#programming
1 messages Ā· Page 361 of 1
I can't say I'm suprised
The only thing I might do is get the thunderbolt bois dealt with so it can be slightly portable
I really need a better way to mod those cables
I swear if I had those done I'd be more focused on actually getting a system that contains one of the framework mainboards and 4 P40 cards
you never did 
i just downloaded several GB of apple tts voices
they changed the va of the mc in the new season of scissor seven 
Btw Daniel (enhanced) is the tts they use for Daily Dose of Memes
me when hpppl micropython libs except theres no documentation for them lol
anyone in the letterswap have a T I could have? I must become The Man
I don't have a T only a t
are y'all still letter exchanging
vaint went back to Vani did she return the letters
or did she steal them and never returned them
vaint did WHAT? 
i guess theres a free t up for grabs
unless she already returned it?
oh
apparently it went back to t
yeah
t is simply a t spawner
or a t despawner

the best we have for rules is by best effort
i was looking through discord's OSS acknowledgements and found the source of so much pain
this fucking library https://github.com/words/stemmer
Fast Porter stemmer implementation. Contribute to words/stemmer development by creating an account on GitHub.
this shit right here https://words.github.io/stemmer/
oh is that why exact string search sucks and doesn't work
exactly
I'm back, apparently you produce infinite ts?
Ohhhhhhhhhhhhhhh that makes total sense
where do think your letter t's come from
I recall it infinitly sucking
Was watching an older clip of Vedal and Ellie talking about AI/neuro and I'd have to say its really interesting, even with him being secretive it's very interesting...
Idk if he has changed how he looks at it but (from what I understood) ellie was suggesting a system that was essentially what I do for thinking (gate model triggering a reasoning model behind the instruct model). But I'm also probably nowhere near as constrained latency wise given I have the advantages that come with 2025 models. reasoning models are only about 13% slower for TTFT (10ms) and 20% slower token generation (7 t/s)
can i have one?
ask again once i've finished my cereal
i hate this
lmfao
i have .gitmodules file that seems to be ignored (i renamed it to .gitmodule12313123 it, git submodule command still works)
I'm gonna be busy in a bit...
...watching (crying bc of) the LIFE mv
report back with your own finds from here
how does it work? it reads it from server where i cloned this thing from?
you've got time
fine š
searching one seems to yield exacts for one and no on, but if i search for on nothing is highlighted
:SAVED:
wdym?
discord moment
would you like a T?
i want to add new library to this repo im editing but it completely ignores changes i make to .gitmodules file
i can remove the file it will still git update without problems
oh i mean git package
or how its called
yes please, t is being stingy with their t
if a -ing doesn't get removed, for example an emote only the last layer of -ing gets removed
did you initially add the repo with git submodule?
t
I'm stealing this
no
too late
Also the way Vedal explains stuff it sounds like he struggles with post-hoc rationalization
i edited the .gitmodule file in notepad
is this the problem
i have to use command?
he has become The Manā¢
maybe. git submodules are configured with more than the .gitmodule file
Lemme capitalize it rq
wdym
There we go, kinda fitting with my current pfp
try undoing the change you made to .gitmodules if you can and then run git submodule add <repourl>
alternatively nuke and re-submodule
after a while it gets hard to see where the letters came from and if they're legit
teste e ine searche
it's mostly for fun anyways, at least until it becomes a server event
we need to put this shit on the blockchain 
just be sure to run this in the directory you want to clone the repo in. moving submodules can be a pain if you're not careful
all of the e's minus the solo e get removed
even though supposedly modern git does the dirty work automatically with git mv it somehow still broke when I did it
only in doesn't match ine all of the others do
the stemmer appears to speak italian
(I don't know italian and have no idea what I'm talking about)
More specifically it strips italian from the message
one may say it is $U-I_{talian]$
so e almost always gets deleted regardless of word
ing only gets removed once, so xdxinging should only remove the last ing
I can't search the message
oh no
I have to include the ` for it to find the message
it isn't stripped
okay, that is actually very useful
interesting behavi
r
This one will be fun, the ` is required and won't match without it present affects all of this text styling
it does rstrip but not lstrip
He's added a lot of features and updates that aren't on that list. Vedal constantly finds ways to improve the girls in ways that he will never divulge. I've seen a lot of AIVTubers that use memGPT/RAG, and none of them are nearly as complex as Neuro and Evil. Their LLM isn't the largest, but all the improvements that Vedal has made to them have made them very advanced as streaming agents.
i had to use commands to add packages instead of modifying the .gitmodules file manually. it all worked out now 
You can definitely make an AIVTuber with time, but don't expect them to be anything like how Neuro and Evil are now.
the reason why it's short is just in case people dont have the attention span to read the entire thing and only want the "important bits"
not making that up trust
I should spend some time on that, but it keeps getting sidelined
In case you need it for future reference, in addition to the .gitmodules file, Git also tracks custom config, data etc. for each submodule in their respective .git/modules/<rel-path-to-submodule> folder. It is basically the .git/ folder, but in a centralized place for submodules. Relative path to that folder is also referenced by the .git file in the root directory of each submodule.
tl;dr If you ever have to tweak submodules manually again, you'll want to check & fix paths in / for all of the above files / directories 
oh also unrelated tip but if git-submodule complains about a submodule not existing when it clearly does, check if you accidentally left a trailing / in any path in .gitmodules (in addition to checking that all of the above files/directories have the right content / are in the right place)
Well yah I get that but also idk, it's hard to explain. I get that he's careful but like avoiding explaining why post-hoc realization is a problem with llms isn't specific to the twins.
As for their model size, the more I work with my system, the more my thoughts on what they run on shifts to the lower end
it's fine surely people won't come in and ask that often anyways 

