JPEG XL

Bud spencer Restored

Dz gas do you know the season

easy fab solution is a permanent fab containment channel

seems like they where either kicked or left

we did try that, that's what <#840831132009365514> was made for, but he didn't use it

it is a joke

it depends what you consider a CPU I guess, for instance you can emulate riscv instructions or run linux on an fpga

you can run windows in an emulator too

and the fpga is presumably programmed to run some kind of CPU

im talking about emulating calls themsleves and not really "emulating a CPU"

what do you mean by "emulating calls"?

I *believe* it was this ? https://github.com/lunixbochs/usercorn

I cant rember I have a lot of misc junk saved

that's CPU emulation, just with fancy analysis stuff attached

oh not that one then, one sec ill keep looking

in essence though, it was very similar to what wine does, it will just do system call emulation, and use JIT to turn any architecture specific calls to whatever needs to be done

uh, how do you run all the code, then?

like how do the risc-v instructions get executed

basically something like box86 or fex

that's emulation

sure but it's not cpu emulation

though fex does emulate a full kernel of sorts iirc

it literally is CPU emulation

or ok I guess I see how you mean it

not really? it doesn't do anything like memory management emulation or anything like that, it decodes the riscv calls and translates them to something more usable, it's not like they are reimplementing an x86 cpu

though I guess you could consider the calls themsleves to be the main part of the cpu and everything else just the technicals

yes

hmmm

I am conflicted with that

the main difference between something like qemu, which can boot a full copy of Linux, and something like box86, is that box86 *only* translates x86 machine code into some other kind of machine code, like ARM, and makes sure regular user visible details look right, while qemu handles all the different states a CPU can be in and talking with BIOS and stuff. But the actual x86 part of qemu works the same, it translates from one x86 to ARM or whatever.

of course, anything box86 doesn't handle is handled by the host, so you even if that doesn't count as cpu emulation, you still need a cpu.

well `qemu-system` actually still does a lot more since it needs to handle things like mmu emulation, pci emulation and what not, `qemu-user` however is much more like box as it is also a userspace emulator

the Linux kernel infamously requires the use of an MMU 😔

fwiw you don't need a TPM 2.0 to run Win11, just to install it

Skylake CPU contains a TPM 2.0 that I normally leave disabled

I enabled it when I installed win11 and then I disabled it afterward, and the OS still functions fine

you can technically install it without it anyway by doing some registry stuffs

Yeah, they simplified it to a single registry key labelled "Allow Unsupported Hardware" or something along those lines

https://jkkweb.sitehost.iu.edu/articles/Kruschke2013JEPG.pdf 😁 “JEPG”

(Journal of Experimental Psychology: General)

I can see why they wanted to add to their requirements though, as it means after Oct. 2025 (Win 10 EoL) they can really push things like full disk encryption and device bound passkeys.

for the last 3 years, I have been running a project to save videos, creativity, Russians, on the topic of "gacha life" (is real). it was not a serious "lazy" job. so far, 2 years after the start of work, I have not discovered that since 2020. 20% of all the videos I saved, namely 10 thousand out of 50 at that time. just deleted from YouTube. In September, I took this case seriously. having fuck YouTube through search. even then, I was faced with the fact that the YouTube search does not output 0.001% of all the videos that exist. next, I used Google. and at the end, Yandex Images, to search for videos, and then download all videos from the channels. thousands of channels that I manually checked. Thus, I created a torrent of 451 gigabytes with 108 thousand videos, 99% of which are Russian videos of the "gacha life". 10% of all videos that are unique at the moment due to their removal from YouTube. magnet:?xt=urn:btih:2DDF830E80768BA0B5378840B08EAEAEE62AB2B5&dn=gl&tr=https%3a%2f%2ftracker.tamersunion.org%3a443%2fannounce

my friend has uploaded all thousands of videos of my archive from gacha life to archive.org — https://archive.org/details/gacha-life-archive-by-dzgas ( https://archive.org/download/gacha-life-archive-by-dzgas / ) you can access any video and watch it immediately from the servers archive.org just by clicking on the link with it: https://archive.org/download/gacha-life-archive-by-dzgas/gl/

<:This:805404376658739230> I wrote all this in order to introduce the case of a problem that I cannot solve. maybe someone here knows how to solve it. all my search work is solely the random finding of videos, despite the fact that, of course, I have saved all the popular channels, this is just a drop in the ocean. and all my work was saving random videos that "the Yandex crawler" found even without using the api. The Problem: how can I find on YouTube all the existing videos containing the name gacha life, and having more than 100 thousand views. if just think purely logically, this problem is solved very simply. — all the titles of all the videos are Write in Google databases, the number of views is just too. and Everyone has access to these databases, you just need to enter a word in the search bar. but here's the problem I'm facing — despite the fact that all videos can are sorted by the number of views, (the function is available on YouTube search). .... But I can't even get 0.01% of the videos because the youtube API **literally** limits the number of videos received to 50 pages of 10 videos (in api spec docs)

a lot of people expressly don't want disk encryption. I imagine it might backfire

It's not just TMP either, I'm on a Ryzen 1700 which is 'unsupported' even though it seems to have all the requirements needed

I mostly agree re FDE. It is something enterprise users want, because they will have centralised backups, and a high chance of at least some devices going missing. For consumers it provides little benefit and increases the risk of unrecoverable data loss. Either way, by increasing hardware requirements it does make them able to fully implement these new usecases, even if just for those who want them.

