#programming
1 messages Ā· Page 149 of 1
bwa
That's one thing I kind of like about mojo, if you explicitly tell the compiler to do something it will do it, regardless of if it's a good or bad idea
you also like this about hblang 
why is it so hard to do anything when you feel sick
bwabwa
What about this CTF? https://github.com/jellyctf/challenges
Damn body trying to survive, how dare you 
Fun fact, the body does not try to survive
because you should rest and be lazy 
lie in bed and read something
noe
But the body does try to remove foreign stuff at all cost
That at all cost part is problematic 

huh
I've been wondering recently if Sam was going to be less trusting as an adult, he seems to be more vigilant about people keeping their distance from him, which is good; but it's nice to see he still makes exceptions for people he knows.
I miss my swan

They are L O U D tho
is this not a valid argument why you should bedrot

Hugging face absorbing papers with code but doesn't even have non-machine learning paper smh
At what point does a game stop being a game and go into simulator territory?
i got that with a cloudflare captcha once
turned on 1.1.1.1 and it fixed it
wasnt on linkedin though
I fixed da maths.

I have no clue why the tutorial I followed gave the wrong equations.
I would assume I don't need to make a filled ellipse option as you can just use the paint bucket.
wait i just realized
Not for production use
linkedin please 
Audio???
audio is normal for captchas
it allows blind people to pass them
sideways toad, sideways toad, sideways toad, sideways toad, sideways toad, sideways toad, sideways toad, sideways toad,
that's what I imagine the audio says
if i use my bt headphones on windows i have to remove them from linu and repair them to use them
a bit annoying but not the end of the world
Are we allowed to edit the code for the challenge? Or does it have to be as is
edit it as you please, as long as you dont cheat by forging the fps
you're also not allowed to just not render anything
#programming message also I got 28k FPS with a 5950x, and a 6800xt on arch
Oh, yeah. This happens. Wireless device can be finicky with dual boot
this is cool and all, but the purpose is to note down the gpu that was used to render it
not all 3 in your system
aight i cleaned upt he code for the linux build, it shoul be a lot more orderly now
okay since i'm already on windows to grab some backups, ran it the triangle again and got around 66k (w11)
im trying to combine the 2 repso for windows and linux into 1
jsut need to change the makefile too now
old nix apparently used let { a = 5; b = a; body = a + b; } instead of let a = 5; b = a; in a + b
Thanks I hate it
so it was literally just syntax sugar for rec { a = 5; b = a; body = a + b; }.body
Though I get it
aight the main branch should jsut check for windows at compile now instead of needing to use a seperate branch
havent tested it thos since i did this on linux
aight it works, i can delete the other branch
shiro is a contributor now apparently
shr
stop reading my mind
the windows build is still doing 70K fps, while linux is still only doing 18K fps. the only diffrence betweenthe 2 is this tho

ive updated the repo about 10 times already since he become one. im probabl clutering his inbox
discor

mobile 
using box i have identified they are pixel aligned

i think the top one is a tiny bit more to the left
it not

guys, i may not have a brain, but i have an idea....
instead of THREAD_PRIORITY_HIGHEST, what if i used THREAD_PRIORITY_TIME_CRITICAL?
it matches perfectly
don't
it causes bsod
I tried


I freezes my whole windows
lemme try
(LOL @olive sable ]
i get 72053 fps normally, and if i dotn crash ill say the other number
Let me test @midnight sigil modify script
In a100 gpus š
ok while i didnt bsod, all my windows did become unresposive

it did ask "... is unresponsive. close?" so i did that
trying to launch it again im stuck at 10k fps now
windows might have secretly nerfed it
I don't think this triangle utilizes cuda cores

even if it did it wouldnt matter
yea it's cpu capped
also doesnt all 3d acceleration (on nvidia) run on cuda behind the scenes?
you jsut dont program it in cuda
at least my windows tells me your program is only using 3% cuda
yep, cuz it aint doing shit
and 100% compute_1, whatever that is
tldr: no
ok but like, then what else?
the name cuda core is kinda shit tbh
they're not bound to cuda, cuda is a level above what they would write to implement the graphic apis
maybe tensor cores? wuh
no
while vulkan does have support for ofloading shit onto the tensor cores, you generally dont do it much
Someone says the tensor core ?
unified shader cores

ye but those unified shader cores are just the 10496 cuda cores on my 3090 no?
yeah
For example, the unified shader is referred as "CUDA core" or "shader core" on NVIDIA GPUs
they all come down to cuda cores
at the end you're calling a function on a bunch of inputs, whether you do this in a general purpose manner via cuda or via specialised graphics apis doesn't really matter to the ones that end up executing it


