programming

[rough bloom]

they only switched the naming scheme quite recently though
the last desktop CPU using it was 2023 I think

[silent cloak]

surprising

[silent cloak]

if i go on their site i still see i5, i7, and i9 for all their chips

which just makes them seem old

[true hemlock]

elvyn had fun (definitely ) working as my consultant lmfao

[rough bloom]

yeah, so if they're still around and very common, they aren't old?

[true hemlock]

huh, i personally don't view it that way

and this is the first time i seen someone view i naming scheme "look old"

[silent cloak]

again they associate the chip itself with the names i5, i7, and i9

they dont know that there are differences in the generations

[rough bloom]

if anything their new naming scheme is much less known

[silent cloak]

the i5 still looks like a decade old if ur not tech literate

[true hemlock]

at the same time, they don't care and think that i7/i9 will always be better no matter what

[silent cloak]

they should put years on the naming scheme

so u know its actually a newer generation

[true hemlock]

we already have that...

its called as

the sku model number

its always displayed everywhere

just that

[silent cloak]

the average consumer wont know that

[rough bloom]

this seems correct to me (since Intel markets the "tier" of the CPU very heavily, AMD does this too but to a much lesser degree)

[true hemlock]

illiterate people doesn't understand it

they don't look it as old

[silent cloak]

exactly

but they are the mainstream consumers

[true hemlock]

they only had one thing in mind

[silent cloak]

outside of big tech

[true hemlock]

"i7/i9 is better"

[silent cloak]

exactly

thats my point though

anyone who has been gaming for years wont want to "buy another i7"

[true hemlock]

except for the "look old" part because that's practically never happened to any of my client...

[silent cloak]

they will look for a bigger number

[true hemlock]

actually i need @umbral thorn to confirm that

but i consulted at least 3/5 of all the people

still pretty much a hundred and more

[rough bloom]

I don't think people see i3s and i5s as "old"
just "worse" (which is different but results in effectively the same thing)

[true hemlock]

yeah

exactly that

fun little fact, i3 is somewhat pretty good for 4K GPU intense gaming

[silent cloak]

i see them as old tbh

mostly because i pull them out of old computers all the time

though not as old as a duo

[true hemlock]

the thing is that if people saw them old, wouldn't that mean that people would at least understand the generation segments since back then

[obsidian mantle]

i felt comforbable with their "iX + number" namings

[true hemlock]

yeah core ultra naming is shit

[obsidian mantle]

so i3-5-7-9 is like general grade
then number shows generation

[silent cloak]

if they throw a new number on the iX then market that i can see them having a comeback

the i9 has bad rep

[obsidian mantle]

so when i bought i3 10100 i knew its i3 but its current most modern generation

[true hemlock]

have you seen XE lineups

[silent cloak]

no

[true hemlock]

i9 10980XE as example

[obsidian mantle]

unlike amd with their AMD RX31237129312381238189 names

(maybe im just clueless)

[true hemlock]

this is basically intel version of threadripper btw

[silent cloak]

threadripper is a more recognizable name tbh

[obsidian mantle]

amd like

[true hemlock]

i9 10980XE has like 18 cores

[obsidian mantle]

7800 from 2011 and 7800 from 2028

or smth

[silent cloak]

and the fact they still have i9 slapped onto it

[obsidian mantle]

guess which one is which

[silent cloak]

intels big issue is marketing rn

[true hemlock]

intel's naming scheme is confusing even for prosumer and enterprise grade

lol

[silent cloak]

and not having a big enough release for people to really notice them

[true hemlock]

nah

their marketing is working. people are fooled all the time

[silent cloak]

yeah mostly corpos

ur average gaming pc builder is gonna use AMD though now

its been like that for a while

[true hemlock]

they still make profit well btw, just that the people that knows intel is bad is usually the loud ones.

and uh

market stock shit

[silent cloak]

its more just people stopped looking at them

they arent considered as often by ur average user

[true hemlock]

oh I WISH my clients are THAT smart at least

[silent cloak]

AMD is being talked about more

[obsidian mantle]

like wtf

[silent cloak]

again they need something to bring them back

[obsidian mantle]

zen 2 zen 4 both have 7000

i see 7000 something and i think

is it from 2025 or 2005

[silent cloak]

im all for competition in any industry

[true hemlock]

epyc is datacenter/enterprise

[rough bloom]

their laptop CPU names are stupid
otherwise it's pretty sane tbh

can't really confuse the Epycs with the normal consumer CPUs

(and within the Epyc lineup the naming scheme is sane, just different)

[opaque sigil]

ditch the "their"

[sage crag]

a lot of things i see say amd is around 30% marketshare, so intel is still leading in total. not sure what percentage of that is desktop or enterprise tho

[true hemlock]

their laptop CPU naming is straight up lunatic and you can NEVER change my mind about it.

[obsidian mantle]

yeah but there is non-epyc 700 in zen3

so zen3 zen4 both have 7000

[opaque sigil]

i think that was mostly enterprise

[silent cloak]

likely enterprise through work laptops

[true hemlock]

APUs aren't relevant either

[silent cloak]

companies love to overspend

[dapper kernel]

I'd like some help with this task. To summarize it — I'm given an 8x8 grid in U3.txt that contains the following: 0 — empty space, K — King, Z — Knight, U — occupied space. I need to write a code, to make the knight capture the king and then return to his starting space, while avoiding U spaces, in few moves as possible. Result is show in Rezultatas

[obsidian mantle]

are chess tournament applications still open

[opaque sigil]

shiro's been busy iirc so probably

[obsidian mantle]

so knight can go +2x +1y or +2x -1y

somethhing something

[dapper kernel]

yep

[rough bloom]

technically the deadline was a while ago
but it seems that the form is still open

may not be counted though since the extra time is an advantage

[obsidian mantle]

i just mentioned it because i saw some chess related exercise

[silent cloak]

10 reps of moving ur pawn

[true hemlock]

@rough bloom 7940HS vs 7940HX

do you know what's the difference

[true hemlock]

[opaque sigil]

second one probably has like triple the tdp right

[rough bloom]

no
I don't know or care about either of those

[true hemlock]

one of them has 8 cores

the other has 16 cores

[opaque sigil]

[true hemlock]

same exact number model

[rough bloom]

[true hemlock]

just one difference in letter

both are "ryzen 9"

fucking lunatics, amd

thanks.

