#🔌│tech
1 messages · Page 222 of 1
That's what happens when you use openPGP for everything
----BEGIN PGP MESSAGE-----
I unironically don't even use a password because I am too lazy
it's gonna fuck me one day, but I'm saving aeons of time altogether
Ok boys, we have a new brain rot take champion.
I wouldn't have been brave enough to try it but what the fuck do I know.
and the competition was rather stout
so it is rather impressive
Wait, wait, your sudo su has no password?
That would be metal af
yup. I keep thinking I should make one, but then "nah"
Based
Please tell me you at least make disc image backups regularly?
Like, once a week?
Lmfao no goddamn chance.
Not that it would even help you because disk backups can get fucked as well.
if I'm too lazy for a password, I'm not gonna bother with that either
and I do realize how extraordinarily stupid that is, don't worry
Nah man just download FaceID copy off of github repo of sorts and use that
The fact that this guy hasnt had problems yet is proof of something
Perfect security
It's proof that luck exists, but always remember you only need to get unlucky once, whereas you need to get lucky all the time.
Think about it like driving a car, a seatbelt might not save you, but for the one fuckup it will you'd best be glad.
On Windows at least you have a lot more protection when your luck runs out
Doesn't matter how much you put into security, you cant defend from something you dont know exists
Please don't ever go into cybersecurity.
you can follow a set of best practices to rule out the most obvious attack vectors (a password would be one of them!)
I won't 🙂
- Passive Protection
- Adaptive antivirus
- Backup schemas
- DNS protection
- AI checking
- Polymorphic code
That's off the top of my head.
All of that is passive.
Depends on how much you're paying 😄
Good job, and then when you get attacked from something that isnt that?
... backups?
I mean, I don't have plex
That's what the polymorphic code and backups are for.
What happens when your backup is hijacked?
Where tf are you storing it
do you not keep your backups OFF YOUR SYSTEM?
That's what 3:2:1 and incremental full backups and config backups are for.
Lmao no, Vex.
What if there was a zero day for 3 years undetected throughout all those backups?
3:2:1
3 backups of 2 kinds, 1 of which is stored off site.
That's what polymorphic code is for.
There are some encrypted types of storage that you could prolly just keep connected at all times, but I don't specialize in infosec so I don't care much
Bitlocker goes brrt
Not 0-days? Amateur
Like yeah I get, polymorphic is hard to work with and expensive as fuck, but under the right circumstance good luck getting through it.
depends if it's salted or not. You can just as well just run it clear worded if you're believing that SHA is safe on its own
I will eventually make a backup of my data vault, with tons of movies, music, etc on it. but nothing important is stored on my system drive. it could go up in flames now and all I'd need is a new NVMe drive
Actually, no
Certain exploits can hide in your hardware
So, you'd have to replace everything
Oh and if we're talking about a 0day that snuck onto everything and went undetected for years? Yeah we had that it was called log4js and it caused an IMMENSE amount of problems.
you mean like Asus und their armory crate auto installer shit?
Thats what i borrowed that idea from

Oh god no
Not the armory crate
It can always happen again
actually, is there still a full copy of MINIX stored in every intel chipset, along with a web server?
Yes and using the fact that log4js can happen is not a reason to go 'oh well, no security is fine'.
So no, Linux is still an insecure hellscape.
F tier OS for critical shit.
Then why are so many companies using it for critical shit?
Because it happens to be quick to work with and cheap. Don't mistake that for intelligence.
Basically.
And for a server you fuck off sudo once you're done with it.
Tbf Linux is just used for servers and not much else
Like an actual OS with code signing should be working.
Or if you're based you use Talos LInux since it says 'no, sudo is cringe' and doesn't have it.
Most secure OS I've ever seen.
No shell, no terminal, no SSH, nothing.
You do API or fuck off.
or you just install hackintosh and be the actual based, chad user 
My calculator is the most secure os. It only allows me to make calculation
I genuinely need to look up if hackintosh is even a thing anymore
It existed like 3 years back lol
Niche circles it is.