Bye my #### academic (i hate academic)
I will ask my Dad to my academic
you do need to do some very specific stuff do do matmul with tensor cores instead of just localy
normally you jsut do mat1 * mat2
for tensor cores you need to do this whole ass thing
#version 450 core
#extension GL_KHR_memory_scope_semantics : enable
#extension GL_NV_cooperative_matrix : enable
#extension GL_EXT_shader_explicit_arithmetic_types_float16 : enable
#pragma use_variable_pointers
layout (local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
layout(set = 0, binding = 0) coherent buffer Block {
float y[1024*1024];
float x[];
} block;
void main() {
fcoopmatNV<32, gl_ScopeSubgroup, 16, 8> m = fcoopmatNV<32, gl_ScopeSubgroup, 16, 8>(0.0);
m = m + m;
m = m - m;
m = -m;
m = 2.0*m;
m = m*2.0;
coopMatLoadNV(m, block.x, 16, 128, false);
coopMatStoreNV(m, block.x, 16, 128, false);
}
i doubt it would be faster anyways for just a singular multiplication
since you need to transfer memory to it and such too

you'll need to move the memory around either way, unless you hate yourself i guess
its just glsl, so normally you dont have to do much memory stuff yourslef
bucket
the next step of the vulkan triangle optimization is optimizing the shader binary 
OpCapability Shader
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main" %outColor
OpExecutionMode %main OriginUpperLeft
OpSource ESSL 310
OpSourceExtension "GL_GOOGLE_cpp_style_line_directive"
OpSourceExtension "GL_GOOGLE_include_directive"
OpName %main "main"
OpName %outColor "outColor"
OpDecorate %outColor Location 0
%void = OpTypeVoid
%3 = OpTypeFunction %void
%float = OpTypeFloat 32
%v4float = OpTypeVector %float 4
%_ptr_Output_v4float = OpTypePointer Output %v4float
%outColor = OpVariable %_ptr_Output_v4float Output
%float_1 = OpConstant %float 1
%float_0 = OpConstant %float 0
%12 = OpConstantComposite %v4float %float_1 %float_0 %float_0 %float_1
%main = OpFunction %void None %3
%5 = OpLabel
OpStore %outColor %12
OpReturn
OpFunctionEnd
Nai
Have fun
since getting linux my windows clock keeps getting unsynced
consisntently off by exactly an hour
sometimes 2
Linux distro name ?
cachyos
Try auto sync
cachyos isnt the problem here
its windows that keeps getting unsynced
also i broke my windows somehow, its locked to 10K fps untill i resize the window slightly
Restart ure pc easy
i did
https://support.microsoft.com/en-us/topic/august-12-2025-kb5063709-os-builds-19044-6216-and-19045-6216-96d99cf6-f8b5-4798-9892-4e3eb8f11548
This win10 update will drop your storage device upon large I/O, and correlates to this win11 update:
https://support.microsoft.com/en-us/topic/august-12-2025-kb5063878-os-build-26100-4946-e4b87262-75c8-4fef-9df7-4a18099ee294
classic microsoft
what
On my PC, I am currently using
6
11
1
Linux (NixOS)
1163912230585237655
enub
they're breaking win10 cuz they want to be a bitch on win11?
it was a security fix for win10 
so nix actually got 45% votes
i did?
which is still a lot for a single distro
you arent a nixos user to my knowledge 
yesh its not a trivia question 
and the new Edge Game Assist thing hook onto your game, allocate 2000mb more ram and lag the game on an interval if you don't intentionally kill it in your process manager 
why lag the game on an interval?
What ure uptime in WINDOWS
a couple hours ussualy
Windows 10 Enterprise LTSC 2021 is affected by the update btw
YOOOOO
SATA'S ACTUALLY SPEAKING ENGLISH NOW LETS FUCKING GOOOO
-# sure its still nonsense BUT ITS ALMOST COHESIVE
"I hate that" is so real tho

