wouldnt you?
Most of my steam library had a PS4 port at some point...
the ps4 is a close to jsut being an x86 pc
Small
119B
well, x86-64, but you get the point
ah ok so it is a made up number these days
ye kinda
Later console generations are based on groupings of release dates rather than common hardware as base hardware configurations between consoles have greatly diverged, generally following trends in generation definition [...]
Well, i guess the bright side is... if a game has a PS4 version, then my laptop will guarantee run it at 100+ fps
As oppposed to PS5 only games barely running at 70 on 1080p medium


EL DIABOLO 
im not consistent on this, i only use the correct term when the difference matters.
Are devs getting that lazy? Cuz games barely looked any different from each other and somehow ran worse than others...
i guess i use the wrong terms according to the poll
amd64 if i dont want to press shift
x86_64 if i want to conform to standards
x86-64 if i want to do maths
x64 if i want to play minecraft
first option if I miss my shift key when typing in a target triple
amd64 just feels wrong to me
amd64 also technically correct I think


yes but amd
ye but Intel pretty much copied that AFAIK
idk I forgot half the history
something something Itanium

what do you call arm
the million arm versions 
yo @olive sable can you accept the request?
sure
x := u8.[1, 2, 3, 4]
v: ^@simd(u8) = @bitcast(&x)
for e := v {
e.* += @splat(1)
}
example
tetrahedron
i can now write a simd swisstables hashmap
but im lazy

look look simd in codegen
only sse at the moment
it uses the largest vectors available on the target, but its dumb
not really auto vectorisation

simd also implemented with non-simd fallback
simd instructions implemented for wasm and x86_64
wrr
oh recently there was an inlining heuristic added
we dont call HBM ram, "samsung-HBM"
idk, amd64 jsut feels like a dumb name to me
im not entirely clear on the specifics, but it appears to ignore alignment
it works in the tests so its probably fine
complain to AMD
Samsung probably could've named it SHBM if they wanted to
-# though apparently it was a collaboration with AMD, so it would've been ASHBM or something idk
that seems
not good


i agree
probably fine
anyway the feature isnt fully implemented yet
so
maybe it wont be that way 

Okay I think some of the DLSS 5 backlash is the usual overblown response to the word "AI" being used but this got a chuckle out of me
Pfft
I saw someone say Nvidia got high off their own supply with this one and I think I agree there
Its clear the technology has potential but they decided to demo it using implementations that are as maximal as possible without any tuning or deliberacy
For their part Nvidia claim the yassisfied Resident Evil screenshots are the result of capcom https://x.com/MrBenB/status/2033626943929913587
@tomwarren This example worked on by the team at Capcom - game developers have detailed artistic control
guys it's not ai 
gamestopslop