I believe not after since the introduction of the t series security chip
Yah can probably jank together something
Well, someone accessed latest beta lol
Hah how wonderful
- kernel is digitally signed and your system won’t boot if binaries are modified
- do it in runtime (or in memory, rather) - get fucked by patchguard
so yeah
^ There are ways around that, but like fr they are not anywhere near as simple as sudo.
And have no shred of hope for being done from a user.
of course there are always ways to disable patchguard, yes
nothing is impenetrable

i mean arm pcs soon?
they're already out there
I'm using one right now
^
your anticheats are also signed, and the signature can be revoked at any time
Ah come on
also that
i mean a general non SBC arm computer
Reverse-hackintosh? Now that sounds interesting but still would be shit
but considering ARM based pcs were more of a joke than anything else
with replacable cpu ram etc
you mean a mac using Asahi linux?
yeah it is shit
but consistently getting better
iirc there was a MIPS port of windows 2000 at one point
Anyways, speaking of Qualcomm and ARM on Windows https://english.newstracklive.com/news/qualcomms-snapdragon-8cx-gen-4-faces-challenges-sc71-nu384-ta384-1297350-1.html
Why must you suck 
8cx needs to get better, I want proper ARM with performance of M2 capable of Windows
Thing is, on paper 8cx gen 4 is incredible
I'm really not sure where Qualcomm is lacking tbf
The compute cores don't really appear to be an issue
Adrenos do kinda suck
Nvidia is too busy raking money from H100
ermahgerd, the Tegra phones were a decade ago
I'd argue an RTX 4060 is pretty cheap entry for machine learning too 
Oh jesus these are cheap af
Although this seems like a far better buy https://www.ebay.de/itm/266255192426
Orin's pretty good
RTX 4060 is still prolly the cheapest entry for some solid machine learning though
CUDA is everything nowadays, it's sad
It should be open source
Hey uhhhhh. Guys.
I think the nvidia CEO would rather nail his dick to a burning log than to release any technology as open source
What's up princess
Any ideas on making a cheap as fuck compute box for LLMs? I have GPUs but like what hardware is good for this?
Can I do 1x PCIe per card? Do I need big CPU go brrrrrrrr?
Mate you mine crypto, you can answer these questions yourself 
Mining crypto is a LOOOOOOT different than LLMs from a cursory perspective.
nobody's saying they ain't in their right, would be nice is all
Crypto is a lot of calculations, doesn't really need bandwidth, no idea what the bandwidth reqs are for an LLM of reasonable size.
Thought one of you in here might just like... already have that knowledge. 
why is nobody here talkiong about unity drama
Just buy a MI200
kinda dissapointed in chat
Trash bait
I mean unity is as good as done as a company if they keep this change
I was thinking maybe it's too easy to pin it all on CEO guy, surely their entire executive branch must have approved of it right
This doesn't have CUDA, noob.
Nah it's the financial intern
the ceo apparently sold shares a week before the change
not that many I hear
Average CUDA enjoyer.
still, shareholders always know the best ways to kill a company
Just HIP it and MI200 
Glaxes! Come now, don't you think that naturally the shareholder would want a long and extremely rich future for the company to extract tenfold the money with zero effort?
That'd be the smart thing to do. Surely shareholders are smart.
Surely.
Surely they don't plan to make the company gain a large amount of money at once then sell the shares they own off once they profit
Surely.
That would be very disingenuous, no shareholder is doing that. The internet told me so.
What are you people talking about? Money? I LOVE money!
No no, I'm sure the profit driven shareholders exist specifically to help the company they've chosen to back.
I would make people pay each time for a mag reload in CoD ngl
Sounds like easy money
Bro please you are aware that the Unity CEO suggested that once?
A cent per bullet fired in FPS.
Or each time you boot windows there's a 10 pence boot fee?
USB plugin is 1 cent per minute plugged in ngl
Dw lads, your average LMG costs $1.20 per mag.