My phone was too far away and I saw "first imma cat"

please do not the cat
uhhh
manually installing things and dependencies (aka no package manager)
Windows, good answer
windows do have a package manager
it's winget

macos
theres homebrew but
its not really official
and macports ithink
or whatever itscalled

not until 14900KS owners show up 
maybe i should also try the macos
but only apple product i have is an ipad 
maybe i coult rent a mac?
M4 max, $47 AUD per week
monkas
dont rent shit for this stupid project lmao
also according to vani mac struggles for performance
probably cuz vulkan to metal conversion or soemthing idk
i spoofed my mac address and ip to imitate my doorbell
Oh Sam, the things we do for shit and giggle 
that's quack smh
you're speaking names wrongly

you mean to tell me the guy named KuShiro, with the word Shiro next to his name, is not shiro?
quack's username is kushiro


:SMH:
That's obviously
, just look at the bird
Anyway how the triangling going? Is Linux faster than Windows yet?
nope can't figure it out
-# Chat, I may start NN work again soon. I have some things I need to do before then but it's coming.
its odd, vulkan is supposedly faster in linux
i might wanna make a one click compiler for linux ngl
@nocturne olive can you try the triangle on windows
For the half of the linux users here, just include a nix shell 
- I'm not local
- I've fully switched over
- my Windows install is dead
-# Unless I use my remote laptop
also i suspected that nuking 80% of the vulkan lib actually makes most of the lib files recompiles and optimized from source hence an even faster performance
lmao fr
technically if I SFTP over the file I can run it on my remote laptop, which has a UHD 620
yeah you can try
I'm gonna need the file though
.
i still find it hard to believe that my es cpu performs really good at triangle lol
So as it turns out my remote laptop lacks the headers for the C++ STD, who could have ever expected that
rip
no
linux multithreadign sucks apparently
that and there might be a problem with the drivers
something to do with linux using mesa and windows having direct hardware
Or maybe you just need to handle it differently
idk how
Do a funny and write custom kernel
How is MSVC this massive? 2GB for a compiler? The remote machine only has 128GB total for Windows
And still no compile
This "portable" compiler sucks, it needs 40 billion things to already be installed and still complains after that
Would i be crazy for making a web-ui for ableton?
I don't know what Ableton is
Music making program
How silly
I hate MSVC now by the way
How has it downloaded 15GB of garbage and still not basic stuff like float.h that are required for the most basic hello world program??
I don't have the storage to be dealing with this on my remote machine
And then like optimize audio transfer between client and host using the QUIC Protocol
@true hemlock I don't think this thing is running the triangle, I don't have the storage for all the garbage MSVC wants to install just to try to mabe get the basic float and string headers required
11.00MiB
Yo
This thing has installed 15GB of garbage and still no float.h, string.h or any of the other ultra important things
This thing couldn't even run the basic hello world since iostream depends on float.h
Wait i have gpp INSTALL in windows pc without wsl unit
What was installing from the visual studio
Not even Visual studio, this is all the compiler
how do you have 15GB? i dont think i even use half as muchs torage for that
MSVC just sucks that bad
True
And again, it still doesn't even have the most basic headers
So I can't even compile triangle
Give me details to install all models ? If i can assist that
Whar?
huh? This isn't AI
Shoutout to discord for splitting a permission so one can finally pin messages without also having access to deleting other people's messages.
I got free data yea
What would I need that for?
whats the fastest and most effective way to learn python
fuck around
and find out
honestly basic syntax is like the easiest
so if i fuck around and basic syntax im good to go? 
It's one of the best starts you can get
just install gcc
That also didn't come with the headers
Or I installed the wrong GCC
learn basoc shntax than play around and teach urslef concepts
It takes time lol
its weird though, clang++ should've already included all the headers by itself
Maybe one should learn english syntax 
i can speak english im just mot used to mobile kb
Not the executable alone though
True, mobile sucks to type on. Sounded a bit drunk there 
Iām used to a physical laptop kb
I donāt have a phone (smh)
And ipads have wide screens
So add them together and its hard to accurately type without sacrificing speed
Still better than apple watch se tho
I still rely heavily on autocorrect to be anywhere near fast
Canāt type on that at all
I have autocorrect disabled
Itās usually an annoyance
let me try something
I've tried on a watch as well, did not end well at all
pov me taking 10 minutes to write the funniest joke to the group chat only to accidentally send it halfway through the punchline and ruin the entire operation
That's me forgeting shift + enter on discord desktop
Who'd ever forget you need to do that?
And this is why I hate Windows, making things work is just impossible
try just running the prebuilt
while i try fixing it
I assume Windows 11 sucks so bad
Unless it's 10 then I have no idea