anyways I found a set of interesting python packages
mostly because despite being counterparts to each other, one has a proper server mode and one kinda doesn't
I'll work on it once I'm done with my NN project, it will happen trust
as in like there's an example
They aren't that low, but probably aren't huge. Likely around 8b but a lot of that was guessed a year ago, before all their current upgrades. They're definitely big enough to efficiently call functions and such.
My guess was originally 70B but that's shifted to 13B. I tend to overestimate though since my gate /sentiment analysis model is 8B and the core is 2 dense 24Bs.
They might use a larger model now, or at least have a larger model as part of their systems so idk tbh, but they probably aren't under 8B.
The problem is that based on the time frame, and various other factors it's likely not a 24/36B. That most likely restricts it to either 13 or 70b
shit bro we got gravity gun in the big 26
ye
half life 3 confirmed
it literally works the same way the actual gravity gun works
set the mass of the grabbed object to 1
give it a fixed constraint
???
profit
this is just how half life 3 will look 
the graphics are so realistic that your gpu cant even render them
its rendering each individual molecule, but due to lack of floating point precision in gpu hardware it gets rounded to 0
the backpack is there because i want to have death stranding's cargo mechanics, but idk if that would be too much and just focusing on the chud daughter is a better idea
i fcked around making like a garrys mod type thing before too, cuz i always rly liked garrys mod
its reall fun messing around with physics
i dont have the files anymore so finding clips of it is hard
what surprised me is how simple valve's solution is
compared to my own old one and what i've seen online
the physgun was actually not a garry's mod thing, it was a scrapped hl2 weapon
ye ik garrys mod is pmuch just half life sandbox
in fact the first gmod version all it did was enable the weapon_physgun.cpp
YAYYYYYYYYYYYYYYY
either way, it's surprisingly simple, just attach a physics object to a joint, teleport the joint to the correct position and have the physics engine solve it for you
oh does this listen to streams without opening them?
It can. But why would you do this?
oh i was just wondering what the purpose of it was
is it to just open the stream or
oh i found a clip where i added the gravity gun launching mechanic, shooting the cubes out of the air was fun
I have a list of streamers I monitor. If any of them go live it yells at me (as heard in clip) and opens their stream.
oh thats neat
I use in conjunction with specific streamers where I also have hte program kill itself upon first found streamer, which allows task scheduler to first run that (and open the stream) and ping my discord and then start a very loud alarm.
okay it seems to work 
i ownder when they'l add ai to blender
so i dont have to manually choose my physics settings
Everything clipping through the floor 
the idea is to fix clipping actually
https://github.com/ahujasid/blender-mcp (I have no idea if this is a real project or not but there was at least one sketchy fake download page on Google)
claude
MCP is definitely a real tech, interesting one as well
minecraft pocket edition

In this case it is Model Context Protocol
Allowing to place functions and details of a program in the context of an LLM 

I'm not doubting that MCP exists, I just also know half the ai tools that exist are probably scams for gullible people if you do it wrong
i have an idea
first physics, then subsurface modifier
so the physics wont take 20 minutes
MCP "interesting tech" 
It's a JSON RPC schema
I always laugh when I see people talking about MCP security as if the messaging protocol is the problem. If your MCP server allows Claude to read ~/.ssh and upload it pastebin then rip bozo, you deserve it.
i'm pretty sure there are plans to add AI to blender. but like the good type. so things to help you UV unwrap or remesh etc. that kind of thing
And how many secret tokens that should never be public are passing trough unencrypted network?
current vite release version :evilWYSI:
Is that the version with rolldown? I've been using the beta for a while now

the rolldown version is vite 8
vite 7 honestly wasn't that big a deal in terms of user-facing changes
the only changes were node and browser support
Wait, what node support?
Vite 7 dropped support for Node.js 18
Ohhh, I thought vite 7 can be used to make backend packages too
nah
you still gotta use express or hono or something like that
actually, the closest thing we've gotten to proper backend support for vite recently has been the environment api
but that's for running the code, not building it
Welp, still tsc it is. Can't wait until the golang version is released
tsgo is going to be SO fun
it'll finally (probably) make type-checking worth it to integrate into faster bundlers
e.x. esbuild, rspack, rolldown, etc
I didn't know I could use AAC passthru instead of re-encoding it so I changed it. I also didn't know what the advanced options parameters like qcomp or bframes did since handbrake picked them automatically so I looked them up to check. Then raised qcomp to 0.6 and bframes to 3 but there was no size difference between bframes=2 and bframes=4 in my testing. I use 27 crf for videos I watch on PC or for memes I share on discord sometimes since discord video embeds are small and look good even at lower quality unless fullscreened. 25 or 23 crf if I want both good quality and size
AV1 decodes fine when playing even on old hardware, but it's way harder to encode. I tested it with [this 80MB video](#livestream-chat message) because it had a mix of nuro stream and irl footage, and AV1 at preset 6 took almost twice as long as h264 with about 40% larger filesize when encoding both at crf 25. Thanks for all the suggestions 
I have a 1360x768 monitor which could be why it looks fine to me so can you check the h264 encode (crf 25) as well:
Why would you prefer jxl anyway?
it's just JPEG but better
useful everywhere JPEG was useful
"jpeg xl"
Yeah, thats what I mean. I usually just use jpeg, so why use jxl? What advantage does it have?
smaller
Oh, so better compression?