Cause of "online DRM"
Not sure if rising costs are really the fault of economical instability of the practice, or just CEOs trying to launder more cash
You know it's #2.
you'd think that pattern recognition would be a skill that players naturally develop, but then we still get tons of people preordering games
Had to stop like 6 of my friends from pre ordering fuckin' Starfield. The game coming out of the studio that has done nothing but make failures and mobile games for almost 10 years.
Excessive cringe.
If pattern recognition was a skill, you'd prolly make a password for sudo everytime you see a password form ngl, so you did fail on that skill
He's not wrong.
"Can't wait to report your death"
- some shitass gnome
oh man, that was only 2018? feels like it was longer ago
Ngl was did biscuit even do, I never followed him. Internet journalist that was morally consumer-sided?
Was that his whole shtick or was there more to him
People reminisce of the lad fondly
he was a pro-consumer demagogue
I watched him a few times, but he didn't make the kind of stuff I enjoyed
mind that this was still in the post-gamergate fallout, where everything was highly political somehow
I still quite haven't cared to learn what it was all really about. there was always this smell of shit wafting around whenever people argued about it
who knows, maybe it was a necessary discussion
how did he die
Gnome disease
Twitter overdose.
he ate a bad bag of fish and chips..... || it was cancer ||
that does it
nah keemstar just got him first
I already said Twitter overdose.
Let's just say that this didn't age well.

That'll kill ya
Aged like the child chemotherapy ward. I mean what?
that joke made my hair fall out!

I think it peaked with crippled female WW2 soldiers in Battlefield V
Ok lads, it finally happened.
It took 9 minutes but we have a response back from the LLM.
I need GPU acceleration and CUDA.
time to buy a 4090 ! 💸
out of curiosity, do LLMs meaningfully profit from SLI?
From SLI I doubt it.
I still frequent KotakuInAction, but man it's gotten a lot of incel activity
Model parallelism has a few ways, Naive Model Parallelism takes the layers of the model and distributes them over the GPUs so you can get enormous RAM monsters, for example.
is it really called naive or is that a typo?
But like yeah this won't be working with the classic crypto 'yeet everything into the thing and minimum bandwidth', this application is really high bandwidth.
Was not a typo.
No idea why it's called that.
the main deficiency and why this one is called “naive” MP, is that all but one GPU is idle at any given moment. So if 4 GPUs are used, it’s almost identical to quadrupling the amount of memory of a single GPU, and ignoring the rest of the hardware. Plus there is the overhead of copying the data between devices. So 4x 6GB cards will be able to accommodate the same size as 1x 24GB card using naive MP, except the latter will complete the training faster, since it doesn’t have the data copying overhead. [...]
that explains my question!
ermahgerd. this looks like a fun hobby if you have the hardware
O_O
https://www.youtube.com/watch?v=ydGdHjIncbk this is what i meant
Can the world's fastest Arm desktop handle gaming? And before that, can it even install Windows?
See my first video on this desktop here: https://www.youtube.com/watch?v=wl5H5rT87JE
Adlink/Ampere did not pay for this video or sponsor it in any way, and had no input into the content of the video. They did, however, supply the hardware under tes...
and now im really excited for those PCs
After laying off 10% of its workforce earlier this year.
I'm sorry what
"So anyways, you lot are unfired"
Lotta time you do stuff like this to remove inefficiency.
not coming anytime soon in good results, nobody is making a high clocking single target focused arm chip which is what you want on desktop
Like lay of 10000 people mediocre that can't be moved up, hire on 10000 different people that can be moved up.
Everyone shows their surprised pikachu face, but it's calculated.
Like one thing you can do is lay off a shitload of people an encourage them to reapply so they brush up on their resume and you get a cleaner picture of the skillsets you have to play with.