Well I disagree, enterprise can always purchase hardware with TPMs

They need some features of the newer cpus for proper core isolation feature. It's not just TPM.

Sure but it's ridiculous to refuse to install on consumer hardware that lacks a cpu needed for optional enterprise features

My CPU is from 2016 but win11 would not install without me screwing with it. think about that

can't install android 14 on an old phone

also requires tpm

most people aren't sticking with old hardware and people who are will do the mods anyway like get real

the restriction isn't about the users installing the os

right, but what about a user with, say, a skylake CPU who isn't sophisticated enough to do the registry hack to allow win11 to be installed

what happens in 2025 when Win10 is EOL and it stops being safe to use?

you're telling people to purchase new hardware even though their existing hardware *can* run the operating system, it just won't for, some reason.

there's definitely been a paradigm shift in windows

for example win10, released in 2015, still had support for 32-bit x86 which stopped being manufactuered in 2005

the point is you dont want a user to buy it and wonder why a feature doesnt work

like if you buy a phone and you can't use any banking apps

the paradigm shift is the same as all tech today

What will end up happening if you don't allow users to upgrade in practice is they just won't

people won't buy a brand new computer because win10 won't upgrade to win11. they'll just use win10 after it's EOL

except full disk encryption is not a generic end user feature. people aren't going to be blindsided by that

passkeys are a generic end user feature

passkeys don't require tpm tho

the OS still can generate entropy and passkeys can store keys on disk

that goes against the point of a key

security devices like yubikey also don't need tpms fwiw

because they are the tpm

I don't think a yubikey is a platform module

and the point of them is so you dont store the keys on disk

Software passkey would probably encrypt everything with your login password

Which wouldn't even be that much less secure than windows with tpm and FDE

it makes it not a 2 factor

By that standard the most popular 2fa option, phones, aren't 2fa

yeah that's why most are moving to phone passkeys

unique to the device rather than number/account

My point is software passkeys gets you like half the benefits

Don't all the vendors have passkey cloud sync?

why would people want to attach authentication for their accounts to their phone‽‽ that's just asking to lose all your accounts

house key same reason

ideally you'd have two

but if the phone one syncs then 🤷

that's not good

looks like microsoft doesn't allow that

which is correct

I've heard so many horror stories (including personal ones from friends) of people attaching authentication for something to their phone and then their phone breaks or dies

and that's not even counting phones getting stolen

Google 700 links Yandex 1300 links

Apparently the Matrix bridge is down. Does anyone know if there's any plans to get it running again?

huh I was wondering why I wasn't seeing Matrix messages posted anymore

how long has it been down?

Looks like middle of January. Last message that I'm sure is from here (Discord) is Jon muting Fab, which was the 13th.

Pepper and salt. You can tell by the size of the container what kind of proportions we prefer.

Also: black pepper is the best pepper and this is not open for debate.

i like using multiple peppers at the same time

the other day I made a cream sauce with black pepper, white pepper, green pepper, and pink pepper

I like https://www.migros.ch/en/product/106206100000

pretty much the mixture mentioned by <@167023260574154752>

In Sweden, there is something called "five pepper sauce", that is those spices, plug allspice, which is called "spice pepper" in Swedish.

I wrote a python script for beats swapped and reversed and it's beats... inspired by https://www.youtube.com/watch?v=OyEa1DFM5gk and https://www.youtube.com/watch?v=09tzb8lkMwE

Omg, it is painful for me just to look at this photo

That is nice — all peppers are good (I think the pink one is not really pepper but it doesn't matter, it's also good). I'm just saying, if you have to pick a single one: black pepper.

mm

wait there's different kinds of pepper other than black pepper?

I just know of black pepper, and then crushed red pepper

White pepper is the default for many people where I live...

I don't visit enough restaurants to know for sure, but my feeling is that in a modal Belgian restaurant, the salt and pepper on the table (or that you get if you ask for it) is plain fine-grained salt and white pepper.

(often pre-grinded white pepper, which is even worse)

i like black pepper then white pepper

green is my pepper

It's not something I've tested yet. However, if you want to test it, you'd be best off testing with a Linux guest first.

weird ~~typical vbox i guess~~

VMware best

I use lts kernel for vm hosts

VMware works better than virtual box

Especially when using more than 4 VMS at once

Even for running single VM, VMware worked better. Virtual box crashed during heavy workload.

qemu supremacy

(I don't actually know what's best)

Qemu is also good.