also shouldn't have named it DLSS 5
horrendous branding decision
the 5 in dlss 5 stands for slop
Why does NVDIA want to yassify any video game that the question
google it
to appeal to gamers
Who want non determistic rendering that undermine the artistic vision ?
gamers
I don't think you understand what this actually is
By definition yeah, AI upscaling and rendering tools need to invent details
it is deterministic, just not accurate
When you reduce an image from 4k to 480p that a lossy process that info is gone, upscaling can't retrieve it it has to guess it
deteministic mean "same input -> same output"
It is accurate insofar as developers actually do have control over it and its grounded in the lighting and geometry information from the game engine
The demo implementations just sucked
I don't think the same input / same output is true here but I haven't read on the tech in a while
computers are inherently deterministic im pretty sure
In Nvidias own press release they call it deterministic
that being said, i still dont like dlss
The yassified grace kills me 
I kinda love DLSS but I'll reserve judgement on DLSS 5
we're just trading variable render time for static ai upscale time
Honestly I'd rather they improved ray reconstruction to not look like arse below the quality preset before they worry about this stuff
oh jesus
they gave her face more wrinkles
I mean yeah even if determistic it will invent details
Interesting that they got it to be deterministic tho
its kinda hard to not make it deterministic
unless you feed it noise
Apparently that implementation was actually designed by Capcom
And it does look closer to the real face model for grace
But subjectively it looks silly
its more acurate to irl humans, but far from acurate to the original
Like you can see the potential in the technology more clearly in this comparison
NVIDIA DLSS 5: Resident Evil™ Requiem GeForce RTX Comparison: NVIDIA DLSS 5 On vs. NVIDIA DLSS 5 #003
Where Capcom did a less stupid job of it
Many of the current image gen stuff isn't deterministic, thought they slapped something like that on but that must be more complex
that because image gen uses noise as the start of the diffusion im pretty sure
Image gen models actually are deterministic
also mabe noteworthy that NVIDIA probably doesn't mean determinism in the strict sense
more like "similar inputs produce very similar outputs", which is much more important for this kind of upscaling
If you feed it the same noise, prompt and seed you get the same image
can't have the whole frame change significantly just because some random pixel is slightly more red than before
honestly aside from this being a clearly gamer(derogatory)-centric feature I couldn't stand the example it showed where it was misinterpreting the background as the foreground
as a gamedev, i should be careful about being derogatory to gamers
if people ask me to implement dlss i will ban them tho
I gotta clarify otherwise the kids here start crying
wrr
gamers and gamers
The general gaming audience are morons, but I think the intensity of the backlash to this tech is misguided too
it's essential that my characters must have pimples or else I can't play my highly stylized game
does the same apply to XeSS/FSR?
sure idk
please implement waifu2x 
there's the difference between gamers; people that play games, and gamers;
PC masterrace RTX 4K 240hz
that's what the volumetric display is for


Looking at DLSS 5, I guess in the future graphics pipelines are just going to render a low-quality frame as reference and feed it to an AI model to generate the finished frame?
nvidia has said many times their goal is to replace the exponential compute increase that comes from increasing fidelity with constant time renders
every pore on the face must be simulated such that you can zoom in and see the mites that live in them
just add more cores to the gpus
you will get more tensor cores don't worry 
As long as I get more VRAM too 
gamers when they do not get their chicken nuggies (vram)
vram sure, but i feel like for gaming 32GB is more than enough already
Not for gaming, I want more VRAM for ML
but what if the game has AIs?!
and tensor cores are absolutely useless for me, unless we start using ai for npcs
My GPUs are dual use
tbf nothing nvidia does really matters anyway since they're not used in consoles
good
now when amd makes a worse version next year we'll see
They partnered with Palantir yes
But honestly I don't know what to say if you thought publicly traded megacorp #2 were moral bastions
they're just as bad
i think the ultimate in game development will be when you throw four pixels on the screen and pass it to dlss
the people that make the gpus are shit companies, the people that utilize the gpus are shit companies
who would have guessed?
the companies do really well 
you could probably make a very cool music visualizer though with it. just set a grid of pixels up and morph them to the music... then let dlss turn it into something trippy.
genuinely, dlss will just make it slower
but it'll be prettier...
sorry I mean X, the everything app
Nintendo uses Nvidia based chips
ther eis no way to make dlss go faster than 0.000007 seconds per frame
don't care, game is too woke without realistic characters
i dont care about being woke
You actually can kinda do that already https://youtu.be/ge8K6m6JX9I
Using Ultra Performance upscaling using DLSS 4.5 Preset L, I attempt to run Red Dead Redemption 2 at increasingly decreasing framerates until my framerate increases slow up and go down.
0:00 - 1280x720
0:26 - 427x240
0:55 - 213x120
1:22 - 160x90
1:51 - 107x60...?
2:04 - 140x47
2:32 - 137x77
2:48 - + Lossless Scaling
3:55 - + FG
4:24 - The futur...
Not four pixels but pretty close to the jokes intent if you remove the hyperbole 
guess triangles gonna have to go on the woke game detector steam curator profile
thats crazy
I think the end game will be just rendering reference frames consisting of basic shapes and textures for the AI to use as basis to generate the full frames.

🔺 
should've known when it started using nix
but the triangle is also packaged for nix
i made the triangle
is it? i didnt make that package
konii did iirc
my terrible attempt at humour is failing to land
gonna go get ragebaited by my uni flat now
why is the uni flat? 