can losslessly reencode JPEG into JPEG XL and save some space
I'm sure JPEG XL has some other additions too, but this is a big feature of JPEG XL specifically
Hmmm, what about the performance? Is it faster too or just around the same?
it depends on your encoder settings
but typically jxl would be slower at its highest compression settings compared to jpeg
for encoding
decode is about the same iirc
pretty much "negligible"
note that jpeg and jxl are still a lot faster than png
Hmm, if it saves around 50% space then I think that is a reasonable tradeoff. So, how much better is jxl usually compared to jpeg?
and jxl scales with image size really well
like 20-40%, i would say
though, you get more savings on larger images
Oh, so it is worth it
you might even get a 20% saving on lossless jpegs, jxl is a better format
I did a test for jxl and you can survive a lot better at tiny file sizes
though, if you have a lossless image, reencode to avif instead
avif is pretty ridiculous
I've never used avif. Why is it good for lossless compared to png?
storage size 
lossless webp also really good in my experience
but that was a very specific (limited palette) test so mabe doesn't apply everywhere 
its like, not even close
Got it. But is it widely supported by browser tho?
though, avif is quite expensive to encode
we need .jaxl
^this is avif
tiny dog
this gif was 2mb at 128x128, and 70kb reencoded to avif
this is a cool site to compare
the biggest kinda stuff for me when I was testing things was this
Noice
its an exported image from darktable
which is always huge
also jpeg completely falls apart at tiny sizes
still, looking forward to jxl going into the browser
can finally revive progressive image loading

Damn what Skyrim mod has quickhacks
Yeah thatās possible.
Pro tip: lowsrc and src :)
Yes, thats html3 era stuff
But who cares, it has no reason not to work
See how it doesnt say why?
At that point ignore and use anyway. It has no downside, most browsers ignore it, because its so unused, but those that dont profit massively
(most Software Development is just enshittification anyway..)

why would I use the feature that tells you not to use it
instead of the working one
this is literally better
Not every browser can do progressive jpeg. Ofcourse best bet is to do lowsrc + src (progressive)
Which is why its stupid that they say "depricated" they should say "use this instead, and consider still using lowsrc"
It SHOULD say so
it is deprecated
also like
which ones??
as far as I can tell all of them do
Its stupid that they say Depricated. Thats what i said, it costs you almost nothing to also add lowsrc
but there are better alternatives?
Older ones, but those older ones can handle Something like wikipedia fine and there isnt a reason why Something like wikipedia shouldnt offer lowsrc as example
doesn't it require encoding a smaller version of every image?
even then its not complete lack of support
Its not an alternative though. Its am Extension. Using as example a heavy compressed bmp or png/jpg (normal, non progressive) as lowsrc (greyscale, low res, low bitdepth), and then having src as a progressive streamed in image would work.
its just "this will load like a normal jpeg"
Ooh, discussing optimisation are we?
of course it's Internet Explorer 
Yes, but the 1KB of a old super low Quality image does nothing storage wise.. Worst case you compress at runtime and keep in ram
ok but a progressive image has similar negligible cost
I mean my Websites are IE3 compatible Out of spite but yea
this does not sound like "costs almost nothing", implementing this sounds very annoying and unnecessary
This reminds me of the idea of memory area arrangement, limiting specific areas to specific purposes rather than dynamic allocations
Akin to the days of DOS
Somewhat yeah. Although lowsrc does allow it working as a fallback for unsupported image formats.. idk if HTML5 has anything for that.. probably tbh the "fallback text" could probably be abused for that
Its a case of having to reimplement it, older servers did this on the fly.. if you enabled it iirc
PDFs recently adopted the JXL format so that will pretty much force chrome, firefox and others to add full support for it, since they all want to be able to render PDFs
Idk how that reminds you off that, but it's not a particularly bad thing.. probably not the best but it works.. and it worked with the limited memory and compute they had..
Yea
ooh finally

I was just testing if that api can actually make skyrim connect to some socket
But yeah its funny too 
I think the severe limitations forced many Devs to optimise. There might've been a point where new hardware came out that ran the incredibly optimised software incredibly well, before the developers of the time took advantage of the more abundant resources of new hardware
If you tell people you surfed the web back in the day on 4MB of ram, and that you want a simmilarly experience nowadays, they look at you funny.. why must something like firefox eat 1gb of physical ram, and snowball into 40GB of virtual memory.. it kinda pisses me off ngl. I can understand complexity, but if you consider we did very simmilar things waay before "Half a gigabyte of ram" was a thing..