windows 11 is fine these days
its fine
Didn't they just recently push an update that wipes disks?
Either way, I hate Windows, I hate MSVC, someone go tell Microsoft how to make a proper tech stack
MSVC doesn't even come with the headers to compile hello world and obviously can't run triangle
aight, imma profile the trisangle and then go to bed
i should probably make interactive .bat ngl
This laptop only has 128GB for Windows and most of it is already used up
for the choice to also delete the whole thing that's only used for compiling after it compiled the triangle
My VR system in a nutshell
Windows gets one 128GB sata drive and that's it
Whar?
i think the windows 70K fps stuff may be fraudulent. i need to research first tho
what renderdoc is showing isnt too promesing at least
either we arent doing drawcalls or renderdoc cant see them
i need a second pair of eyes.. preferably someone who knows rust...
I've been trying to debug a issue that i've been having with a c++ to rust port for the past 2 days.. and i just can't figure it out.
https://gist.github.com/Adesii/3c574eabb151282832a90083669a558b
its a binary greedy mesher for a voxel game im making. but its not meshing the voxels correctly.... it works in the c++ version and i literally can't see a difference in the functionality between the two. yet they have drastically different outcomes..
im unsure if the hashmaps are just not being populated correctly or if the loops in the greedy mesher are wrong... i got no clue
Maybe Windows doesn't accept drawing identical content as an update
thats probably not it since that should be a decision on the vulkan side of things
but idk
Well, Windows is Windows and it does odd stuff
couldn't you add a bit of movement in the shader to check that?
but it does sound odd.. i don't think vulkan does that kind of stuff. if you render something its gonna render it
i like how making it do 7 frames in lfight makes 7 of a bunch of diffrent objects
and quack was apprenlty making 4096 frames in flight
madman
Draw calls: 1
Dispatch calls: 0
API calls: 6
API:Draw/Dispatch call ratio: 6
the frame statistics show one has happened so false alarm
but i guess renderdoc couldnt fit it on the timeline cuz of the frame only taking a couple microseconds to render???
Maybe the one is rendering the trinalge and then it just skips rendering the rest of the times?
Or I don't understand what your thing means
I don't know
Figure out what it's messing up
it does seem to make a new trianlge each frame
But would Windows do something weird, see that the content is identical and tell the GPU to not refresh?
well to be honest, i dont think thats possible withoutmaking it slower
cuz you only know the output once it has rendered ont he gpu in theory, and to check if its the same you'd have alreadty rendered it
I really have no clue how Windows could actually be that much faster
the only possebility is the diffrence in how it handles threading, or the vulkan api not going directly to hardware commands
windows and linux are pretty tied in most benchmarks that I have seen, linux maybe better in some productivity benchmarks.
there shouldnt be a huuuuge difference
well there is
Well the difference is like 60K FPS
ye, I have seen
18K fps vs 72K fps ye
but like I dont see how the gap could be so large either
I'm pretty confident Windows is messing something up, like could it somehow mess with the timings of when the frame count is checked?
How do you determine when to calculate FPS?
the renderthread just does +1 to a value, and the control thread changes that to fps
renderhtread is uncapped, control thread does 20fps
like this
Do you account for race conditions anywhere? If the render thread is constantly updating frame count, it could end up having just copied the frame count value as the FPS update thread updates it, adds 1 to the old value and pushes that to the frame count, counting two intervals of frames
i dont account for race conditions at all
Then we can't be sure if frame count is being race conditioned
The frame count could be showing corrupted values
the only thing i have is
std::atomic<bool> running = true;
std::atomic<uint64_t> frameCount = 0;
but i dont think im using them correctly
atomic suppsoedly helps with race conditions between threads
Hm
It should maybe help, not sure
Then it seems possible you have to look at the FPS update function in case that's being scheduled inconsistently by the Windows scheduler, causing it to miss the intervals
It's always possible it's making a syscall, Windows takes the thread and goes off doing other stuff, and only returns it a while later
this should be plenty enough for simple counting system
even if it was scheduled inconsistently, the SDL_GetTicks() should still make the fps acurate
adding a mutex would destroy the performance
true
well im done with all the info i na get out of renderdo
im trying nsight now
nsight takes a lot longer tho so im waiting on the results
40 seconds of "gameplay" is taking 5 min to compile the results of
this shit is using 6GB of ram for statistics and api calls
i lied its at 11GB now 
oh damn
"showing 466261 out of 466261 cpu frames" ah yes, i can see why now
oh i discoverd why it would start up in 10K fps and switched to 73K when resising.
i accidentally made them use the same core but they automatically switch when resizing the window apparenlty
ah yes, the windows NT kernel is givign us 3 microseconds of interupt