I use qemu when I don't have VMware. It also handles more VMs and heavy read write loads.

I tried to use Qemu many times to launch non-x86 stuff, but it never works.

I’ve been thinking of moving away from VirtualBox for my Arch VM

then I could enable Hyper-V and switch to WSL2, I guess

(Arch on WSL2 could be an option but has the drawback that I would pretty much have to commit to it before being sure I can make it work)

I did arch on WSL2 by grabbing a container image

https://images.linuxcontainers.org/

i just used https://github.com/yuk7/ArchWSL

ArchWSL, Qemu, Hyper-V, Vmware workststion/player all work well together you can run all of them simultaneously with no issue, virtual box is the only one that gets caught up when you do

gpu perf is best with wsl2 and hyper-v (Hyper-V is a bit involved setup however) then vmware and Qemu are pretty tied. Cpu perf is mostly the same on all of them, they all use whpx for accel on windows. Disk speed is marginally fastest with qemu but not enough to matter

Wsl2 uses 9p for accessing the host data, which is better then the other VMMs that can't do it, but its slow as molasses so try to do it as little as possible

I don't think most WSL users are using gpu stuff tho

at the very least I never launched an X server on it

one would be surprised, it actually works quite well and I winded up using it a bit, it's also common for compute stuff too

seems like MS's general strategy with wsl is to give you linux dev convenience + windows hardware support

with stuff like directx + vaapi support within wsl

well directx is mainly used as a backend for vulkan and opengl support

that’s actually the reason why I’d be interested in switching to WSL2

so that I could try some JAX stuff with the CUDA backend

NumPyro (a library for Bayesian inference) uses JAX, and I’m interested in trying it out either directly or through PyMC

ya i use arch on wsl2 to use things like openai-whisper on CUDA, it works surprisingly well

I use whisper on my phone for "real time-ish" text to speech lmao, flatpak still doesn't support either rocm or oneapi, and doesn't support rusticl either. I actually made a blog post the other day about it xD

https://www.theregister.com/2024/02/09/240_4_ipv4_block_activism/ Instead of moving to ipv6, some people still like to milk ipv4.

I swear, we might end up adopting JXL before people fully accept IPv6

At least on that front, AWS charging for IPv4 addresses is pushing everyone to finally move

https://www.livescience.com/technology/electronics/universal-memory-breakthrough-replaces-ram-flash-next-generation-of-computers-major-speed-boost

TIL: Enable `Fastfox` user.js in Floorp (or other Firefox-based browsers), seems like it makes Firefox somewhat faster it's already integrated into Floorp, just needed to change one setting https://github.com/yokoffing/Betterfox

I wonder if they address cold boot attacks with the new tech

I assume they'd encrypt that partition with /dev/urandom or it's equivalents like you'd do for swap without hibernate

https://x.com/phoronix/status/1757041961330958521?s=20

https://www.phoronix.com/review/radeon-cuda-zluda

lol quietly funded

HIP is literally CUDA but with the function names renamed. They even have a perl script that can regex CUDA code to HIP code.

So making their driver work with the original CUDA names couldn't have been much of an effort...

er, wait

apps already compiled with CUDA? Ok, that's useful.

the funny part was the cuda version's still faster than the native hip/rocm

lol

still worse value than nvidia's but it's something

except the news comes from them stopping funding and it being abandoned

double lol

this would be neat if rocm wasn't a massive pita to use

https://www.reddit.com/r/jpegxl/s/VCJYR7dYn2 Is it fab?

seems plausible

Seems more like a standard bot to me

there is not enough comments for it to be fab lol

if it were fab it would have responded to itself with 'd0.8876 e3.322 gab=0' or something

https://arstechnica.com/gadgets/2024/02/mozilla-lays-off-60-people-wants-to-build-ai-into-firefox/

how long before firefox switch to blink ?

considering that they've been increasing investment in Firefox development and laid off people doing work unrelated to Firefox, probably never?

ah ? why didn't we read about this ? development of web engine ? or AI stuff ?

what

what ? 😄 you said they are increasing investments in development, but development of something useful ? and do you have some source for this ?

the way you worded it is why I said "what"

my source is I saw Firefox developers on mastodon talk about how they've been getting more funding

that's a nice source... I shall use mastodon too

People always hope for something when "new CEO" pops up. But things always either remain the same or get worse when there's "new CEO". How many times Disney got a "new CEO" at this point?