Isn't that crazy?
I've seen very simple programs needlessly taking up to half a hundred MB, and this just for meagre tasks
Don't get me wrong, i love av1, jxl and so on, all for more efficiency, lz4 is my favorite. But please make it run decently.. its why i still use win 95/xp, cuz those were still limited.. putting myself in those limits forces me to make better code
Windows 11 task bar being Node.JS as example.. it makes me wanna cry. I must be scared to open and close an app too often, and end up with a JS memory leak
Dotnet compilation is scary. I made a program to simply write to console and I got a file in the size of MBs, just a small binary program that merely writes output being no bigger than like 50b
I wonder if the recent sluggishness we experience from the latest windows operating system is akin to the days when MS revealed vista over xP
Like, maybe ms has higher hardware standards than the complacent consumer base idk
That is an interesting thing. I do know some .net devs that are insane and managed to absolutely slim it down. The funny thing is, the main thing i noticed? .net 4 and such
For sure.
I mean, just install xp in a vm and allocate one of your many cores and a fraction of your ram
why do you care about virtual memory usage
Even though your pc is now running two OS's, through translation and so on, it will still feel snappier.
I have seen chrome reach multiple terabytes, it doesn't correlate at all to physical usage
I could run that on L3 cache if that were possible it's so small
Because its rediculous? Virtual memory is more or less swap if shit hits the fan. Not to mention, the more memory you have, the more memory your cpu has to look through (if you even have to look at the data ever again, which often you dont, so you could also just discard it)..

its not physically allocated
Is it true most people use AMDs for CPUs now?
do you have the UK secondary school teaching understanding of virtual memory
it's the only place I've heard it called equivalent to swap
Hmmm, I wonder why my drives are so full
Did you see that one EPYC cpu? 1.3GB of L3, and with older AMD microcode and coreboot modded in a way called "L3 as RAM", you could install XP tp L3 and run it from L3 and in L3
amd desktop marketshare is about 33% in late 2025
I have a desktop tower from 2012 I think. I think that was in the era when Intel was standard
oh
https://www.goodreads.com/book/show/15951196-what-makes-it-page
Very nice book :)
I'm wrong about everything it seems
well, it doesnt hurt to google something before you say it
though asking a question isnt by any means bad
oh mate server CPUs are no joke
wrr
Someone made a software for that if I recall
cache as RAM
speaking of windows
A fortochka (Russian: ŃŠ¾ĢŃŃŠ¾Ńка, IPA: [ĖfortÉtÉkÉ] ) is a small ventilation window generally not greater than 35x45 cm. It usually spans the frame of one window pane and opens on hinges independently of the whole window. Fortochkas are in common use in Russia, other post-Soviet states (Ukrainian: ŠŗŠ²Š°ŃŠøŃка, kvatyrka), and Fin...
Fortnite reference?
Wild what you can do if you know how to work with computer data
I said "if shit hits the fan". Either way, if it stores data, you gotta store it on physical. If its just boilerplate, you might aswell allocate it when you need it to actually hold data. Where do you think the "virtual memory" stores its information about what virtual memory is supposed to be what? Exactly, in ram. The filesystem as example also uses storage to know where files are located, even if the files themselves are 0 bytes in size





ermtosis
idgi
even then how is lots of virtual memory an issue
Its slow









ermeiosis
it dentge
easiest way to manage memory is to virtually allocate a terabyte and and write a bump allocator
Unfortunately I'm not such a person ... I tried running arch on that 2012 computer, which doesn't have bad hardware by any means, with the dream of it being buttery smooth. Alas, it has jittering here and there running plasma wayland.

I dont, no. But its also a difficult thing to benchmark, as what i tried to explain depends heavly on what the OS wants to do.. and my main point was "no program ever needs 100GB of virtual memory" if it would, it would slow to a crawl.
allocating virtual memory doesnt cost a program any performance
at all
the cost is zero

Because by then, when those 100GB virtual, on a 16gb physical machine, are fully used up, you hit swap.
Maybe it's an NVIDIA issue (shifting the blame)
I think you have a fundamental misunderstanding of what virtual memory is and why it is used
some people just like to show up and say thing that make no sense
Is it part of ram or no
it depends
ram is physical memory
I could mmap a file for example
If 2 programs allocate 80% of ram each size what happens
The cost is 0, aslong as your allocated or active memory is under the physical amount of ram. Which often is the case. If you have active memory larger than physical ram, on virtual memory, you get slowdown.
virtual memory is an os abstraction
allocating a terabyte of virtual memory is completely fine
I mean allocate virtual
it doesnt cost any physical memory
Ooh ok but then wtf does it do
Yes, but USING the actual terabyte is slow.
its an abstraction
ok, so we are in agreement, that allocating a large amount of virtual memory has no performance cost 
Just let everyone allocate 500 tb
this lets me access the file as though it's an array
but doesn't load it into memory
Notepad and discord and browser give everyone 500tb
virtual memory allows memory sharing too, a program that says it uses 10gb of shared memory doesnt typically use a full 10gb of physical memory
if one program only would do that, yea. However, since a modern os runs THOUSANDS of programs at a time, you end up with tons of "virtual memory descriptors" which take real memory.
If you arent USING the virtual memory terabyte ACTIVELY you should free it.
So you wanna tell me the Memory Manager of an OS doesnt use up any ram, yea sure.
consider reading the memory manager section of windows internals
oh no, it does, its just essentially cost free 
Are we talking about page file?
We had some server issue that was out of file descriptors
page files are a separate concept to memory pages
basically the OS divides memory into memory pages of ~4kb and those pages live on virtual memory adresses, each program also runs on a arbitrary memory adress that is invisible to the program. so the program can imagine being on a certain memory adress, when in reality it lives on another adress in the actual memory. Also having pages means that the OS doesnt have to allocate programs on contiguous adresses, a program can just live on multiple pages scattered across memory, even with some of those pages living on the swap disk
To me, even kilobytes used is a waste. Why waste those kilobytes when i could just implement something as streaming, or manually keep everything in real memory, and constantly keep everything as small as possible?
it should be noted that not all virtual memory is paged virtual memory, virtual memory is a broader concept
But who and why would let it take more than physical ram
its patently stupid to not use RAM for the reason RAM exists
if you want to compute everything in just one register you can