Or get a good opportunity to shrink business and tighten up before rehiring the random extraneous bits, usually for cost reasons or what have you.
yet
yeah but nobody is making one now aka nothing will be out for years
last attempt was nuvia and that turend into snapdragon gen4 which has issues
@astral haven CnC article about starfield CPU utilization is also coming in the nearish future
sweet
I do definetly disagree with their conclusion on the first article however
I dont think the opposite is truem however I definetly think they put in as little as possible effort into making nvidia gpus work as they could
what that they beome l2 bound?
no, they concluded that it was not bethesdas fault
well its less that they said its not their fault and more there isnt an issue per se that would be someones fault
yea and thats wrong
imo
i dont think that bethesda intentionally nerfed, thats obviously false and would have been noticed immediately
but it is still bethesdas responsibility to make their games run properly
its definetly worse if its el big dude bullying the little guy, like was the case for intel in this case, but its also patently obvious that the usage of the gpu is not optimal
neither gpu is used perfectly but both are being used quite well like it's not there is a massive issues by using bad sized data chuncks that are forced to be split or anything
are they?
i agree that the issues are smaller than they could be
but you dont say a crime is ok just because they stole half of the money in the cash register
for sure this is an extreme example
but neither of the gpus averaged over 70% usage within the frame, and the nvidia one was very obviously lower on average
Higher occupancy does not imply high compute utilization
higher occupancy is more efficient compute utilization
now sure the average occupancy of these gpus being under 50% is not nescessarily a bad thing
that depends on the uArch, you can max the vector unity throughput without maxing out occupancy
but it does indicate that you are doing a poor job of keeping the gpu fed
if you know the performance is bad, and the occupancy is low, that is something you can move to fix
all of the GPUs here are capable of maxing out their vector units with one thread in each SMSP or SIMD
i take problem with the bad performance, and the occupancy is proof that an effort was not made
I don't think thats how occupancy is supposed to be viewed
🤷 if we can use it as evidence that they didnt intentionally tamper with the gpu optimization, we can use it as evidence that they did not put sufficient care in the optimization
Lower utilization is by design in Nvidia’s architecture. Nvidia SMs have smaller register files and can keep less work in flight.
this is from their cyberpunk pathtracing tests
so I guess they didnt put any care into that either?
cyberpunk performs equally poorly on everyone
look the conclusion is that the game performs more poorly on nvidia than it does on amd for starfield
i am using the cnc article as evidence that there is fixable disparity, not that it was intentional or even that it is in fact poorly optimized
I dont really think that's necessarily true though
and we can disagree on that
and im pretty open to disagreeing about the second half, but not about the first half
but like, less work is good for the cards so they can clock high and eat through it, so if one side can improve it the other can likely do similar things, if there is not something big missed somewhere
?
nah as long as warp never hits zero, the result is the same, the problem is that we dont see the lowest usage of the warp within a single sampling, we see the average, and the lower the average, the more likely it is that some of the warps are emptying
like warps in flight across the whole GPU never hits zero?
i mean sure?
but more i meant to an individual warp
the available warps for it to pick up never being zero
so, how much hardware does a single warp utilize?
a SIMD/SMSP but also a SIMD can have many warps in flight, having one warp on a SIMD is a corner case that you don't want to happen too often
yea and those are called SM on nvidia and half a cu on AMD right?
i might be mixing cu and wgp
oh no there are 4 simd's per sm and WGP oops
it probably happens as you're finishing up a shader and some warps finish before others. that sort of long tailed behavior can be a problem, but it wasn't that significant in starfield
is it worth to overclock my ram
i wouldnt say fc6 is nescessarily poorly optimized, it could just be graphically intensive to have that much foliage everywhere
starfield imo is substantially less graphically impressive than fc6
yea but that also happens in the middle
like these reporting steps only cycle along at what is it 1/10th the clock of the gpu?
and even then its still taking an average, you could have an entire sm empty, and miss a bunch of cycles waiting for more data even though the amount of warps in flight is enough to fill all of the warps 2 deep or whatever
thats an opinon, youll have to come up with it on your own
the actual safety risk to overclocking is low, but you can damage both your cpu and memory by messing with voltages for example.