I assume that's time-based interrupts
im gonna flip... my binary mesher was working all this time. but my voxel array wasn't....
I saw you set the thread priority to real time at some point
Tip about setting thread priority to real time: don't, it can literally lock up the entire system in a way where you will need to hard power-off the system to recover
i have tried it but it didnt work out
i put it on realtime... it did lag insanely but i was able to close the process. (tho the fps didn't update)
first tim using it after reboot makes the system choppy, and after that it jsut behaves the same way as highest prioroty
so im geussing it just decides to not do real time after locking up
It probably just doesn't give threads back to the kernel so the kernel is not able to update the window title bar
put the triangle in windows bootloader so it doesn't get affected by windows fr
you can write it as a native application if you uh
Like dealing with only ntdll calls
which means software rendering only


i need to be active here
Evil my oshi
how are we even supposed tog ive you those?
we're just programmers
it says that the profiler overhead is that tiny grey bar of 13 microseconds, but this was running at 3000fps instead of 72000 fps so its a bit more i think
Lmao
@olive sable i figured something that might be a bit helpful

you mean to fix it being stuck at 10K fps beofre resizing problem?
cores
after resize
i fixed that already, i made a stupid mistake in copying code for the multithreading
Genuine Intel 0000
just might be helpful to figure out why windows gives way more fps for some reason
prtsc
is the program still running here? cuz your gpu went down form 100% to 24%
thats a bit sus
right, there's even difference in GPU load for some reason
before resize
yeah
hmmm
i need to check GPUZ aswell
ye i was putting both threads onto the same core by accident, resizing redistributes that for some reason
i think i might have an idea of what's going on
need to check first though
im not sure anymore
hold on
before resize
bro has a dirty taskbar
gpu clock goes down, memory clock plummets.

no i have them in the wrong order lmao
gpu clock goes up
and memory clock....
memory clock looks like cpu capping issue
right, task manager measures utilization differently
my GPU draws more power after resize too
seems reasonable
GPU load seem to go up
memory being that slow before resize tells me that it attempt to fill the SM on the GPU out really slowly, which then it just splits saturation between clocks
resizing seems to "refresh" something
i thought it had something to do with the multithreadign but putting everything into a single thread doesnt fix it
it is 100% vulkan and shader thing, failing to utilize the GPU before that "refresh"
On Linux the program stops rendering and responding to inputs instantly after resizing
okay i got an idea
roughly how much shader memory does this thing take?
because i had a suspicion that the low clock is probably related to the GPU's L2 cache, although that doesn't help our case lol
vkAcquireNextImageKHR() or vkQueuePresentKHR() are both able to see that the window resized, and are able to report it as VK_ERROR_OUT_OF_DATE_KHR or VK_SUBOPTIMAL_KHR
in repsonse you ussualy rebuild the swapchain but here i dont.
the question becomes how vulkan behaves if you keep running it like this
that reminds me of one part of your code that handles something related to resizing...
ye i deleted all the resizing stuff this morning
oh? have you pushed that to github?
but its still there in old commits, just commented out mostly
only the very first commit handles resizing properly
There's definitely nothing handling resizing, resizing on Linux just breaks all rendering
It stops rendering and I assume on Linux halts on the render call so the FPS goes to 0
hm
i have an idea tho, im gonan take the VkResult result = function() from the previopusly mentioned functions to see if they actually keep rendering if the window resizied
If Windows instead skips the render call when rendering errors out, that would explain the significantly increased FPS on Windows after resizing
Weird, must depend on what it is. My laptop had no issues scaling any of the binaries
it should still do, evident by my GPU actually having higher utilization after resizing
Also compiled from source
ye windows doesnt either, but the problem is if vulakn peaces out after it resized
that is true
dunno why it doesn't have such utilization before resizing though
making it full screen also does the same thing (with this one visual bug in the middle for some reason)
If Windows responds to that by simlply skipping the Vulkan parts that would explain why it has issues
ye
Can confirm that one on KDE minus full black, last frame was a ghost and moved to top left
Hardcoded cpu cores mhm
How Windows handles unhandled resizes on Vulkan contexts is the key here
Does it forever stop drawing?
im not sure
That's my guess personally, and whatever jank happens with the window happens
Hmm
Temp RGB triangle for debugging 
That supports my guess
it's just sam didn't clear buffers idk
A war crime against those with epilepsy
Yep, it's probably that
So we have the answer to what if you don't render the triangle
ok so we cracked the case wide open and the 124000fps was fraudulent.
so the higher gp clock comes from the gpu not rendering anytihng and judt presenting the already rendered framebuffer
coming back to this and the error, I found your code accidentally created a double double bubble, which is not supposed to happen 
that would also explain there being 2 triangles here
I just stared at the debugger for 30 mins lol
ok so the only options i have is re-add framebuffer resizing handling at a small cost. or find a way to disable resizing completly
On windows only 
time to nuke leaderboard 
on linux it jsut crashes apparenlty when resizing
try RGB triangle btw
windows > linux
My god, you're still debugging the compiler?
u mena rgb as in it changes colours or its just diffrent colours?
change
perhaps
uhhhh
every single frame
changing colours would mean addig uniform buffers
Folks at #programming truly builts different
Simply tell Windows that the window does not allow resizing, that is an existing property a window can have
full screen button also did the same thing lmao
Make it a compile flag mhm (or just override the const "#FF0000")
Simply don't spawn the maximize button either
what are you guys doing 
ok so ill see what i can do about rgb triangle, but dont expect much
šŗ
You can render a title bar without a maximize button