Me when some dumb dataset wants to load into memory and so I end up with a 300GB pagefile
Wasn't it you who wanted 10kbps internet konii
here I'll save you some trouble - Windows Internals Ed.7 Ch. 5 Page 302
(this causes the memory manager to remember what pages belong to what program, costing real memory. Fine when you only run 10 programs, but when the os already runs 1000 youre outta luck imo)
That actually happened, when I was still on Windows
This is true aslong as ram is free. Oddly enough, your program isnt the only program running at a time.
Are the kids discovering virtual memory now or what
Its gotta store these virtual addresses somewhere physical, no? 
page table
This.
a limited resource is limited, shock.
just because a resource is limited doesnt mean you shouldnt use it.
And page table is physically located where
this conversation
Oh in os
Your process
It also means you shouldn't just use it uselessly. Free anytime as soon as you dont need it anymore

So process stores its virtual adresses in its virtual adresses?
i dont get it
Its stored by the os as part of your process
Ok so its os
no, the virtual memory is invisible to the processes, it's the OS's job to handle them
Yes
And os limits it
And the os uses? Real memory..
ye
at least in the case of browsers the high virtual memory usage is mostly for security
you can read up on gigacages and similar mitigations and sandboxing strategies yourself but basically you just allocate a ton of inaccessible memory that nothing is ever supposed to touch, if something does touch it then it could've otherwise accessed some internal structure outside of the JS heap, which you don't want
none of that memory is ever backed by physical RAM
Therefore we cannot actually allocate million terabytes of virtual memory
Yeah, Canaries.
If we have 1gb ram
it just cannot be physically backed
wrrr
not really what that is, no 
the amount of dentge here is hurting
So it looks like "from this physical adress lies 500 terabytes" but actually its 1gb and when it exceeds program crashes out of memory
Its not. The os has to store page tables. If you make a billion 2mb virtual memory areas, youll OOM.
Its functionally very simmilar, just "at runtime"
you'll have a bigger issue if you spawn 1 billion processes
I need to buy physical version of that one day


when you allocate memory from the OS, or ask for lets say 100GB of ram, the OS gives you the permission to use the RAM, but does not allocate anything yet, only when you start writing / reading on the memory, the OS does the allocation of that memory
note that it doesnt allocate the full memory
Considering modern osses that sit at 1000 processes just to be booted, we arent far away from it.
Let me clarify, i dont mind virtual memory, i mind excessive use.
This is why os doesnt crash when Firefox uses all ram
Right?

Very cool
im running osu, librewolf, discord, and im currently sitting at
user@nixos ~> ps -e | wc -l
333
333 processes
that's all processes, even paused ones
every household that uses windows should be legally required a cooy
this is with firefox, task manager, spotify
Thought just occurred to me. A byte is technically just a number, meaning you can use them for any quantifiable application. 1Bms is 255ms
tailscale, sunshine, some extra stuff I don't need
One or two reasonably sized virtual memory areas are good and all, but some of the shitshows of code, are rediculous.. either you havent seen the horrors i have as a system Integrator.. (i have to get the shit code from others working on a underpowered machine because corporations be like that), or im talking Chinese..
A guy recommended uh
Practical Malware Analysis
Rootkits and Bootkits
Windows Internals
What Makes It Page
When I asked for advice on bachelorās thesis source books
I read them all and still donāt understand shit about Windows
windows internals because most malware books are pretty ass
i dont think task manager shows kernel processes does it?
you'll get hands on when actually looking at malware
Well thats linux, and its one of the few thats still sane. Linux used to be at about 30 a few years ago tho so.. yeah.
i was wondering why my number would be higher than yours
you'd need system informer with kernel driver enabled
no it didnt
alternatively windbg with kernel debugging enabled
ive been using linux for over half a decade at this point
Linux L
its simply because linux isnt hiding the kernel processes from me

Yes it did. You can still get there with LFS nowadays if you don't believe me lmao
that's not generic "linux", that's a very particular setup 
Linux Userspace became cancer real quick lately.
Idk, gentoo gets there aswell
ArchLinux with a very careful selection too
wsl is cheating a little
isn't this entire conversation ragebaiting
i think so

or this person is just stupid

one of the two
I love ragebait
It absolutely isnt, linux used to boot to desktop with something as big as kde3 with less than half a gig of ram and use a browser. Try that nowadays lol
So rude 

i pledged to be more rude this year
the consequences of moving this channel next to genchat
move it into neurotic neurons

do not
Can we go back down
replace nn with programming
Anyways, this all just proves my point
One day people you didnt enlighten will install software on your isp server konii
It was possible with 4MB back then, easly. We dont do anything more special and suddenly were at a point an OS itself wants 4GB to run horribly


i cant stand this anymore please call me back when the brainrot is over
Windows moment

type
type
š«©
i dont like bad types 
can you make your linux machine run on 4MB soup box
Woah, even chay is not immune to letter swap
not if you want to load the nvidia driver blobs
om
I haven't tried but certainly less than Windows
I wish it was just windows and not like 90% of distros. There are atleast alternatives with linux so thats something to hope for.
I did with 8mb on a 6.1 kernel, modern display server bumps it to about 120mb tho.. DSL linux back then managed with 14mb or so iirc