The problem remained unsolved, I've parsed 12k of links using Yandex images and 4k of links using Google images

in total, 10k unique links, after downloading each video, I parsed a link to the channel, and also downloaded each channel (in the process, but the result is about 300 thousand videos)

The matrix bridge has been dead for quite a while, will it not be back up?

bad spam has also been dead 🤔

What bad spam? That guy was from here

i mean evil spam

the article doesn't say they're unrelated to firefox does it

Yeah but ive only seen that fab guy doing spam and he was from discord

we had a major spam problem since matrix didn't have good spam filters

fab is fun spam

there were a lot of spambots that discord admins had no way to ban

without going to matrix and banning them

it was a real pain

fab isn't a spambot, he's a human who is a little bit spammy

Yeah i understand killing the bridge for the spambots but it just stopped working after fab spammed so it was a bit random

the power of the fab

we disabled before fab got worse

Im a bit curious how it could be worse

I think the bridge died at Sat Jan 13 13:07:07 2024 with this message

I thought the bridge had been down for months

see: https://discord.com/channels/794206087879852103/794206170445119489/1118879169607127111

Maybe it was janky, either way it was just the jxl room / channel

i still don't see the purpose of it

I didn't do anything to remove the bridge, nor to try to bring it back up

I dont think its that necessary but for the people who like it then i guess its nice. The moderation for it is not great even with the new bots and stuff. The GrapheneOS rooms are frequently spammed with very bad content, so if you arent going to have the resources to actively moderate then matrix is not a great place for a community. Another fun issue of matrix is that rooms just slowly brick themselves over time, the server just forgets users are in a room and they "leave" (not on their end, clients need to clear cache to figure that out, so for them it just fails to send and receive messages)

yeah where's the jxl mumble server

I think the bridge just naturally broke on it's own, I remember seeing in some other Discord servers that some Discord bots where breaking randomly because of internal changes happening somewhere

there where some people in here who actually used the bridge, I remember one person spotting a typo with chromium's libjxl implementation

https://twitter.com/OpenAI/status/1758192957386342435?t=veHEo9R4hFH5orR7zFeL3A

firefox will only heal when mozilla adopts linux' development model

where companies pay/hire their own employees to work on it, instead of being mozilla's burden

splitting development of the gecko engine from the firefox browser itself would make it so much more manageable pumping gecko-powered browsers and applications.

that could be cool

sorta like how we have electron apps?

gotta create a human centipede of AI

man, sinus infections suck

I made a cut

gecko is the worst part about firefox replace it with something new :D

Replace it with Tiir's JXL image of a gecko

[dogelol](https://cdn.discordapp.com/emojis/867794291652558888.webp?size=48&quality=lossless&name=dogelol)

The Gecko is my gecko btw

not tirr's

Ah, just seen it in Oxide tests so much I must've gotten mixed up haha

that's in part because a lot of weird jxls that test oxide were generated with libhydrium

my experimental encoder

whaaa, the nightmare sequence ! reality is so distorted in this first 20 sec

come back to reality <:PepeOK:805388754545934396> Neural just a dreams

Generated dreams <:H266_VVC:805858038014672896>

No idea about this. I saw this at printing shop.

https://youtu.be/iZiX1a7uYmg

https://www.youtube.com/watch?v=cM_sAxrAu7Q

https://arstechnica.com/tech-policy/2024/02/air-canada-must-honor-refund-policy-invented-by-airlines-chatbot/ 🤣

I came across the classic XOR texture in <#824000991891554375>, reminded me of the demo video of IBNIZ https://www.youtube.com/watch?v=aKMrBaXJvMs

not as constrained as jxl art, but very fun

https://twitter.com/nixcraft/status/1759303937478496543?t=ViFc38vPhcl4dOfsFyW1VA&s=19

genius

I did a fix to fix church images on fB

https://www.phoronix.com/news/Linux-Disabling-RNDIS-Attempt

A person has done a encoding

A photo of me leaked on the web

74% less psnr

I encode even for cringe content

But I think I have 5 ideaz left on my side

speaking of very hot and bright stuff, other than welding there's also glassblowing with pure quartz: https://www.youtube.com/watch?v=VT9cmyIB61c

much higher temps than regular glassblowing

> my family computer in the early 2000s had a second drive that was 4.5 GB and I remember thinking "there's no way I we'll ever fill that up" meanwhile, on my computer: ``` NAME PROPERTY VALUE cirno allocated 838G cirno free 330G cirno size 1.14T cloud allocated 23.8T cloud free 12.5T cloud size 36.4T data1 allocated 3.98T data1 free 579G data1 size 4.55T ```

I'm juggling data on a single 2TB SSD after vibration killed my hard drives and now I only have 70GB free

I was about to link you to [Compactor](https://github.com/Freaky/Compactor) or [CompactGUI](https://github.com/IridiumIO/CompactGUI) but it seems you already know. I will say though that compactGUI is out of beta now (I didn't know it was until someone told me recently)