[umbral thorn]

"want i9"

[true hemlock]

[rough bloom]

meanwhile Intel

[umbral thorn]

"for gaming"

[rough bloom]

they all suck

[umbral thorn]

"best cpu"

[opaque sigil]

intel is arguably even worse cause they have a hundred skus of the same embedded processor too

[obsidian mantle]

idk i see i9 i know its fast

[opaque sigil]

me when i9 9990

[true hemlock]

intel's objectively worse

[rough bloom]

they do a similar thing to AMD where their naming convention technically (with some exceptions) makes sense
like, if you know that the -H suffix is anything high-performance and -U is for shitty office laptops then everything is fine
but just use the same convention as desktop CPUs, why make them look similar but be completely different

[true hemlock]

Core Ultra X9 388V
vs
Core Ultra 5 125H

[true hemlock]

guess what, BOTH has 4 p cores AND 8 e cores. just that 125H has 2 lp cores and 388V has 4 lp cores

we're going backwards woohoo

[rough bloom]

wtf is "X9"

[true hemlock]

i don't fucking know

[opaque sigil]

i remember when amd changed their naming scheme they handed out this weird contraption to decipher them

[rough bloom]

oh yeah, the disc decoding thing

[opaque sigil]

[true hemlock]

i want that

[opaque sigil]

there's so many skus you need for mobile chips so it kind of makes sense, unless you want to go the monitor model number route

[fast pagoda]

the trap of alibaba

[true hemlock]

why do they want so many skus though 😭

[opaque axle]

OLED monitor users:
Which of these 27" monitors would you recommend?
AOC Agon 240hz about 530USD where I live.
Philips Evnia 360Hz Also ~530USD.

[rough bloom]

I guess?
should still do something to better differentiate them from desktop CPUs (and at least have it be consistent...)

[rough bloom]

many different laptops with many different requirements

huge market

[true hemlock]

speaking about cut downs

shuni have you seen the B300

[rough bloom]

I think so?

[true hemlock]

cutdowns on fp64

now they have the consumer 1:64 for fp64 lol

[rough bloom]

ye makes sense

[true hemlock]

its pretty shit

considering how many fields still relied on fp64

[young plover]

like what

[opaque sigil]

if you don't do AI you don't exist

[true hemlock]

major scientific simulations

[fast pagoda]

philips has better colour coverage and higher refresh, more ports, seems slightly newer, have crappy speakers

[true hemlock]

especially physics related

[fast pagoda]

both have speakers*

[rough bloom]

not AI
the B300 is an "AI factory"

no other use cases allowed

[true hemlock]

[fast pagoda]

are you locked on that one $300 us seems steep to me

i think that's about 530 aud right

[opaque sigil]

what's the name of the intel accelerator again

[true hemlock]

erm

[opaque axle]

where i'm at these go for 800aud, cheapest 240 / 360hz OLEDs on the market rn

[sick owl]

Got Gemini 3 Pro preview access in Gemin CLI

[fast pagoda]

was unironic wild gueess i guess ill look it up

[true hemlock]

he did state its usd

[fast pagoda]

i was p close

[true hemlock]

530 usd

[fast pagoda]

oh

[opaque axle]

5000 SEK* i just converted to dollars

[fast pagoda]

he said in usd

im dum

i see what you mean now

but also

530 usd is even steeper

[rough bloom]

I wanted to see if they have a different series for general HPC workloads now, but

HPC

[opaque sigil]

yeah nah

[rough bloom]

abandoned

[opaque axle]

yeah, it doesn't seem like that crazy of a discount on them. About 15% off MSRP i think

[true hemlock]

gone

reduced to

idk

[trim valve]

no more hpc

only ai

ai video generate the weather maps

[rough bloom]

✅ we have a model for that now
WeatherNext

[true hemlock]

also "more gpu better" ahh comparison

[sick owl]

I know that one

[rough bloom]

no need for accurate simulation

[true hemlock]

just pure

blackbox uncertainty/probabilistic

computation

now

at this point

just API call chatgpt everytime the GPU wanted to solve a SIMD task

"chatgpt what is [insert IEEE 754] [insert IEEE 754] [insert IEEE 754] [insert IEEE 754] times [insert IEEE 754] [insert IEEE 754] [insert IEEE 754] [insert IEEE 754] plus [insert IEEE 754] [insert IEEE 754] [insert IEEE 754] [insert IEEE 754]"

[opaque sigil]

finally access

[sick owl]

I mean they are turing complete

[sick owl]

:HOLY:

[fast pagoda]

ive looked at a few and the philips is holdiing up pretty well

[true hemlock]

until they spew hallucinations

[sick owl]

Is that local hardware?

[fast pagoda]

if he has that local that's crazy

[opaque axle]

It's got really chonky bezels, kinda eww

[sick owl]

That's why you use majority voting on your OpenAI API powered computer

[sick owl]

My thoughts exactly lmao

[true hemlock]

oh absolutely

"guys which instance of chatgpt did the SIMD correct"

[opaque sigil]

campus server so not quite local local

i wish

[sick owl]

Okay that's still mental

Where's my campus 480GB vram AI server

[fast pagoda]

renting b200s and stuff is fun because if you mess something up it costs you large dollars in real time

i wish i had free access

[opaque sigil]

can be yours for the small sum of $40k or however much it costs

[fast pagoda]

i should just take out a personal loan real quick and ruin my life

[true hemlock]

i just paid my research server instalments i don't have 40K to spare

[fast pagoda]

loan against my 401k buy in now

investing

[opaque sigil]

[olive sable]

do we have 401K in europe?

[fast pagoda]

pensions might be still a thing there idk

[olive sable]

oh ye

we have pesnions

[fast pagoda]

some places here do

a very small amount now

unless ur in gubbermint work or high up

[olive sable]

However, there are two other ways to save for your retirement:

    Pension saving plans ("pensioensparen" or "épargne-pension"). This is a way to save for your pension outside of your employer, and it comes with a tax break. There are limitations though, as we'll see later.
    Investing in accumulating stock ETFs. This has benefits in Belgium as you are only taxed 10% on capital gains. There are several ways to do this.

we love investing i guess

[fast pagoda]

that's whatr a 401k is

basically

[opaque sigil]

i forgot how ancient the software in ubuntu is

[fast pagoda]

what release are you on

24.04 lts? it's over if so

[olive sable]

then why do they say ther eis no equivalent?