TinyCore gets my respect as a DSL "second version". Its not quite as small but its still crazy for how much more modern it is
DSL2 however kinda missed completely, iirc that doesn't even attempt to boot with 128mb ram, unless they fixed that
So the conclusion on virtual memory was that system actually doesnt have trillion processes so virtual memory is basically free for each process
Yes, aslong programs arent stupid they can pretend to have a shit ton of memory.
Why theyd do that i still dont understand but im not gonna bother as i target win95 anyway..
Why stop there, 4B
Yes why
1 32-bit integer
Because its convenient and easy to use and gives freedom
this conversation dawg
I will torture the tlb and you can't stop me
Memory leaks have nothing to do with this, i do not support memory leaks
tfw 4070ti can barely train 4B
4 breads
In case of any federal investigation
its okay to leak memory as long as your process isnt persistent

4 banana bread
i support leaking ddr5 memory from the memory store into my pc
I misrembered, it is 4kb not 4mb, I just remember it was 4 something
I want banana bread 
I need to look into buying banana bread around me or my workplace
it's been a while since I last had one
Id guess 4mb because of a tradeoff between "memory waste" and amounts of pages possible using up more space..
half the grids content and you gotta remember double (or 4x depending on how accurate you wanna think) of pages and their identifiers or so.. idk yapping and guessing here, im comparing it to sector size on disks..
Nobo i just wanted to give example that you can go even lower regardless of mb or kb
It would be stupid to run entire os on 4mb when you have 128 gb of ram
Imagine how crammed it would be
And slow and
Ineffective
Tell that DOS lol
People use more ram because there is more ram
Happy with less than 640kb
Same with disk space for games i heard recently
The ram days are over
can I have more ram so I can have twice as many dos installs
aight chat here's a theoretical question:
you want to become the smartest little programmer.
however, you have to choose your first PC at a specific time of your life - and that will be the maximum specs you can use for any task until you're a senior developer.
if you choose it early, you will be starting your programming studies early on small hardware
if you wait, you'll have better hardware to program on, but you won't be able to do anything until you're old enough for that hardware to exist
when do you buy (and what)
Sure, at that point you kinda loose quality of life but yea
when are we gonna get 64gb consumer sticks
Damn added the consumer
you can program on any relatively modern hardware.
unless you need compute for some reason, you can use any computer.
current PC market is
idk if I would want to buy anything currently
Oh I just found out my nvidia GPU is no longer supported by pacman install. That aside, why are gpu drivers so massive anyway?
proprietary blobs presumably
Can you really choose when
As soon as you can
Unless the wiki is not updated š¤ considering it happened 2 weeks ago apparently
You're going to have to grab the 580 driver from the aur
~2005-2010, probably a 4 core machine with about 8gig of ram.
My first machine i got at age 3, a pentium laptop with 120MHz and 16mb ram.. wouldn't mind still using it ngl
I got pc when i was 10 and i hate it
idr what I got first for programming
they exist
I know that Crucial sold some, but they're kinda
now
I think it was a pentium laptop with a hdd?
hated that thing but at least it played gmod

couldn't keep terraria at a stable 60fps om
I was playing dos era doom on my pentium machine lol gmod wasnt even remotely doable for me
I have a 1070. Is 580 part of Pascal too?
i mean pentium is pretty wide
š
Yea mine was 120Mhz like pentium 1-2 era or whatever
my first pc was a pentium 2 300mhz with somewhere between 16mb-64mb ram too
i don't think i'd have enjoyed having that for high school š
the CT64G56C46U5
very good name 
Wtf 220 gb?
I did PowerPoint on mine in 2014 lmao
With careful software selection it was enjoyable
those ai companies taking everything away from me
Surfing the modern web was the worst
2014, modern web.. hehe
https://archlinux.org/news/nvidia-590-driver-drops-pascal-support-main-packages-switch-to-open-kernel-modules/ that's the last driver that supports pascal 
more HBM
I like that zram made downloading more ram real

Pretty cool that I get to be more involved with my system as the hardware support reaches end of life. I wonder what percemtage of the hardware on the tower can no longer receive updates
why is my computer making new noises
You have angered the machine gods
Thats usually a good sign
PSU?
Im gonna scare some html developers with this sentence: ||i willingly put tables inside of tables in my free time||

if you ever need help, i know a good asylum
my psu makes coil whine too
LMAO, i think im past that.. like said, i target Win95 without updates and IE3 xD
wait chat it might have been my door
That does sound less than ideal
Bro found doorwhine
ok no it still happens
Bro did not find doorwhine, nvm guys
something seems to make cool whine noises when the door gets slammed in the hallway
I sure hope I don't have electrical issues in my walls
Perhaps an another door or the pipes vibrating?
who knows
parts shopping continues, and i have to ask
does 5060ti 16gb version make sense
@true hemlock
(to buy new)
going for cheapest 16gb vram?
consider that it will be overall slower than a 70ti etc. cause it has less of everything else, but also idk
not really, itās just the matter of going one step below the 5070
i mean it's both a step forward and a step back
i know memory bandwidth sucks on the 5060
sorry I was snoring in the walls
its decent
Hey, Iām Bella. Independent, gym lover, and all about good vibes. Iām into music, cooking, and chill late-night chats. Always open to meeting interesting people and seeing where things go
is there a convenient way to make a table in c++ which will have pairs of "int and string"
what are you going to use it for and how much you'll pay for that 5060ti 16gb
and you can take ints using strings as keys
only map?
Pretty sure they exist
bot
std::unordered_map I think
that one i know but it seems a bit over complex
just do struct array and index by iterating it. that'd be the fastest implementation
that just how C++ is
yeah idk why i said that
if you just do iteration and never lookups then ye a plain list works too
itās $560 cheapest here, weāre at like an electronics market so it is literally the cheapest you can get here
weāre also considering the 5070 itās around $725
choose appropriate data structure depending on how you access it 
its because im working with alien code which is c++ and i dont know if its going to be easy to use another language in pair with it
I can't think of a use case where that'll be a large scale table do i presented the laziest method
all prices for new