I've been sticking to Compactor since it attempts to compress 4KB blocks of the file to decide if the entire thing is worth compressing. CompactGUI does it brute force and then sends the result to a database that has to be manually updated. Currently around 150GB over my drive capacity with it

https://github.com/Freaky/compresstimator

I just wish they added the threshold setting for the compressability

I left it in my case since it was dead anyway, can't be bothered to check but it was a Seagate Firecuda and a standard Seagate. Firecuda had random corruptions even with a clean SMART and sector scan, the normal one got the click of death when I plugged it in one day

Firecuda was roughly 3 years old and the normal drive only 1 and a half

Exactly

On a Samsung 860 EVO currently, at 98% writes left after 3 years so not bad

Although the motherboard decided not to detect any storage devices a few months ago so that was fun

The same, with exception that there are no HGST drives available here, WD\Seagate only. And according to Backblaze seagates have the worst failure rate

50 years today since the release of https://youtu.be/EhQpXD2Z9WQ

What discord ui is this

Oh yeah, I didn't even notice on my phone earlier

It looks like htc sense UI. Pre sense 5.

Drive sizes have stagnated. No one sells 100GB drives for more than a decade, but 1TB are still being sold for 15 years already in most PCs.

long gone are the days where we always needed more storage, bluntly, the vast majority of people do not need greater then 1tb. and those who do can just run multiple or "enterprise" drives

ofc gamers are the largest market for larger then 1tb drives for consumers. speed on the other hand is still a noticable increase in perf even for consumer workloads

People don't "need" anything, tho, and yet other tech is improving, while other is stagnating. I hoped HAMR would improve things, but so far it's enterprise only. Also, information is accumulative. If you need 1TB today, you'll need more later, because old information isn't gone.

oh we did need it though, Many times customers would come in complaining that they ran out if space. It was extremely common so advertising "Double the space" with a 250gb drive was a massive selling point. now the vast majority of people don't even saturate 3/4s the storage they have.

especially when digital cameras and "digital camcorders" were just becomming a thing, people would fill their drives in months

Smells like retroactive justification, IMHO. One single game is 100Gb nowadays. People would gladly buy more if 5TB drives had the price of 1TB drive. 1 TB drive would simply become extinct.

anyone who thinks gamers are a sufficiently large market to drive consumer hdds I would argue is the real cope. You can easily get multi TB ssds for sufficiently cheap, and still growing cheaper, but the default is 1tb because for the vast majority of PCs 1tb is sufficient. I can go to amazon and get a 2tb nvme for like 140cad, if I go with an ssd I can get it for like, 100cad. But that's it because the market to drive consumer storage is now mostly dead.

don't get me wrong, I have multiple 1tb drives in my PC and am looking to get another nvme, but I realize I'm not indicative of the larger picture, If you want larger storage, you need to innovate on larger storage solutions, but for the consumer market, there isn't a reason to innovate there anymore, so we just get the trickles from the enterprise

You don't get my point. If HDD continued to improve, it would have been economically unviable to sell 1 TB drives, because they wouldn't be any cheaper than higher capacity drives. That's why 100 GB drives are extinct.

all im saying is that there needs to be a reason to improve, that reason is long gone.

We'll see.

I mean, we have seen it, it's exactly as you noted, 1tb has been the standard storage size for an extremely long time now when you put it into comparison against other storage size periods. 60gb drives, 100gbs, 250gbs, etc. we did indeed stagnate at 1tb, and it's going to be a long time before we see 1tb starting to limit the average consumer. Phones are currently the "larger issue" but even then we have seen manufactures back off of 512gb devices infavour of 256gb devices

in the future we may see the 1tb average limiting the average user, but I don't really see that happening, gamers for sure, but the average desktop/laptop? I kinda doubt it

The other thing is that the cloud era coincides with the 1tb era. When streaming large files is the norm you don't need as much local

[yep](https://cdn.discordapp.com/emojis/721359241113370664.webp?size=48&quality=lossless&name=yep)

and ofc cloud storage in general for pictures/videos for long time storage, even if you had to download them, would be a tradeoff people would be willing to take

I looked at the place I buy computer stuff. They barely even have consumer HDDs, except exernal ones.

But the cheapest 2TB drive was cheaper than the cheapest 1TB drive 😄

at some point, HDD tech become good enough that a single platter is enough to satisfy most consumers. At that point, fancy improvements stop making sense. HAMR makes sense when you have crammed 10 platters into a single drive, and literally no other way to improve it, but anything 2TB or less may just be a single platter with CMR. First gen HAMR maybe you could get that up to 3TB. Not super interesting exactly.

aye

the store I looked at to compare prices only has 3 2.5" HDDs 😄

but consumer 2.5" HDDs are mostly for laptops, right? Those should all be replaced with SSDs anyway.