The title bar is something the app explicitly requests on the create window call
The resize property is the same way
Contribute to samvanmaele/very-fast-but-very-bad-code-vulkan development by creating an account on GitHub.
lmao
@ivory plinth should have pinged
It's a bit silly but completely possible to just omit asking the Windows API for those entirely
(probably a bug)
suddenly i feel like i want to run it
either i have to spend a day or 2 on adding uniform buffers, or we choose between one of these variables
in int gl_VertexIndex; // only present when targeting Vulkan
in int gl_InstanceIndex; // only present when targeting Vulkan
in int gl_DrawID;
in int gl_BaseVertex;
in int gl_BaseInstance;
and base the colour of that
When absolutely no Vulkan at all happens:
apparantly, I might have implemented this wrongly and made it to create a double double bubble when the input is a:[10, 4] b:[3, 2](be aware of my bad code)
i forgor
idk how 
nah none of these change per frame
just for good measure make it render red for like 5 second then start changing color
oh
the magical triangle
Hmmm, that does indeed break the spec lol
It's so silly how #programming has just become "make the triangle go fast" simulator
so rgb trianlg eno tpossible for now. ill try the sdl lock window stuff now tho
insert quack's not a virus zip file
Where are you going
ā³-----------------
Based on the area around the triangle not being updated, it's definitely not rendering
tower
If it was still rendering the triangle would still be normal in the area the default window takes up
it's rendering, it's just not clearing previous buffers
do you perhaps play any source engine games?
it's like the null buffers
ok so while the bug being fixed is cool and all, i only get 8K fps with multiple threads and 10K fps on a single thread now
bwaa

but ye now linux is faster since it wasnt cheating
No, the vulkan presenting frames is just being halted
send me that
i gitpushed it
ye
As dev very much intended
And so Linux was once again superior and everyone was satisfied
exactly
nah
Nah, Linux'd win
lets fucking goooooo
so how does this leaderboard thing work
Just run the thing, then show the score you get and state your hardware
i have to edit it manually sadly
Perfect first time project
i3-8130U
UHD 620 
also cachy
Learn how to make discord bot that updates scoreboard with your submission
peaked a bit higher tho
13K is respectable, certianly for that hardware
you're still getting 50K even with the no resizing?
yeah
on windows?
windows
what

also i saw a ram column in the board 
Even without the insane advantage of not having to render the triangle
still has insane hardware
honestly i should have resizied the window while profiling to see what it would do
but im geussing it jsut completely ignored all vulkan comamnds tot he framebuffer in favour for jsut full throthle on refreshing the same buffer
fixe š
this bug also exists in other arithmetic operations
Your parcel will be delivered today between 14:43 and 15:43 

lower while minimized 
could jsut be getting less priority
almost 13k when apps open
no errors, perfect 
@ivory plinth what's your cpu
i3-8130U
ye thats seems correcter
correcter
im going to nuke windows leaderboard
is that not a enlgish word? superlative of correct?
oh
wait
is that why my laptop is getting lower than 5K
because i didn't resize