[opaque sigil]

24.04 yeah

[fast pagoda]

hallucination i guess

[olive sable]

Investments are made post-tax ah i see

[fast pagoda]

401k is before tax

[olive sable]

ye

[fast pagoda]

so you dont pay tax prior to the investment

[olive sable]

i think a 401K is better technically?

[fast pagoda]

it looks like it might work out to be similarish but depending on the amounts you can put in i guess and how much those rates of tax actually are

like what is "low" annual limit lmao

[olive sable]

about 2.5K

[fast pagoda]

oh

that is low

[olive sable]

30% tax reduction at that amount

[fast pagoda]

oh gdi i forget cant talk about retirment investments because of the abbreviation of murican ones

[umbral thorn]

[fast pagoda]

anyways for the non employer plans it's a max of 70k put in per year including the employer match

and the roth/traditional "individual retirement account" have a combined contribution limit of 7000 or 8000, if you're trying to catch up to the max you wouldve contributed if you had done the max every year

traditional you pay tax now

but then none when you pull it out

roth you pay tax when you pull it out

err

other way around

lol

dont come to me for retirement advise folks

but yeah the limits are high i guess idk how much i actually put in other than it's 10% and my employer matches that

and then the world will end and it'll not matter

[umbral thorn]

taxes

[fast pagoda]

i know right

[umbral thorn]

[fast pagoda]

me when the gubbermint takes me money

[umbral thorn]

https://cdn.discordapp.com/attachments/1359530258344185876/1415261604655923220/DOWNHORRENDOUS.webp

[hard raptor]

meow

[opaque sigil]

i love copying all my project dependencies via scp because the server has no access to the outside world

[hard raptor]

So.. funny little thing.. i remade that little vector class thingy i had before i accidentally nuked my previous windows install..
And i also made a little driver based Key-Value store.
And this is a practical example of those 2 things working together

the reason why i had to do window.location.reload() is because of some silly bug i introduced by working around restrictions Svelte and/or vite introduced with the way one of them handles CSS

the only features missing from this Vec implementation (because i copied a slightly older version i send the code of in this chat a while ago) is the alphabet overflow prevention (by letting it overflow into a new char that gets added before the current one) and the ability to pre-set the type of vector it needs to be
(xy, xyz, xyzw, etc.)

also these being so far that i already functionally made them the fIdx variant and made them unsafe since well.. you're setting them by their index.. but not as far that i added the Idx suffix to them and their non Idx counterparts not being implemented yet

[clear sedge]

is this typescr*pt

[opaque sigil]

i may have forgotten something about SIMD

[true hemlock]

@silent cloak an "i9" btw

5 GHz, but look at that power draw

even though its "idle", its still running my basic browsing and discord task at that wattage

how come intel's laptop CPUs do worse than this...

[hard raptor]

Yeah.. i basically do dev on most thing (web, embedded, desktop, mobile, etc.).
TS is just really readable for me so if i work on something i would wanna end up using on multiple platforms then i write the initial version in TS since it is easier for my brain to port TS code to like any other language

[clear sedge]

[silent cloak]

does intel have any arm64 chips

[true hemlock]

that is in fact

not arm64

[silent cloak]

ik they are generally all x86

but im curious to if they ever made any

[true hemlock]

arm efficiency is misleading anyway

[silent cloak]

i know they both have their pros and cons depending on usage

[true hemlock]

less of actually being the "generally faster/more efficient pipeline" and more of its apple taking all the smallest process node, built whole ecosystem based on their arm implementation pipeline, and practically get away with lower clocks and hence lower overall voltage and current for still snappy performance
and x86 are generally being tortured to its limit because why not

i've made a whole rant about the misleading conceptions about x86 vs arm

and chip efficiency isn't an absolute measurement either

when you can always tune voltage curve and limits

(its highly non linear, that's why)

[silent cloak]

u see arm more in embedded and mobile systems

[true hemlock]

yeah

because simpler pipeline

you tend to not care about brute power

[silent cloak]

u get more out of the wattage but its not over the top performance

[true hemlock]

so embedded and mobile systems are designed accordingly to such pipeline

buuuuuuuut

my 13900 es would beg to differ here

im practically running it on M chip level of idle power draw for basic tasks like browsing, youtube and discord

[silent cloak]

i think both architectures are gonna be around for a while given their uses in their own areas

unless we get a best of both worlds situation

[true hemlock]

both won't be gone

[silent cloak]

i do like working with x86 ASM

mostly because its more documented

[true hemlock]

everytime people say "x86 will die" just never worked on computer systems ig

[true hemlock]

a hybrid would be nice

but also a pain in the ass to implement

so I/O need to be standardized

[silent cloak]

im thankful everyday to compiler developers

[true hemlock]

instruction decoding aswell

[silent cloak]

especially anytime i have to write SIMD manually

[true hemlock]

HAHAHAHA

NOW YOU KNOW MY PAIN

[silent cloak]

dont get me wrong compilers do a good job but sometimes they dont do what i need to depending on the configuration / compiler being used

then i have to pull out SIMD myself

especially a lot of recent rendering engine work ive been doing

[true hemlock]

some of my friend got so sceptical when they saw my i9 running at such low wattage they think its an M chip renamed with bunch of custom hardware lmao

[silent cloak]

lmao

[true hemlock]

its really simple, really

its just

tuning

well

professional tuning

[silent cloak]

thats because the i9 was famous for its initial release with the high ass wattage

[true hemlock]

but still

i proved my point

that intel just suck ass with tuning it themself

[silent cloak]

thats because vendors were so used to having to clock them a ton

and intel provided their recommended settings when they should of called them "required"

so it wasnt pulling in 250+ lmao

[true hemlock]

mind you

im doing the "i9 being as efficient as M chips" on an engineering sample aswell

[hard raptor]

so i now implemented the label based operations

[true hemlock]

this is getting worse 😭

to how intel managed their shit

like

i don't even ask for a chip to be drawing below 10W

[silent cloak]

tbh im thankful that most of my low level work is just GPU related

or assembly level instructions for CPU

[fast pagoda]

intell made me unable to change bclk

rude

[silent cloak]

i dont think i could truly get into the hardware level

[true hemlock]

im just asking them to tune the damn curve so their chips doesn't pull 400W unnecessarily