you're also a bot 
5070ti.
for $560 bucks its not even worth it
Can't you get anything secondhand instead? I think you can get a better deal
not considering second hand because the guy building the pc does not want to
if you want to actually use the C++ features and can't expose a C API to use from that other language, then yeah, better stick to C++ for everything 
ye there are much better deals second hand
Oh, if it is not for yourself then just go ahead
nah
weāre together say hi
What another language and how are you interfacing
Hi
i could probably use python

No cats?
okay, would only say that $560 gor a 5060ti 16gb is pretty shit of a deal
5070ti has double the compute and bandwidth
the thing is, i will probably not need it much anyway, most of things will remain in cpp
so writing 1% of program in python sounds stupid in this case
Hi chay! I think you missed my message about your name lol
I think just using most natural data structure is best
E.g unordered map
i'd help optimizing overall budget for that
just wait until you hear about the price of a 5070 ti 
Well in the current market, everything new is shit deal
Just sell 8GB of RAM that'll cover it
i misread
what is it, 500$? 
What is the use cases anyway?
gaming and occasional ml
well i don't see any here
thought he said 5070ti on the $725
so as far as i'm concerned they're unicorns

If only 
nvidia says 5070ti is 500-750$
That is indeed what he said
I've seen them replace 4x32GB kits here as the chepest 128GB options
no he said 5070
not many options left then
occasional ml will like the 16gb vram and that it's not a radeon card
for purely gaming a radeon card might do better in the price range, idk
i'm talking about single 64gb sticks
is 2 sticks better than 4 sticks
yes
4x sticks is gonna be slower clock, and usually can't xmp/expo
Yeah 2x64GB kits replaced 4x32GB kits
Oh you mean as in sold invidually
yeah well they don't exist here
oh and this exists. i think crucial made them but very rarely there's any batch coming out
so 1 stick would be the best
No
DDR5 yes a lot
but why no
No that's half the speed of 2 sticks
single stick is far slower
2 memory controllers
you'd want 2 at minimum
Dual channel fastest
do they lose their ddr power
DDR5 really wants one stick per channel for good signal integrity
consumer CPUs have two channels, so you use two DIMMs
if the ml is just running random models ig just take the 9070/xt
or its actually 2 sticks each is ddr and in pair its x2 more
No there's just two physical memory channels on the CPU so if you have one stick one goes unused
oooohh
Am blind lmao
i see 
same i misread it aswell
i didn't get to sleep last night
And with 4 DDR5 sticks it can be too much for the memory channels to handle 2 sticks each

So 2x DDR5 is optimal
so 2 sticks is the best while processors have 2 memory channels
On DDR4 there isn't too much difference between 2 and 4
nah, ddr4 is optimal
-# because of ram shortage 
Unless you have a big boy CPU with more channels 
can confirm 9070xt is good for gaming and fine for ml
Yes
There's also processors with like 12 memory channels where 12 RAM sticks are optimal
Wdym? DDR4 is already more expensive than DDR5 due to supply
Now with shortage, everything got worse
i can confirm this. my board used to able to do 4000mt/s on 4x16 ddr4 but neither 2x32 nor 4x32 would do xmp
ddr4 in many places is still way cheaper
I swear to god this country is cursed
ok that might be big but also itās more expensive than the 5070
Less demand I guess but in return no production
ye like here
even the 9070?
They should really have just waited for this to launch the thing
The 1X World Modelās latest update is a paradigm shift in robot learning: NEO now uses a physics-grounded video model (World Model) to turn any voice or text prompt into fully autonomous action, even for completely novel tasks and objects NEO has never seen before.
By leveraging internet-scale, video data fine-tuned on real robot experience, ...
Their reputation is shot now
9070 same price
DDR4 32GB: 250ā¬
DDR5 32GB: 400ā¬
DDR4 still cheaper
as 5070
is 1X
Still, that's 100⬠more than I paid for 64GB of DDR4 just a few years ago
oh, it's the spy robot makers
god damn
1x developer sounds like a shitpost title
cunt where did you get these prices at in 'straya 2x32 ddr5 cost $600
we would've had AGI by now if they just named themselves 100X
They made the home humanoid bots that at launch only really worked through teleoperation from some guy on minimum wage in the third world
i hate that my screenshots appear smaller than they are because upscaled monitor
At launch it was basically slavery with extra steps
This jimms.fi we have in Finland?
ye I looked it up
I forgot the name, but I will never forget the robots dead eyes
wtf