it's more correct

@olive sable max score of yours?
well, i cant get above 11K rn 
i used to do 13K on windows if i made taskmanager kill everything on my system
I wonder how that is, cause we have pretty similar pc's so I don't see what would stop you from getting higher
what ram do you have? I'm using 4x16 @3600 cl16
I'm on 16 19 19 36
i dont really wanne opent he biuos rn so idk
cachyos is bloated, confirmed! 
I have 2x32GB 3200 CL16, running at 3800 CL19
But I could get it down a bit to CL18 if I bothered to do that
Probably lower too
noted
hmmm, i used to get 11K fps acording to my chat history here
on windows
but now im getting nearly 11K without killing everythin in task manager
Genuine Intel 0000 my beloved
13K is probably psossible then
ok so now that we fixed the resize bug, is igpu prefered again?
Though depending on what Windows is doing it might not work if Windows decides you only get the dGPU today
nope, way lower on iGPU
if you're running it with makefile you need to do ARGS="--use-igpu"
And on my system it definitely won't work either based on it not allowing to run on iGPU before
"way lower"
i dont have an igpu so wouldnt matter anyways
well this is still decent
Can i post articles here
30k on the most recent version 
if its programming related or something you can loosely link to it sure

not the psp
It is programming and llm related
im still surprised how well this thing runs games
scratch this, switching to gcc gave me 31k
aight
gcc strats
The fact that it can even run an LLM makes it a powerhouse that got underutilized
You could probably do more with it
i dont have gcc on my linux
updated

Interesting
Whose Linux score is the highest right now by the way?
I forgor
Oh yeah
The leaderboard
its up

lmao
Silly commit name
i murdered one of microsoft's background process
@olive sablescore btw?
also, what did you even do to get that score of 50K before? cuz increasing the frames in flight makes it slower for me
it was actually faster with flames in flight 8
and for how i got 50K above, i don't fucking know i was even surprised when i do make run and it hit that kind of number
all the foreground processes will be killed
ye im already on pbo 3, im not gonna stress my cpu even more
time for nixos
Pkilling discord makes my ram usage spike to 32gb for cashe???
prolly a dumb question but what model was neuro trained on? ._____.
What is it caching???
We dont know i think
we dont know what model
uses, probably some heavily customized open source model
not public
I got 32k when setting to 8 frames in flight, you might be onto something
aight imma update your score
Hmm, setting the core_id on both threads to the same number seems to be better for merformance
But it was already set to the same before by accident cuz i copy pasted the code without thinking
ok so unless anyone else has ideas on how to make thei run faster, the currenct code is peak performace while stilla ctually rendering?
should be
god damn using only 2 sticks look ugly
might wanna get dummy stick just because my ocd is acting up

People use fakes with rgb these days to fix that
I just use 4 sticks
Damn im really tired now
I think ill start actually implementing features again instead if optemizing. But not today
According to linux im also running 1.2V
low-profile cpu cooler 
SPELL: Spelling error - "const" is not a word. ahh yes, thank you, Visual Studio 2022

that's pretty funny
Typical XMP is 1.35 I think. though seems to depend on the kit
wait you didn't know?
its a surprisingly good cooler