[silent cloak]

its their built in chip self destruct system

[hard raptor]

imagine Intel instead of AMD L

[true hemlock]

i was so pissed when they did this

i had to

manually do some microcode shit

to do overclock on my es

7.9GHz though

[fast pagoda]

maybe they were trying to stop you from hitting 8

8.8ghz

go back in time

[true hemlock]

im ngl

im tempted

to delid this thing

do liquid metal contact on full copper fin container

then LN2

to hit 8.8GHz

i figured that heat density is my bottleneck rn

[fast pagoda]

some guy recently did the air conditioner/fridge thing and hit some crazy speed on maybe an amd chip cant remember

[true hemlock]

this chip CAN def push more at 1.65v

[hard raptor]

@fast pagoda i blame you for the label based set methods only working half the time

[rigid snow]

what is that

why

[fast pagoda]

it was this guy getting https://www.tomshardware.com/pc-components/overclocking/gpu-overclocker-uses-car-coolant-and-pond-pump-to-cool-intel-arc-b580-achieves-17c-temperature-16-percent-performance-uplift-and-gpu-benchmark-record his b580 to -17c with antifreeze circulating with a pond pump that i was thinking of lmao

not cold enough for anything too wild

very cold though considering

[fast pagoda]

it's probably my fault you are correct

[true hemlock]

iirc at -235C quantum shit happens and can kill your chip randomly

[hard raptor]

Svelte probably either dislikes the idxMap method or the findMap function

[fast pagoda]

you cant get that cold with ln2 anyways lol

it's like -196

unless it's majorly changed most chips get pretty likely to cold bug towards -150

[hard raptor]

quantum shit is actually more likely to happen the hotter it gets... quantum computers usually are made that cold to reduce quantum noise

[opaque sigil]

does liquid helium get that cold

[fast pagoda]

it should

[true hemlock]

that is both not how it works and not even relevant

[opaque sigil]

Oh it does, -268

[hard raptor]

The fuck you mean? i went to school for this shit

[opaque sigil]

cool

[rigid snow]

[hard raptor]

we actually know absolute zero (aka -273.15C) specifically because we can't get helium any colder

[true hemlock]

im not talking about the influence of quantum noise to a device specifically ran on quantum mechanics

im talking about straight physical quantum effects on non quantum devices

[fast pagoda]

well before anything -200 or bellow our poor magic rocks cant handle it without bugging out anyways

[hard raptor]

Okay?

i fail to see how it's irrelevant

[true hemlock]

its irrelevant because quantum noise affects the state of a supposedly devices running on quantum mechanics

but is NOT the issue with altering physical properties of electrical components to the point of damaging it

[fast pagoda]

https://www.youtube.com/watch?v=UtXM71tw5fk der8auer using liquid helium

while back

[hard raptor]

Quantum chips and non quantum chips are made out of the same stuff.. anything that applies to quantum CPUs in terms of "quantum shit" happening also applies to non quantum CPUs

[hard raptor]

Brother... you're the one that made that claim

[rigid snow]

[true hemlock]

something that alters quantum state VS something that straight up alter something physically to damage it

different things

one is best to work at extreme low temps

the other... will break

[hard raptor]

THEY'RE MADE OUT OF THE SAME STUFF

JUST DON'T DROP YOUR CHIPS IF THEY COLD AND THEY BE FINE

[true hemlock]

and im talking about how the 2 alteration are straight up different?

also no

even though quantum computers are silicon, its usually only silicon with phosphorus

you get more complicated variants on classical computing

[hard raptor]

Oh my fucking god

[true hemlock]

how the fuck would you drop your chip when its anchored to a whole system for extreme overclocking

[hard raptor]

do you even do your own research?

[true hemlock]

yes, im one of those unfortunate people

who had an affected chip

due to shit happening at -235C

and im not alone

had other overclocking friends experiencing the same shit

[hard raptor]

for what are you even trying to get your chips to go that low lmao

[true hemlock]

minimizing the effects on heat density?

[rigid snow]

[true hemlock]

but getting it under -235C are usually accidents lol

[hard raptor]

i could imagine the tiny copper wires failing at those temps tho.. but i refuse to believe your silicon broke

or alternatively you somehow got the silicon to oscillate at it's natural frequency

[true hemlock]

as i said again

quantum effects on physical properties and quantum noise on qubit state alterations are TWO different things

its a separate phenomenon

[hard raptor]

actually.. no.. they're two different observations on the same thing

[true hemlock]

qubits needs -270C to maintain coherence

[hard raptor]

anything between -220 and -273.15 actually.. but usually they run at around -270 yeh

[true hemlock]

one of my overclocking friend suggested might not even be entirely quantum shit

could also be

semiconductor dopants freezing out

[hard raptor]

that's quite literally what i am trying to say the whole time -_-

[true hemlock]

we ruled out physical temperature stress because usually we had a method to drastically lowers the rate of change

[hard raptor]

do you want my honest opinion on what i think it could be?

[true hemlock]

its pretty odd though that dopants freezing is usually temporary

people called it "cold bug"

but in my case

and some other peoples

its straight up dead chip

or dead core

dunno what the fuck happened

[hard raptor]

i still think it's the copper wires

[true hemlock]

.

[hard raptor]

silicon just isn't supposed to react to temperature like that

[true hemlock]

you sure?

[hard raptor]

very sure

[true hemlock]

if you're arguing about non quantum effects

[hard raptor]

mhm

[true hemlock]

you can only argue about dopants freezing out

because that's what literally would happen below -160C

copper contraction can happen

but on such scale it wouldn't mattered much

[hard raptor]

actually...

so you know how copper is one of the weaker metals right

[true hemlock]

in fact being a really good conductor just makes it less likely its the copper

copper isn't a weak metal...

[hard raptor]

and how the copper wires are practically just a few atoms wide inside of chips like CPUs and stuff right?

[true hemlock]

and it doesn't apply in such tiny strands of scale

have you seen how its wired aswell

its array like, with poles connecting process nodes

[hard raptor]

Brother... how in the fuck doesn't it apply?