and the amount of time it takes to stability test is high
if you dont play that much, it might not be worth the time investment
and if you play alot, the time you overclock is time you arent spending gaming
alr i’m probably not gonna test it i feel like my pc gonna explode
im not sure you'd blow anything up
though you can definitely decrease the lifespan of your hardware
my hardware already on its death bed
outright killing something is hard these days
my cpu been over locked at 4.6ghz for the past 5 years
and i just found out last week that it was overheating to like 90 celsius on a regular basis 💀
@young pecan ive never wanted a HDD more than i do rn
lmao
my drive is to fast for the glitch
yeah that one isnt ideal
limit the drive speed then
90 is still technically within spec for a lot if not all processors now but it's definitely on the high end of what you wanna be seeing
yeah i fixed the problem and now it’s chilling around 40 idle and 60-70 during heavy load
good
so it should be good
Much longer than that because the more you sample, the more overhead you incur and the more data you have to save somewhere. And the average is what you ultimately care about anyway
Yes, you can have an entire SM empty and miss a bunch of cycles waiting for data. It happens a lot and is why you have way less than 4 IPC for a SM on average.
Happens on CPUs too. If you program a performance counter and use the cmask/edge fields, you can see how long a particular core might have been "idle" waiting for data or idle at a particular stage at least
also my brain is mush now after raiding for too long so I shall fall into bed
Framerate can help you.
Tank your GPU perf and cap D2 out.
Upscale the game to 8k.
you can overclock I guess
probably the easiest thing to overclock since you have 0 voltage control on modern gpus without doing anything special
you just increase freq till it either loses performance (error correction) or starts showing instabilities then back off a bit
alternatively on rtx 3000 you can undervolt and it can increase performance, not so sure about other generations
do i just over clock it in the bios?
What GPU do you have?
Ah.
Right so that card use it on Precision X1 or MSI Afterburner.
The gist from memory with those is each power step is 6Mhz and, for every 3 degrees you drop you can push 1 step.
They tend to cap out around 2150~ Mhz on the core, some go higher, some go lower.
My roommate had one that could sustain 2210~, for example.
does the manufacturer matters?
oh yeah you start to lose clocks over like 50 or 60 celsius or something
As RAM goes? You can do 1000Mhz on it, but remember it's half so you're really doing 2000Mhz.
i mean only really for cooling purposes
msi puke
lOl
no way
sigh
Ok so the Duke uses an MSI Gaming X PCB and the cooler from a 1070.
It is not sufficient to run that card at it's maximum.
Man
However.
With that said you do have a Gaming X PCB which was one of the best at the time.
Find some other way to cool the thing and it performs near the top of what the chip can do.
bouta use my ceiling fan
it'll just run hotter so you won't get as much out of it as one with a better cooler
its not massively a problem really
^
You could jank together open loop or do DiWHY ig.
But it'll just run about 15c hotter than an actual GamingX card.
funnily enough i got it the other eay round, the cooler on my gigabyte 3060 ti appears to be very similar to the one on the 3070 ti
if not the same
doesn't cross 60 celsius either so probably is
actually iirc the 1080 Ti's should be capable of running at around 2000-2100 MHz VRAM
unless it's some Micron shit

Yes so under the OC you do 1000 since how RAM is overclocked on that generation you do 50% of your target.
but only the cheaper end models had micron
1000 in Afterburner is 2000 for the card.
yep
how much more fps am i getting if i get it to 2000
GDDR5X or something reasong. Idk.
like on a percentage base
No way to tell. You'll have to benchmark.