Guys, prepare for something absolutely dumb we just discovered. So, turns out on Rust 1.87.0 and later (including nightly), a format string that is too long stops rustfmt from working.
So, here's our fucking solution 
certainly a bug of all time
What the actual fuck lol
Does rust really hates improper solution to anything? 
I wasn't able to find the right issue yet. Otherwise I'd attach the screenshot
most likely it's simply a minor bug, as this hasn't been an issue before
and, we have other files with the exact same code (except for this splitting), and those format correctly
but yeah, the original error formatting string was over 70 chars, and the moment we only gave it 70, suddenly everything worked again
time to manually format your code as god intended
This seems totally unenforceable
hahaha the fun part is, it doesn't just stop formatting this piece of code. it stops the whole file from being formatted. and rustfmt -v ... doesn't even show an error
our government agency for like cyber stuff (BSI) even recommends ad blockers. I don't see how this goes through
The case stems from online media company Axel Springerās lawsuit against Eyeo - the maker of the popular Adblock Plus browser extension.
Axel Springer says that ad blockers threaten its revenue generation model and frames website execution inside web browsers as a copyright violation.
lmao
companies training AI with copyrighted books and artworks completely diregarding the rights of the creator is fine, but blockin ad servers from sending garbage to your pc is now a copyright violation
and frames website execution inside web browsers as a copyright violation.
website execution web browser. Hmmm, I wonder why is it called web browser
This is grounded in the assertion that a websiteās HTML/CSS is a protected computer program that an ad blocker intervenes in the in-memory execution structures (DOM, CSSOM, rendering tree), this constituting unlawful reproduction and modification.
are they this dumb
this would make my darkreader extension a copyright violation too
Yes it will
They are technically correct in the sense that HTML and CSS is a copyrightable material
oh boy do I love the enshittification of the internet....
But the execution and modification of its in-memory representation remains to be solved
wouldnt it be comparable to ordering a painting and then doodling on the painting after you have bought it?
yes, but I can buy a book in a store, it is copyrightable material, but then I can still draw around or rip out pages in the book as I like...
see, same thing. hah
This is where digital stuff analogy breaks down tbh
probably
I'm trying to ponder about it
I just hope they find actual experts
Can it be modeled as a compiler I wonder
Because even if the raw code representation IS a copyrighted material, the display / presentation can be anything
Because it is equally valid to render the HTML / CSS code as text
Hmm, modding games also falls under the same category
Yes & no at same time
I think I have pondered it enough. They cannot enforce how other program should interpret the HTML + CSS data
Unless they build their own HTML + CSS interpreter
Because otherwise, opening the HTML + CSS in text editor would also infringe their copyright since their representation differ
soon we will enter the epoch of "render it on the server", but this time not with its original meaning, but instead actual rendered images that the browser just displays 
I would not mind that, people are already complaining browser is bloat, if it can be reduced to image viewer, why not 
I would complain. and don't get me started on blind people :D
you can always use gemini
flutter web
cursed by your name for ever having suggested ggwave to me
you cannot imagine how deep that rabbit hole is and its still not working reliably
Probably WAV, no clue
I don't think they have that many codecs in the old days. Probably some form of vorbis?
95 is 1995 right?
Vorbis is around 2000 IIRC so probably not hmm
avi with cinepak and pcm audio
š
raw audio data
You should look into more standard method of sending data through audio
Like in the old days of telephone modem
its also going to take ages
itās meant to be sent over telephone lines, not sound waves
Well, codecs were extremely heavy for CPUs back in the day
Yeah, I didn't think of that lol
I feel impostor today and got cooked with a research project I have to submit like I said
hopefully this works
So, I think the best way is to just make your own modulation 
it should be machine learning related and I have to draft a research paper with a couple of peeps 1.5 years later
now i gotta get a 95 iso and install it on a vm
I'd suggest learning typst rather than LaTeX
and then figure out networking because it never works
Bro, what are you doing with windows95 if you needed to install it anyway? 
the bitrate should be way, way more
im making a twitch proxy so you can watch twitch on... anything basically
with a web browser
do not start saying that this is false advertising because it wont run on your toaster
Note that browsers on 95 do not support opening modern sites
streaming pcm 
http/1
Will it run on my STM32F103? 
ssl is overhead
html might have to worry about
because im unsure where to get docs on earlier versions
I sure would hope you're not using ssl in 2025
This is getting closer to reality 
yeah ssl is a waste of everything just use regular http
twitch proxy so you can watch twitch on... anything basically
actual rendered images that the browser just displays

everything https is handled by reverse proxies nowadays anyway, nobody implements any of it themselves
I imagine the client would need to support it too
And that is the bigger issue than the server
all iām saying is thereās nothing to even think about here, just donāt do https
if you need it itās included with nginx or whatever youāre using
flask is the server
you put it behind a proxy and you have https handled
transparent to your app
We should have a dedicated cursed project thread here in #programming 
Hm?
i spent 5 minutes attempting to figure out why qemu woulnt run
found out i didnt have it installed
š
real
can we use python instead of css
like I mean streamlit or taipy
holy fuck
CSS as in Cascading Style Sheets?
everytime i tried using the nixos installer, it either froze the WHOLE system, stuck at 46%, randomly fails then refuses to wanna work with partitioning my ssd, whatever all the random problems
NOT a single time it wanna work
NIXXXXXX
NIIIIIIIXXXX

Ask Elvyn for help
USBs are slow
ITS NIX
ITS SUPPOSED TO BE QUICK
IDK HOW MANY TIMES I HAD TO FORCEFULLY KILL IT BY CUTTING OFF POWER

IT WAS USED TO BE QUICK
I wouldn't exactly expect a system that rebuilds itself for every installed application to be the fastest thing ever
have you tried manually installing
yes
ok
and it refused to work
skill issue
no drawback on runtime performance, its a onetime cost every time you rebuild
not that you have to rebuild to install
buy a new nvme
Or maybe Nix doesn't like your weird CPU
maybe
should be the same as other linux distros, they all use the same firmware
try to run a VPS
and check
oracle cloud maybe
it has an always free tier

????
run a vps on your nvme 
you might be onto something