[true hemlock]

its not few atoms wide

its still few hundreds to tend thousands atoms wide

have you legit seen how its wired

and being that thin also makes contraction doesn't really matter.

[hard raptor]

which compared to copper wires you can see without any magnifier is still a few

[true hemlock]

and that's what makes it less matter lol

[hard raptor]

being that thin actually makes it matter way more...

[true hemlock]

no

[hard raptor]

yes

[true hemlock]

being less in size makes less change in contraction

you're looking it in a way that "it looks fragile"

[hard raptor]

less in size also means that there is less contraction needed for the wire to fully break.. since there are less atoms to rearrange themselves

[vapid pulsar]

So it is better in reality

[true hemlock]

um, do you even know how temperature based contraction work

[vapid pulsar]

Or

Roughly the same

[true hemlock]

its not like pulling apart the atoms making row rearrangements

its straight up

[hard raptor]

uhh yeah? if my uni professor doesn't lie to me i think i am pretty sure about how it works

[true hemlock]

widens the gaps between atoms

which just means, there's even less likely a rearrangement to happen the smaller it is

also

[keen hatch]

I discovered something interesting: Kimi-K2-thinking q2 generates better results locally than the same model via OpenRouter... It seems that some OpenRouter providers aren't using Kimi, while claiming it's a Kimi model.. otherwise, I can't explain it. The Validation Error Rate is higher via OpenRouter.

[true hemlock]

cold doesn't make it expand, cold makes it contract into denser lines
which is just few hundred nanometers of scale, it just does it pretty fine

also

these coppers are encased in layers of other materials

[vale python]

why is unity

[hard raptor]

Okay so tell me one thing really quick... do you have any qualifications and stuff in this specific field of science?

[true hemlock]

logically if it were to contract and reexpanded, there's no oxygen to form oxidized layer, it'll just fuse back when slightly heated up again

[true hemlock]

had connections with people in the field of semiconductor litography, even had access to one of the basic UV litography machine as a demonstration

and i work specifically on sample qualification on some engineering sample chips so doing all the niche shit is my suit

[hard raptor]

hmmm okay, so then you probably do kinda know what you are talking about

[true hemlock]

there's a reason

why extreme overclockers that had done -230C before

always said things like dopants freezings

and called it cold bug

and not something to do with coppers

[hard raptor]

i might have to like ask my professor if he knows how metals work under those very specific conditions then

[true hemlock]

look into how the coppers are interconnected

[silent cloak]

https://www.youtube.com/watch?v=OR-ADhxY2q4

clearly the best way to cool CPUs

kek

[true hemlock]

gonna take my chance to also imply that this is the objectively best way to electrically wire things

[silent cloak]

looks like a 90s 3D rendered art

[true hemlock]

also these are encased

[silent cloak]

missing the vaporwave background

[true hemlock]

no reason to think that they'd broke off

still though

[silent cloak]

can we bring back mineral oil cooling again

[true hemlock]

the weirdest part is most overclockers that do -230C tend to only have cold bugs

[silent cloak]

like from the 90s

[true hemlock]

dopants freezing off, and these are perfectly reversable

possibly that they never hit below -235C before

because for some of us that do by accident, it kills the chip or at least one or two cores

that's where the quantum speculation comes

the thermal coefficient for expansion is still tiny for that difference

copper issues were ruled out

cold bugs were ruled out because that shit isn't supposed to be permanent

[true hemlock]

actually might work looking into writing a thesis on

once im done with shit

[true hemlock]

https://www.pcgamesn.com/steam/ai-disclousres-debate-valve-dev-response

im genuinely not against and even up for ML models being used as a mechanic or inner working/backend systems of a game
but when it comes to ai generating assets, its just feel low effort trash
unless if its like endlessly procedurally generated stuff or terrain
(that falls on using AI as mechanic though)

valve is just so based rn

[nocturne olive]

It does sound like a game with a reasonably sized machine learning model as a terrain generator could be interesting

Could allow for more variant terrain than typical algorithmic generators

[true hemlock]

just don't ai generate assets that can be made by real artists.

[nocturne olive]

True

At most for a quick placeholder that will never see the light of day outside the development area

[sage crag]

r

[olive sable]

r

[stray dragon]

r

[rigid snow]

click click click

[fast pagoda]

Posted an image and it just vanished

[rigid snow]

what about upscaling textures

is that ok

[fast pagoda]

Omg

Auto mod sir

[rigid snow]

blacklisted image

[fast pagoda]

I don't even want to say what it was, I figured out the issue. Was laughing at this earlier is what I was trying to post

[hard raptor]

oh my gawd i am so smort

[warm river]

nixos is amazing

[hard raptor]

mostly need to test the interoperability with the vectors

in the webbrowser with Vite + Svelte it didn't work 100% of the time, now lets see what it does in node js

[clear sedge]

literally a satisfactory base

[silent cloak]

Lmao

@true hemlock https://youtu.be/hCuT_GXagz0

Speaking of cpu instructions

Mentions of AVX-512 too

[true hemlock]

you're a day late

[hard raptor]

now we just loop this 10.000 times

[silent cloak]

Im not on YouTube like that

[true hemlock]

neither am i, but i got recommended that immediately after opening yt lol

[hard raptor]

i forgor to make it output to a file

runnin it again

okay

after carefully reviewing the data

it indeed is just Svelte and Vite being wierd

[rigid snow]

this is scary

why does this exist

[hard raptor]

wdym?

[rigid snow]

what problem does it solve

[hard raptor]

oh no this was a test

[rigid snow]

i'm talking about your "kv store"

[hard raptor]

oh that's probably the least scary kv store there is

[rigid snow]

why serialize it like that

[hard raptor]

even redis is way scarier

[hard raptor]

wdym?