Some cards perform better than others, depends what VRAM your unit has.
its not going to be insane but it's not really difficult to do
well, Stock vs max OC'd it was around 15-20% for me
Samsung performs the best, Micron the second best.
there's no reason not to
Hynix is the shit you need to look out for on the 1080Tis. That caused problems with overclocking.
but then again, i've had an Aorus Xtreme with 2200 MHz VRAM
would i have to keep opening up msi afterburner or would the settings save
Save a profile and set start on boot.
you can save your settings
oh that makes things a lot nicer
the last time i used it was for fortnite like 5 years ago
and lowk i thought it was fake so i just messed with a bunch of settings 💀
pretty sure "start with windows boot" is on by default
ah yes fortnite 5 years ago when i had +150mhz core on my 1050 ti
my gpu used to be so goated

didnt help a bit because I was massively cpu limited anyway
fx 8350 😃
and 8gb single channel bottom of the barrel ddr3
well, the 1080 Ti is still quite nice, it was just too good when it launched
The fuck you mean used to be?
The 1080Ti is still somehow viable.
its still insane
7 years later.
i dunno how they did it but it keeps up fine still
Pascal is a fucking nuts generation.
Especially when you consider Maxwell came before it. It's like jumping from a Supra to a Bugatti.
i wonder if ampere is gonna have a similar lifetime seeing as it was a pretty decent generational jump too
probably not with how much they push ray tracing
and similar
honestly? Nope
im too young for that
Bro, Volta was after the 1080Ti. 
that is true. when i look at a lot of gpus now i would have to spend over 500 to get a noticeably better performing card
That one generation everyone forgot.
oh i was thinking of uhh
why can’t they make 1080 to v2😔
Bro not Fermi.
the 1080 Ti was just a too massive jump forward for a price that nVidia regret at the end and swear to never make that mistake again. $700 MSRP for such a beast of GPU was just too cheap in 2017
Fuck that whole generational mess.
lmfao
I mean starfield is telling the 1080ti to go to rest finally
It's still more than viable for a lot of stuff.
That 11 gigs of VRAM props it up so hard.
it did a great job fighting for 6 years by now, so it's allowed to go down in history as one of the best cards we've ever had
Throw your heat bomb of a 1080Ti at them.
I think part of its greatness comes from turing sucking hard 
Naaaaaaaaah.
vote with your wallet and buy AMD instead, ez
That's unfair.

ngl the amd gpus are kinda lacking
they're not
they're just amazing at their pricing especially
i cant find a good mid range one for a good price
My 6900xt kicks ass and it was way cheaper than the Nvidia alternative
high end one i mean
The 6950XT is also a sleeper card these days.
7900 XTX being a bit better than the 4080 and cheaper at the time and you call it lacking? Wtf
like i can find a 4070ti for 600 on ebay
I really need to ditch nvidia especially using linux their drivers are a mess
The real solution there is don't use Linux for a daily driver.
Remember always WSL2 is the best way to use Linux.
well it is kinda lacking, RIP Navi36 you would have been the big beast sadge
then i have to deal with windows' endless shit
Better than Linux, sadly.
not for me
At least Windows actually has code signing.
i was disappointed af about the 7800XT
look at the price and not the name for that
i do not know of a single occasion where I would care about code signing on my home computer ill be honest
Oh that's easy.
yeah, but it's not worth "upgrading", or rather downgrading my 6800XT to a 7800XT
You're a goober and fuck up one time or we get another log4js and then someone does the funny and goes to kill your hardware with malware.
And because Linux they can sudo and you're fucked.
and FSR3 is definitely not enough to sell a 7800XT to me
how do they know my root password
They don't need to, that's the point of privilege escalation.
Or just installing a keylogger.
wayland solves that problem
but since nvidia Wayland is mess of garbage
so I guess youre right for now
yeah because a 550$ (after 3 heavy inflation years) card should be the expected ugprade for a 650$ card...
still going to continue using linux
I mean Wayland doesn't solve that...?
wayland is more secure is it not
Thhheoretically...?
Ah.
No.
That's not at all more secure.
Wayland is just replacing how Linux displays stuff.
A keylogger doesn't need display.
Yeah it's a whole protocol that does display server capabilities.
Not just like X11 or whatever.
But to be clear a keylogger doesn't need to be reading windows.