i have decided to just go with boards with ddr4 sodimm istg
for my next briefcase build
Slavery without immigration
Yeah the only thing it could do autonomously was open doors, you ask it to do anything else and it would go into "expert mode" where one of their "highly trained staff" would take over
also 'stray dollars =/= euros, 400 euros would be around $700 in aussie currency
tony at home
$600 in USD, not AUD
at least according to the exchange rate
You were basically buying a slave in robot form it was pretty fucked
Funny, I found a 32GB DDR4 stick for 70ā¬
this looks nice 
Laptop though
im aiming for something like these
1000mhz 
2666
Oh no way a real desktop kit
2666 too though
Seems to be the 2666 32GB version of my 64GB 3200
where are the extra 16% from
From each thread doing the extra 1%
If you want speed, 3600 is this much
cpu on anime power up moments
Why is there 2GB sticks at 200�
Well let's see what they got on the DDR5 side
So obviously 16GB is already more than 32GB of DDR4
And again like double for 32GB compared to DDR4
More than double even if you count the 2666
And that's only DDR5 4800
damnn
i have 2x16 gb of ram and its lowkey tempting to sell one stick
but like when will it go back down....
2030 probably
This is basically my RAM
Except a bit faster because apparently the 3200 version is just gone now
I got my RAM for 150⬠originally
Just like the 2019 chip shortage takes a good few years
So you can kinda still get 32GB DDR4
64GB don't even bother
DDR5 don't even bother
At least mine is only around 3x more
I just remembered I still have a spare 2x8GB 3200 kit lying around
CL22, very latency
this guy's funny
I love his interview series, esp the python one
1.2V though, I wonder if it'd run faster with 1.35
i'm back... day 3 of making a discord tts inside apple shortcuts...
Because of how shortcuts works, I think the only way I can switch tts voices is by making an if statement for every single voice... oh boy this will be fun
me in the auction house in random mmorpg / warframe selling a 500 worth item for 500k hoping for a rich misclick 
gm
Holy
i could get 2x 3090 with that wtf
i could get 2x 3090 with that wtf
i love how the messaged was doubled just like the 3090
also this presents an idea...
just buy gpus and use them as ram
literally
would you rather
64gb ddr5
or
48gb gddr6x that comes with 2x GA102 chip free

I'd take the GDDR6X
Hey guys!
Anyone knows a software that won't crash while trying to edit/cut 10hrs VODs?
Except custom python scripts...
are you getting ooms?
nah, i tried 2 free programs and they just crash in silence or freeze and stop responding (probably going 1 frame per hour or so, trying to process things)
try some ffmpeg ui
How about this cool thing called ffmpeg?
who knows, the networking implementation is a black box to me
in tls we trust
Hem
ffmpeg without ui works just as well and might honestly be easier to setup 
Absolute ffmpeg
Haven't seen ffmpeg in my google searches, will check that out then, thanks 
ffmpeg is literally what everything runs on
The core tool behind basically everything media processing
blazingly fast inference
12 tok/s on 4 bit, 1.1gb gguf model
its pinning all of my cores to 100%
12 per second?

Itās warming up

i have a mid-tier cpu
apparently its actually 15 tokens per second
its not using avx512
whatever
I see the cursed idea in this channel never stops
wrrr amd cpu go brrr
wrrr
pull out de threadrippa
or my name isn't pipkin pippa
i have an rtx 3060 that i could try and use
but im worried
that it will turn my system into ash

what psu do you have
bash
Why would it turn your system into ash?
im more worried about the gpu than the psu, the gpu is the one that killed my (former) system
Ohh
its some 750w gigabyte

i dont remember it, its a fire hazard waiting to happen
But konii... How did you know it was the GPU fault?
it might not
because i put it in another system

Lmao
30 series are known for the transients
Hello, I'm the new guy. I'm sorry for interrupting. I speak some English, so I'll just be watching. I may join the discussions in the future
i set it back to zotac under warranty and they said they didnt find any issues with it
but im a little concerned to try and use it in this system because i really dont have the funds to build another decent one if this one dies
wrrr
im trying to say that the transient on pcie connector could've affected the whole 12v rail
I need something to do with my 64gb of ram. My Thinkpad is to overpowered for what I do daily.
if the topology on the psu is shit
Hmm, transient shouldn't burn any other components no? Unless the PSU is some no name brand
.
.
whatever happened, my old psu, motherboard, and cpu were completely unusable afterwards
i had to rebuy everything
terrifying
what psu was it though
R. I. P. In Peace
i dont remember, it was a couple of years back at this point
evga or something
not sure what the wattage was
it drove the system fine for like, 3-4 years
Yeah, even some branded psu have their own tier too. Like, the silver is the absolute minimum and then gold and the platinum
some evga n series are mining psu and those are ass
also w series are the same with cheaped out components
the tier doesn't tell shit
and then it wouldnt power on one day, so i tried all the components and they were all dead, and when i tried the gpu in another machine that machine proceeded to die as well

It is a best effort. So at least there is something rather than nothing
it could be a unit that uses group regulation and one transient on your 12v rail the whole shit goes apeshit
regardless of tier
I had a APEVIA PSU in my old gaming pc back in 2012. Gambled having a FX 8350 and 7990 running on it.
the translations im getting with this 1.1gb model are worse than deepl
time to try the "not 4-bit quantised" version
I wonder how difficult it is to engineer an ATX PSU
Very
Itās a small transformer taking 3 phase AV converting to DC and stepping up like crazy to do so.
How different is it from some industrial smps?
dc-dc components are pricy
I don't think atx uses 3 phase 
Same tech different application but at a much smaller scale
Walls do. You need to power to come form somewhere.
Wdym? If it was 3 phase I don't think they'll use IEC connector for it
https://docs.google.com/spreadsheets/d/1akCHL7Vhzk_EhrpIGkz8zTEvYfLDcaSpZRB6Xt6JWkc/edit?usp=sharing
this is a great read btw
My bad itās single phase