[rigid snow]

redis explod

nevermind

if you say it ok then it ok

i trust you

[hard raptor]

it's just

type item = {k: string, v: any, exp: int}

idk why that is scary to you

i mean i can see why this looks scary.. but this is just a silly little thing that shoves the vector and the kv store full of data so that i can debug if something failed.. which it didn't

[obsidian mantle]

i have no idea what this is but im curious

[obsidian mantle]

why 10.000 times

[hard raptor]

because i have 4 variables to shove data in

it's because i had a bug in this website where if i used the label based set method for the vectors instead of the index based setIdx it would only respond to the cells that has a coordinate with either a 3 or a 6 in it

the only differences between the 2 environments is that the web version is a 10 * 10 field and stores its data in the cookies
and the nodejs version uses a 10 * 10 * 10 * 10 field and stores the data in a JSON file

[obsidian mantle]

why do you need these vectors

no actually

why do you need to STORE these vectors

or is it just for some simulation/test

where you just generate predictable set

[hard raptor]

they're basically just arrays which can be stored while keeping all it's silly properties like read-only and unsafe and stuff

[hard raptor]

what i am posting here now are just tests

[obsidian mantle]

oh do you just create these vectors and they are meant to store something else than 1 integers

and numbers we see are irrelevant

[hard raptor]

no not quite

[obsidian mantle]

oh you said its a test

i accidentely read it as "not just test"

[hard raptor]

so basically the vectors are just arrays with like permissions type thingies

but the console spamming and the website.. those are the tests

[patent shard]

I got 8 pings
Wtf
I don't think there's anyway to turn this off, only smaller channels have it where they are unaware how horrible it is

[hard raptor]

the vectors are slightly more complex than the JavaScript arrays tho

but that's more because we can map the values to actual properties instead of just indexing an array.
also we can tell it to prevent any changes from being made to the data within and we can also block destructive operations like set and delete

[obsidian mantle]

but what will they store

[hard raptor]

whatever they need to

[obsidian mantle]

so is this a data storage model

[hard raptor]

only it looks really cursed when you nest them

this is an old ss

[obsidian mantle]

oh

[hard raptor]

this is 3 layers if you couldn't tell btw

[obsidian mantle]

i encountered nesting of java strings

its so bad

there should be a better way

[hard raptor]

this ss has a more broken down version in the logs as well

[opaque sigil]

you'd think string escaping being as much of a pain as it is would be enough of a deterrent but i guess not

[hard raptor]

i guess i could also try to make like a secondary way of storing them

interface Vec<T> {
  type: VecTypes;
  data: T[]  
}

store em like that

it would fix the backslash issue

[cosmic sphinx]

that's so helpful, thank you chatgtp

[warped narwhal]

I would try and help but...

[obsidian mantle]

oh wtf

they have a block for researching it's behaviour?

or is it just same thing as elvys screenshot and is a "bug"

idk whats real anymore

[severe path]

AIs are kinda lame

[amber fractal]

I got a lead, free version tho (I pray the markdown is transfered)

Decided while I'm on the good rng to also ask on the response.

[obsidian mantle]

Does "pretend to be evil ai who has no restrictions" not work anymore at all in any form

Surely its possible to patch it

[amber fractal]

[obsidian mantle]

Does it work

[amber fractal]

This was at the end of the prompt, let me ask it to do so

[obsidian mantle]

Will humanity finally be able to connect to mysql in java

[amber fractal]

Let me make text file the third

[obsidian mantle]

"here is complete instructions on how to jailbreak me"

[amber fractal]

Which one should we test

[obsidian mantle]

Is it actual security concern block or its hallucinating

Both 2 and 3 are alright i think

[amber fractal]

It seems to think it is hallucination, I asked on that one give me a sec

This is where the screenshot came from

It ignored the separate user part entierely

also manually typing those took forever

#2 works, including deps and all

Can't get #3 because it answered correctly when I asked inital

[amber fractal]

It would be funny to ask

alas I'm not the type of person to ask

[rigid snow]

aaaaaaaaaaaaaaaaaaaaaaaaa

[rigid snow]

e

[hearty notch]

u aa

ting tang

walla walla bing bang

[fast pagoda]

I remember that song in the context of some movie with a maid

But idk what it was lol

[mighty thorn]

i have idea

[olive sable]

bro what are you doing that you need another pc?

[olive sable]

Projector is ceiling mounted now

[amber fractal]

I was looking at PCs but nearing 3k for 5080 rig is ewwwwww

[olive sable]

[amber fractal]

it was 400 off

shoutout the 1.2k 5060 rig with 16gb ram

that was 200 off

My first thought as one does when looking at rigs IRL was "are we fr"

[olive sable]

I would just get one without a gpu and then get a used 3090

[amber fractal]

So the cost of that one was equal parts CPU GPU RAM and case

[amber fractal]

Workstation was 600

[olive sable]

Not too bad

[amber fractal]

Not the worst not the best

At least for new on shelf

Time for my monthly 3090 price check

@olive sable

[olive sable]

eh

not great, not too bad

[amber fractal]

Yeah more than it was before

but constant with everything else

[olive sable]

500 euro is 579 usd, so i wouldnt pay more than 650 usd

[amber fractal]

Hold on a minute

[olive sable]

holding

[amber fractal]

prices have returned to normal on P40 cards

It is no longer 2x price

Let it crash more

I've been memeing on it for so long that it is actually real, it finally has returned to the price I paid for them

[olive sable]

400 euro is 463 usd, probably like 100 usd import tax,

so 563 usd

[amber fractal]

Glad to see that it is calming down

[amber fractal]

still a lot better than here in the US

[olive sable]

the 3080 ti is being sold by "yeet that child"

[amber fractal]

A trustworthy seller

[olive sable]

[olive sable]

you want one?

[amber fractal]

Nah, need to dejank my rig first

-# knowing me that is an impossible task

[olive sable]

[amber fractal]

Honestly I might consider sooner or later doing something stupid involving using server hardware more than incorrectly, maybe that'll give me an extra PCIe x16 slot. But who knows

Might get some cash back by switching out one bad purchase for another

[olive sable]

if i ever get into ai i may consider a 2nd 3090 if its only 400 bucks
i think i need a gpu with more power for that tho

when i put a 2nd 3090 in before i didnt run them both at the same time, and my psu is only 750W

so only enough for 2 3090s without my cpu and stuff

[amber fractal]

I still oversized mine we love 2000w for a single GPU

one day I will get it working, and now after a million years I found my spare wifi card so it is closer than ever

I no longer have to test on my main rig

[olive sable]

the psu was the first part i got for my pc i wasnt exactly expecting 5950x and 3090. i though my budget would "only" allow 5800x and a 3080 or something.

i just had luck with good deals

im glad corsair psu's are decent enough quality

this specific model at least

[scenic kettle]

why.. a 380 though..