Ideally all SSDs, especially when at risk of drops or knocks

I just checked and the cheapest 1 TB SSD is cheaper than the cheapest 1 TB HDD lol

and some of the higher quality 1 TB SSDs are also around the same price as that 1 TB HDD

(those were all M.2, didn't check SATA yet)

I'm not looking at anything used

Since the discussion started the question of why PCs don't have large HDDs in them

I see that the store I'm looking at has a number of 2.5" portables available, so I guess they must be semi-popular. Internal ones tho, seems like a dead market where I live. All the mini PCs here probably use SSDs.

seems drives are cheaper where you live, than where I live 😄

The cheapests ones are all $80+ here

Not a single store though, it is aggregator. But there are more 3.5" positions than 2.5, and more internal positions than external

Never buy Seagate hdd. They are not good

https://youtube.com/shorts/G1QLf2o0GFM?si=nPe7laaCxC9gOE7g

The problem with SSD/NVME is that at the moment they're increasing capacity by reducing reliability, by moving to TLC->QLC->PLC flash configuration, and the information which drive uses which flash type/mode is often not outright available. I wanted to buy MLC SSD and they're no longer produced for general customer market.

to be fair they seemed to have stopped at tlc for non low-end

not necessairly, among the ones listed on https://www.johnnylucky.org/data-storage/ssd-database.html there are QLCs and even one PLC.

There were some promising announcements of new nv memory types - faster, more endure, requiring lower power to write, but..

TLC is way more resilient than it used to be.

btw, QLC flash can be run in MLC or SLC mode, that's how you get a gigantic SLC cache on QLC drives

which is needed, because early QLC had very slow write speeds

ye, I'd love a disk which you can low-level format according to your reliability/capacity requirements to whatever mode you desire

Due to the wear leveling feature, I wonder if that actually does anything for longevity, and doesn't only improve speed...

Higher capacity means more writes.

Using QLC cells as MLC would drastically reduce the chance of bit rot or single read errors as well, though...

But that's already extremely low.

the capacity seems to grow lineary (1, 2, 3, 4... bits / cell) but reliability worsens faster (according to "classic" data SLC ~50k cycles , MLC ~10k cycles, TLC ~3k cycles, QLC ~2k cycles) so for some applications eg. storing video monitoring records, where you overwrite whole available space multiple times anyway the longevity may be better. Economic cost may not be better tho, as capacity prices are going down over time, so it may be better to just plan ahead more frequent drive rotation. The configurable type would allow to find optimum tho.

but that data does not represent the virtual SLC cells...

just the overly robust actual SLC cells

doesn't capacity grow exponentially here?

I don't think so, unless I'm mistaking something. 2 vs 4 distinct levels just maps to 1 vs 2 bits of stored data.

isn't it 2¹ vs 2² states per cell?

and 2³ for TCL?

<https://en.wikipedia.org/wiki/Multi-level_cell> says it straightforward: ``` Single-level cell or SLC (1 bit per cell) Multi-level cell or MLC (2 bits per cell), alternatively double-level cell or DLC Triple-level cell or TLC (3 bits per cell) or 3-Bit MLC Quad-level cell or QLC (4 bits per cell) Penta-level cell or PLC (5 bits per cell) – currently in development[1] ```

2^n states maps to n bits

so you're correct, but that results in linear capacity gains

yes, that makes sense...

qlc reserved for low end https://docs.google.com/spreadsheets/u/0/d/1B27_j9NDPU3cNlj2HKcrfpJKHkOf-Oi1DbuuQva2gT4/htmlview

syncing the notes in an audio track with the blinking on a modular synthesiser in a video track is not as easy as it seems

I’m not even 100% sure I’m not off by one note :s

… turns out I am

(reference: https://youtu.be/UzQKoGgJmJA?t=52m20s)

wait, no

meh

https://www.whitehouse.gov/oncd/briefing-room/2024/02/26/press-release-technical-report/

Joe Biden wants you to use Rust instead of C or C++

Interesting for the people here

How the ai is uzed on Social

https://www.instagram.com/p/C3PEM4zr8kn/?igsh=aG5mb2tha3BkYjY5

https://www.instagram.com/p/C3aRDlpKeja/?igsh=dzgzc21yaWJkdHY5

You won't believe they make this to your image

Yes most ai does not preserve jpg Bitdepth

To avoid retcouchong it

Of course JPEG is work on newer AI

I cannot post memes aboutAI here, but there is that type of sort

Ai is a bit stupid, cause i'm alzo as Gemini it is

Before it was more intelligent, but Gemini improved its Italian by huge margins

Now is finally look good on some INNA videoclips

But ai is ai so we need SVT av1 2.0.0 with tab settings

So Jon and other membrrs can agree or DisGree

The font in Discord is pretty small for me and I have perfect vision

Can lonnie send a Feedback to Discord.app

https://www.youtube.com/watch?v=8zi7P8c15ao

https://www.youtube.com/watch?v=PW6QWqErcIk

in week, I wrote an algorithm for shaking the screen based on the parameter of the decomposed components of the sound wave of music (drums, bass, other). shared: https://cdn.discordapp.com/attachments/1149722944537383002/1212353405234970624/fnaf2_shake.mp4?ex=65f18718&is=65df1218&hm=e777538061f67c647a0685ed37699b8ea98285d4ab6bfa954028a851bdba40d0&