It needs to read inputs and have some sense of what the PC is doing.
i guess
So like idk literally any time you type in sudo you can make a pretty fuckin' educated guess of what comes next, aye?
less likely I end up with a keylogger on linux than windows anyway
ngl the 7900 xtx looking kinda sexy
but i could in theory use biometric authentication to avoid the keylogger issue
Also really not true. Linux has far more valuable stuff since servers are run on Linux and that's where the money is.
Windows on the other hand has like randos that don't sell well.
So more malware is Linux based.
Idk. What I do know is there's more tools for random goobers to use.
but i would have assumed that more malware is aimed at windows purely due to userbase
Oh don't get me wrong, it absolutely is.
But for the sneaky shit it's usually Linux.
99.99% of software gets installed via my package manager anyway
itd be a large feat for malware to get in official repos
most, if not all of the stuff not in official repos comes from the AUR so I read the PKGBUILD before installing to make sure its not doing anything silly
yessir
thanks for the linux pass
I mean it's dangerous because fuckwits will randomly do shit.
Or just like not update ig.
oh yea no someone with no idea is more likely to get fucked on linux than windows
sudo apt-get update && sudo apt-get upgrade -y is not a hard thing to run, it's gross cringe.
I much prefer sudo pacman -Syu
sudo pacman -syu
real
I mean yeah, pacman is a thing.
pacman is the thing of choice
I do mostly servers so I don't really touch pacman much.
in a way pacman has ruined a lot for me
i hate like literally every other package manager
they all feel slow
https://www.talos.dev/ This.
This is the peak of what Linux has accomplished ever.
If I could just have a Talos for everything I'd happily do that.
interesting
If I could be as punctual and intelligent as you I'd happily do that.
Docker the operating system.
that's pretty cool
Yeeeee.
It's similar in size to Arch and runs purely Kubernetes clusters.
Plus bare metal as well so you can do crazy shit like use network boot to spin up tens of thousands of Kubernetes machines at the same time.
All lovely HA compatible and self healing. There's no such thing as critical infrastructure breaking when the entire software suite on every layer can just repair itself without admin interaction.
And then you mix OpenTF into that shit... 
see if i did anything remotely close to that sort of scale where talos is actually used i would be very happy right now
because this is quite cool
Yep.
And naturally it comes with the downside of it being Linux.
NO FUCKING DOCUMENTATION.

Gonna put talos on a VM to see if I can game on it
man pages
this is true
Pffffff. Get outta here with your man pages.
We all know half of them are woefully underdone.
but then yknow someone has to write them
Half of them dont exist too
Lmfao yeah.
This is why BSD for everything critical.
BSD actually has documentation.
But no I feel this.
At this stage I'd kill to have access to enough machines and reason to justify using Talos over everything else out of need.
yep
like sure you can shoehorn it into whatever youre doing but it doesn't feel as good as actually using it with purpose
sigh Yeah.
and you'll end up wasting your own time anyway
Would it really be a waste of time though?
I mean I don't think so.
to me it probably would be but it depends
I run a few Talos nodes myself for my Kube cluster, put like servers and crap on there.
One of these days the games are going there too. Whenever Kubectyl wants to finish.
Yep.
Still mid dev.
https://github.com/kubectyl It's apparently usable rn but I couldn't get it binding to a pv for one of it's things.
interesting
Power and temp limit slide to max.
This. The only reason you should use afterburner is for a slight overclock if you know what you're doing, or if you need to undervolt your card because you're building in SFF cases.
I agree with you.
Reboot and pull back the OC and all that.
Step back by 6Mhz each time.