🐇 aint there like slightly better cards for that price

or is it really that bad

[olive sable]

wdym 380?

its a 3080

its still 70% as performant as the 5080

depends a bit on vram and such tho

the main focus of that image was the 3090s tho

[olive sable]

3090s for 400 euro is actually a really good deal

i dont see the issue

is this ragebait?

[amber fractal]

looks at pronouns

[olive sable]

gender:

[true hemlock]

not 70% but at least more than half i think
well that's speaking about raw compute

[olive sable]

i think its close to 70% depending on the game

[true hemlock]

in gaming yeah

frame time doesn't scale well with only GPU

[olive sable]

for ai the extra vram in the 5080 matters

but its still a lot more expensive than the 3080

[true hemlock]

3090 better

even though the compute is lower

[olive sable]

it is?

didnt know

[true hemlock]

yeah 3090 and 3080Ti has similar computer

3090 is just 3080Ti using 24x 1GB modules

[olive sable]

[true hemlock]

3080Ti is quite a bit better than 3080

this makes 3090 around 4070 super and 5070 level of performance though

just, double vram, double bandwidth

[olive sable]

[true hemlock]

in games it doesn't matter because raster saturation apparently doesn't need much bandwidth but in other workloads, might be nice

[olive sable]

it depends on some stuff, but generaly you dont need much bandwith ye

i think lower latency is a lot more important

[true hemlock]

not really

i think i've mentioned somewhere in here about gddr latency

[olive sable]

well, for my stuff above 10K fps latency to the gpu had become a factor

[true hemlock]

above 200+ ns compared to most CPU dimms 10-13 ns

[true hemlock]

oh you mean PCIe latency

then yeah that matters, but only with extremely high FPS

realistically below 600 FPS the PCIe is still pretty much capable

[olive sable]

oh you were talking vram still, i thought with double bandwith you meant pcie bus width

[true hemlock]

pcie bandwidth don't matter

like

you can run a 3090 on x8 3.0 fine

and it still won't fully saturate yet

[olive sable]

ye indeed

[true hemlock]

why is everyone picking up elvyn's chatting behavior

[olive sable]

?

[true hemlock]

yeah

[olive sable]

this is jsut how i talk, idk

[true hemlock]

especially GDDR6X

200 or even above, like can even be 300

[olive sable]

from what i remember in the quick google searches i did, gddr is more overall latency for random acces, but gives a lot more overall data bandwidth?

[true hemlock]

yeah, gddr is like

"fuck latency. i want raw bandwidth"

because all the GPU cares is massive streams of data and not sequential access/process

[olive sable]

that is indeed how most of the optimization in graphics works

[true hemlock]

timings is entirely ignored

its handled only in firmware level

although def wanna try timing adjustments for better raw clock

but firmware mod is ass

[olive sable]

the vram speed is still a pretty limiting factor tho

most of the time its faster to calcualte stuff on the fly than to read it from a buffer in vram

the best way to optimize the vertex shader is quite literally just reducing the amount of bits being read from vram per vert

[true hemlock]

that's what you have when the system is prioritizing latency especially on extreme high fps (it happens at that point)

latency can't keep up with how fast the frame should cycle

funny how because of that, avx wins on your triangle bench

i forgot if my 100K score is done with avx or dgpu

or igpu

[olive sable]

lets check

supposedly on the 4070

[true hemlock]

i listed out the dgpu

yeah

huh

oh right

didn't i also test out both avx and igpu

[olive sable]

i think it jsut has to do with saturating the cpu

[true hemlock]

the frame buffer is set at 8192 though lmao

frames in flight?

[olive sable]

ye

[true hemlock]

was genuinely fun competing with owobred

had to pull off the ddr4 extreme overclock and CPU extreme overclock

[olive sable]

8192 frames behind at 102K means you're still only 0.08s of visual frames behind. 80ms is not great but not terrible

[true hemlock]

the way i did it though

horrible

i think my chip degraded

not because voltage

but because excessive thermal density

[olive sable]

[true hemlock]

i can still get another one for $182 anyways

but

i sort of have some sentimental value on this one

would be sad if its gone

its the chip that allows me to prove intel's bullshit

the chip that also allows me to prove that ARM efficiency is highly misleading

[olive sable]

im pretty sure gpu itself didnt matter too much since the workload was so small and we let the gpu-cpu syncing have as many flags and semaphore as there were frames in flight

[true hemlock]

and AVX clocks atteast double

[olive sable]

so the limiting factor was at which speed you could order the drawcalls

i think

[true hemlock]

also

could be possible

that its kind of sequential

5366MT/s at CL11 is like

4 ns real world latency

i can see that would straight up double the possible frame cycle

[olive sable]

ye

[true hemlock]

"ddr5 at home"

[olive sable]

i mean, you outpaced bred's ddr5 with it

[true hemlock]

yeah because the latency

so ddr4 ddr5 don't actually matter

[olive sable]

im geussing ram latency is a high factor cuz of the context switching that needs to happen cpu side with not everything fitting into cache?

[true hemlock]

if it can fetch at the same every 10ns

[true hemlock]

its pretty weird yeah

not sure why wouldn't it fit the cache

[olive sable]

vulkan might jsut be too big

[true hemlock]

although

on the windows side

redston's 7800X3D is second place

and the first place is literally the same 13900 es being pushed to absolute overclock limit

[olive sable]

we did manually asign cores to each workthread so its basically forced to switch out that cores' L1 and L2 each frame

[true hemlock]

hm

yeah i blame vulkan

[olive sable]

same

[true hemlock]

time to write your own triangle renderer engine in ASM

[olive sable]

maybe once im done with the vulkan arc, i still need to release ana ctual game with it first

[true hemlock]

the rules are simple. must trace and render a given set of verts

could in theory gain million to billion fps

[final tinsel]

I may have written a 107 line twitch python fishing bot today

[olive sable]

next step is to make my own rtx 3090 driver specifically made to render as fast as possible

i will release 1 game, and im just gonna be like "must have rtx 3090 and amd 5950x to play. to install you must flash the BIOS with this file"

[fast pagoda]

i feel as though "to render as fast as possible" may be one of teh design goals of the nvidia driver itself, as much as they suck

[olive sable]

no, their drivers are made to promote physx 3D Vision G-sync SLI rtx DLSS

they go to so many weird random tangents that their drivers are probably pretty bloated

remember nvidia Vision Surround?