My surprise is that even though I didn't invent it, I have a feeling that no one invented it. You just won't be able to find tools on the Internet that would do something like that. Everywhere there is either a manual arrangement of bits, or very old tools. For example, in FL studio there is a "ZGamer audio-shake" that does this effect, I tried it, it is very, very primitive, with a complete lack of settings and anything else. Basically, my friend and I laughed a lot when we found "audio-shake" in the top answers to the query -- solutions that are already 30 years old

pretty cool stuff!

just 90% chatgpt python shitcode, but at least it's "fast"

Cool

License?

shitcode license

I hope someone steals everything I wrote and makes the tool better

for the last couple of days, all I've been doing is rearranging parts of the code so that it can be used without edits inside the code. release soon.

Space Chair cat meme REMASTER

Today I encoded some Drake images

As it was suggested in the ex av01 server

The frame without audio frequency flat optimization

I'm suite satisfied with the result

Now i will try the jpeg xl codec for my persoalimage

here you go ``` The Shitcode License version 0.4 Copyright (c) [year], [full name] Redistribution and modification in source form is encouraged, please help make the software less trash. Redistribution or use in binary form is permitted, however not encouraged. Seriously, look at the code and tell me if you want to use it in production. Redistribution of this software, in source code or binary form, including any derivatives, is only permitted provided the distributor / user retains the above copyright notice, as well as the rest of the contents of this License. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” AND ANY EXPRESS OR IMPLIED WARRANTIES. DID YOU REALLY EXPECT ANY??? THE SOFTWARE IS NOT FIT FOR ANY PURPOSE OR USAGE. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON... ACTUALLY I DON'T EVEN CARE ANYMORE, I WANT TO DROP OUT OF LAW SCHOOL :( ```

okay no more editing im done with this silly joke

https://directory.fsf.org/wiki/License:WTFPL

lmao

WTFPL notably has no warranty waiver which makes it sorta unsuitable

warranty waiver is important to work around american consumer protection laws that were written before software was a thing

don’t forget about https://github.com/me-shaon/GLWTPL

can you submit that to kdenlive or blender?

what do you mean by "submit"

like make it an effect in those

<@1207050979267313719> main self wtire python. if you could find a good programmer, I would explain to him how it works. so that he would write a separate library that could be connected to different programs. To be honest, I understand that everyone wants this to be a part of something. but I don't want that, I like it when such a separate software exists separately.... by the way, I use demucs to decompose sound into components, and you can't just put it in a small extension.

Put your working code on GitHub or gitlab etc

mehhh i have not github

Bitbucket 👌

I have few connection, everywhere tj re is this song @+ Mahmood - Tuta Gold

The air pollution is fair but the 28th feb

Indoor 26NO2 every day of the day

In altarello 9,1a day not everyday

A noegjbour with, autistic problems was got recovered in hospital in the 28th fenruary

so much troubleshooting as to why running cjxl in a Python subprocess was forcing the output of the subprocess to go *after* Python print statements... (only when running through fd or parallel) turns out: - GNU parallel (and fd as well) output stdout first, then stderr - Python's buffering is wild sometimes solution: - redirect subprocess stderr to stdout, and - use `print(msg, flush=True)` for every print statement or run `python -u` for unbuffered stdout

from: https://discord.com/channels/794206087879852103/803574970180829194/1213506528414535710 huh...

why not just flush stderr instead?

I. Thi#i thugbh stereo was t bigddf

stderr still comes after stdout when using GNU parallel / fd

since you are using python, you can use multiprocessing

but for sigint you would need to use Pool.apply_async

or maybe that was to stop the encoding if one image fails?

probably that, since apply_async takes error_callback

Pool.map is like 2 lines of code

Where is the software that removes blur in pixel art? Does anyone know about this??

You can do it with imagemagick (in simple cases like this) `convert source.in -sample [original pixel art size] target.out`

Which pixels will be taken for scaling? all? or only those that are in the geometrically correct point of the pixel grid? original pixel art size - how should I know? count all pixels manually? in height and length?

You might need some offset, not sure if imagemagick automatically uses the centre of the pixel

okay. the first one is solved. What about the second one?

I dont know of a way to figure out the original size automatically. Its possible from a technical perspective, but I doubt there is a tool for it

🙄 I will have to program again the software for this. Computers have been around for 1000 years why hasn't it been invented

To add an offset you could specific a custom filter operator, but its probably easier to just add a border or crop the original image

If you can guarantee the image is scaled up significantly like in the image you've shown, you can just count how many chains of duplicates there are in the first row and column

Otherwise compute the 2nd derivative and check for spikes, but that only works on certain interpolation modes

manually count 👍 too fucking lazy

> "Oh, just use neural networks to pixilate"

for the blind, it will do.

I invented algorithm. Igoing to implement .

try point downscale then point upscale should work I think...

you could run some sort of convolution and check for spikes, right? probably by multiplying in fourier space

Yes, you'd want to use some DFT. If you know what blurring kernel was applied originally you could re-blur using that wavelet and calculate the difference, but since you dont know the size of the blur kernel thats not really possible.

Quick search later and there is actually a tool: https://github.com/0x09/resdet

🧐

almost... but you wrote it somehow *difficult*

This is the simplest and most elementary thing

3,16796875 slightly by accident, I found a period for one art. but I can't figure out how to make the code for this. hmm. strange.

I came to the method of bruteforce of image resolution values and the subsequent creation of a matrix of pixel color differences. I still have a lot of work to do on this. I will continue the day after tomorrow I couldn't figure out how to find the periodicities in the values. if the periodicity is a flatout number... it is necessary to draw a vector line graph of values or something like that... skill issues

811 original art pixels ÷ 3,16796875 = 256 true pixels

getnative.py is a python script that is quite good at this

it downscales with the inverse kernel and them re-upscales

which only roundtrips if thry guessed right

If you think about it, Adobe Stock is the real Adobe Photo-shop

Interesting! You have many pictures like this?

I think I have an idea but would need a few files to test something 😁

Download any pixel art and rescale at random size factor (like original size × 4.75318)

certainly not efficient but a simple dst works

scaling to the number of pixels between the top of the dst and the first black line gives me a perfectly pixelated artwork

caveman way to solve it but ehh 🤷‍♂️

But why 😅

if your only doing it for a few images its simple enough and certainly faster than guessing

I was wondering why the original size was found as the first black line... I understand fft but this feels like magic

because that's the frequency of new information

After 2 days of thinking, I was able to **invented **with a brilliant thing

do scale NEAREST scale BILINEAR calculate difference

At the moment when the image will match exactly in pixels, there will be a minimal difference

adding all the methods works even better. But it's too slow.

just use getnative

😳

<:FeelsReadingMan:808827102278451241>

Wow, that's literally what I invented.

or current vspreview

no-no. what a bullshit dump

I did not understand why it is necessary to have such a large amount of software for implementation. I use only one pillow

pillow is huge

☝️ one and easyinstall

pip install vsjet

<a:trollform:771932238748712979>

no

when I suggest getnative.py it gets ignored <:sadge:916886322029936680>

context: continuing from https://discord.com/channels/794206087879852103/803645746661425173/1215098141322256445

does anyone remember what #include the malloc is from

stdlib

but also, do note that if you try to malloc 10G

it will take a while

one reason to have swap is that you can commit a heck of a lot more than your RAM implies, without it being problematic

because linux will shuffle around its address space a lot in order to find a contiguous portion of ram of that size

assuming you actually have that much free

so to actually test it may be a long time

apparently you can get OOM killed on macOS, but only when the disk just straight up runs out of space for more swap

at least just allocating the buffers is really fast

``` allocating done allocating checking buffers 0: 0x120000000 128: 0x8087fd0000 256: 0x9e557d0000 384: 0xbc22fd0000 512: 0xd9f07d0000 640: 0xf7bdfd0000 768: 0x1158b7d0000 896: 0x13358fd0000 ________________________________________________________ Executed in 4.09 millis fish external usr time 0.70 millis 83.00 micros 0.62 millis sys time 2.63 millis 472.00 micros 2.16 millis ```

does it dynamically grow the swap file?

that's without writing to them though

btw im on `Darwin <host> 23.2.0 Darwin Kernel Version 23.2.0: Wed Nov 15 21:54:05 PST 2023; root:xnu-10002.61.3~2/RELEASE_ARM64_T6031 arm64`

yeah

oops doxxed myself :/

lol i forgot about compressed memory i wrote 223GB of data before the process got killed

and it only used around 250MB of real ram

lol

why would it need to find a contiguous block of memory? it's a virtual address space isn't it

haha

`a.out` using 186 GB

but actually using 4gb

I think this is right... and I think if you do such a large allocation it will do something more like an mmap internally

I would think it would but ig not

it will take some time before segfault though

when i ran it it SEGKILL at 223GB (not segfault for some reason)

probably just hard-coded limit not actual outofresources

howd you compile it

I just used gcc -o test test.c

ah I need to catch up on these older messages

yeah I suspect it's doing something like mmap for your allocations and only making the pages resident when you start writing the data. that's why you only oom when writing the data

sigkill would be expected for oom I think, not segfault

Yes

clang++ oomtest.cpp i also switched to 1gb allocations