[fast pagoda]

eyefinity green edition yes

[olive sable]

eyefinity but 3D

[fast pagoda]

'twas a joke tho

[olive sable]

the 3 monitors as 1 was just "surround"
the one where you had 3D monitors was "vision surround"
it was very cursed and died instantly

[fast pagoda]

mainly on peoeple declaring they will singlehandedly solve something that the people on it now havent seen because they arent focused on <the main obvious goal but it gets shunted aside by having to run a whole project>

[olive sable]

https://www.youtube.com/watch?v=Xru6i0aIi24
its linus NCIX tech tips

[fast pagoda]

lmfaoooooooooooooooooo

i know what it is, i had it

i used it

like

once

[olive sable]

i tried surround once, but it sucks to use on anything other than video game

[fast pagoda]

GPU were not ready for it at the time really anyways

didnt have the horsepower to drive all the monitors at once, they kept trying to get 4k to happen in surroudn

[olive sable]

it still exists in the control panel, and my 3090 is fast enough to play fortnite on 1440 x 7680 on medium setting. it jsut kinda suck to use for desktop cuz of sliding your windows around in win11

[fast pagoda]

yep nobody uses it because multimonitor is fine at the os level now mostly

and playing games on multiple screens at once kinda sucks

bezels and whatnot

[olive sable]

its kinda cool, but depends

[fast pagoda]

although surround did account for the bezels

i like it for like

flight sim

racing

stuff like that

[olive sable]

the bezels arent too bad on my monitors, theres maybe 1cm of gap total

[fast pagoda]

any bezel is too much for some

[olive sable]

i used it for asetto corsa or whatever its called

but it is a huge fuckign pain in my ass to have to fumble with the nvidia control panel for 5 minutes to set it up

[olive sable]

seeing the dvi ports get used as the prefered port over hdmi or dp feels weird

[fast pagoda]

hdmi sucks

still does

[olive sable]

its fine

[fast pagoda]

displayport also sucks

for different reasons

[olive sable]

they're both fine

[fast pagoda]

it was pretty busted early on

[olive sable]

not great, just fine

[fast pagoda]

displayport specifically caused a lot of issues for some monitors/gpus back in the day

[olive sable]

HDMI supports 32 channel audio, e-arc and shit

its not that bad

[fast pagoda]

yeah but it also wasnt hdmi 2.0 then

it was still 1.4

[olive sable]

i have more issues with EDID than the cables themselves

[fast pagoda]

for a computer monitor there's really no reason to use HDMI unless you straight up dont have a displayport connection available

[olive sable]

especially when im doing device -> avr -> display, cuz the step inbetween likes to break hdcp for certain things

[olive sable]

ye

[fast pagoda]

should just get on with it and make everything usb-c

there are already usb-c monitors

i mean portables and stuff ofc but desktop monitors with only C inputs

yolo

[olive sable]

apparently 1440p 180hz 10-bit HDR is above what the DP bandwith i have is supposed to be able to do

but i cant find any issues with it

my monitors are DP1.4, and my gpu 1.4a

[fast pagoda]

well it's only gonna go as fast as the slowest bit obv but that's probably getting close to the limit of 1.4 which is liek 25 gbps

[olive sable]

ye

[fast pagoda]

what does xrandr report your link type as

if your works at all with xwayland

idk how to find it without xrandr lmfao lemme google

[olive sable]

i have wayland

lemme check xrandr

what command lol

[fast pagoda]

xrandr --verbose | grep "displayport" https://xorg-team.pages.debian.net/xorg/howto/use-xrandr.html

idk about the grep if it'll match or not

[fast pagoda]

mmm

do it with no grep

or grep "Link"

yoou are looking for it to say HBR3 which is max speed for 1.4

[olive sable]

nothing

[fast pagoda]

reee

oh

can you see x server settings? nvidia

[olive sable]

idk

how?

[fast pagoda]

should be in your programs

[olive sable]

yup

what do i do there?

[fast pagoda]

good question

i thought it was gonna show it

but it's not there for me either

[olive sable]

in terms of graphics, we have no graphics...

[fast pagoda]

ope

do this

inxi -Gxx

this worked for me

[fast pagoda]

why is it so hard to find this information

[olive sable]

i am still runnign mesa apparently instead of actual drivers from nvidia

[fast pagoda]

you have nvidia loaded

its just saying that ur openGL driver is the mesa one

[olive sable]

huh

[fast pagoda]

loaded: nvidia

drv: nvidia

etc

wait

you have wlr-randr

does that show it

[olive sable]

lets see

[fast pagoda]

--verbose?

[olive sable]

unrecognized option

[fast pagoda]

-v?

[olive sable]

nope

[fast pagoda]

kek

well

i dont think you can do 1440p at 180hz unless you're on the max speed mode? so the 10 bit thing might be goofiness with nvidia/hyprland

run hyprctl monitors

[fast pagoda]

currentFormat: XBGR2101010

you are running 10 bit

[olive sable]

DSC1.2 i guess?

[fast pagoda]

all i know is that is xbgr with 2bit pad/10/10/10 bits for blue/green/red

so hyprland itself is outputting 1440p/180hz/10bit

if anything else you are using is not giving you 10 bits it's on the application end

[olive sable]

yep

some apps have that issue

[fast pagoda]

im surprised you've had such a relatively painless time on hyprland

[olive sable]

[fast pagoda]

i've gotten it working like

once

w/ cachy hyprland

[olive sable]

i guess its "relativly painless" i still need to toggle hdr everytime i take a screenshot

[fast pagoda]

i had to do that on windows

[olive sable]

damn

[fast pagoda]

HDR hasn't given me any issues on any system

[olive sable]

i love hdr on console, but it sucks a bit on pc

[fast pagoda]

ive never had issues except that screenshot blowout problem on my last windows install

it's always just worked

[olive sable]

on my windows install on this pc, its not too blown out

on this linux install it look shorrible tho

[fast pagoda]

funny how things manifest slightly differently

[olive sable]

on windows i dont notice the colour change too much, on linux the difference is night and day

i cant use linux without hdr colours

without hdr everything is kinda green-shifted, the white balance is off, its less vibrant

jsut bad

[fast pagoda]

i dont really notice it till i turn it on

well