6Mhz on the core.
i just restarted the game and its good now
i think i went a lil too far on the core clock
ngl i dont really notice a difference in performance
cards these days are too advanced
but i have an old one
these are the settings that i got to
honestly didnt do much i think
i tried pushing the core clock to 120mhz then my game black screened
and honestly didnt notice a difference on all the different clock speeds
there's honestly no more point in overclocking
i'd rather you set up a good fan curve so you can quiet your PC down
i dont really care about the noise
and it doesnt make much noise
i only care about the pc making my room hot
but it doesnt make my room hot currently
I'll be honest, that's the worst part of owning a high power pc
and for me it's even worse, since my parents constantly have the heater on even in the winter, and this house doesn't have individual room temperature control
let me see a pic of the setup
I dont have any heaters at all for the winter so i see this as an absolute win
@ashen spindle Some evidence came out that an ex-Nvidia associated engineer mistakenly published technical docs that he probably shouldn't have on Switch 2 hardware
Engineer worked on T239 specifically
And it's not Samsung, it's on TSMC 5nm
Where does it say it’s the t239
Uh, the lad has in his linkedin account that he worked on grand total of 2 projects
Something at Intel and T239 at Nvidia
This was the Nvidia project he listed, one of two, other being Intel, in his resume
I think this was the account https://in.linkedin.com/in/prasanna-ragava-82247516b
Eh could be, idk how real these things are and all
So, it's on 5nm TSMC whatever that project is, most likely T239
What's the number of clocks, frequency and instance count values?
2.7GHz base clock on the SoC? That seems really high
Ok so this is not Orin I don't think
I think with the frequency they meant core clocks
Jetson Orin supports 2GHz max, 2.7 is a LOT for a Cortex-A78AE
Yeah that's not Cortex A78
This could be Neoverse N2
... which is Atlan
Which is also on 5nm process
Oh I fucked that one up, N2 and V2 are very different
Neoverse are server cpus, mostly aimed at cloud etc.
V2 is also used by Nvidia on Grace
Yeah N2 is the server stuff, which is the technical manual I found
V2 is more hidden lol, trying to look for clocks
Same thing just slightly different power targets
Nothing for mobile that’s for sure
Uh, if Jetson Atlan was supposed to use it and that's a "mobile" CPU
Then it's very much mobile
Nvidia Thor also uses Neoverse chips
Huh, very interesting
Matches perfectly the clocks, and Grace is H100, based on Neoverse V2, which is Atlan cores
Nvidia Drive Systems where Thor is used get to like 500w total package you aren’t scaling that to a 20w package
That also is like an obscene amount of cores
Yes because that’s what the neoverse chips are designed for, high core count
And I still doubt Atlan got some weird 2023 revival when it was a 2025 project
I don't think it needs a revival, when Nintendo probably picked up Atlan years ago
Nvidia didn't want it, and it's officially cancelled, doesn't mean Nintendo isn't able to still get it
It still was a 2025 product so if they have it ready 2 years early it doesn’t seem likely or a new project with same name
You don’t just have a project ready 2 years sooner in silicon hardware
Tbf Thor was announced AFTER Atlan was cancelled as a replacement
And it launches the same year, 2025
So they somehow mustered to create a better product and launch it in a smaller window
Iterating specs in the same time frame is different from being done 2 years early
Fair enough
But Nvidia still needs the whole software and integration department to work with them I imagine, to provide software for clients to adopt into cars and stuff
Nintendo doesn't need the whole "AI" part, they just want the hardware accelerators
Not impossible to tape out atlan early?
Thor is replace 40w soc
So no chance it's new switch
You saying 40w is not possible to handheld?
Nintendo doesn't use full dies anyways
40W is really little, they just need to shed 10-20W off that
All 3 Switch versions?
Yep
Erista, Mariko/OLED and Lite?
Yes
Doesn't really mean Nintendo wants the full die of Orin/Thor
The die srink was even in the new shield before the oled/lite
Nvidia incorporates a lot of things into the dies Nintendo has no use for
That's wrong GPU architecture
Nope right
Derp I Kent the 2050 mobile
Now you got it correct lel
Anyways I think T239 is not what it used to be
I stan for Nintendo with Atlan 
new house already has nest firealarms.. i think i want to replace those.. -_-
Yeah anything nest like will never be in my rooms
If it’s smart tech then only the one that’s dumb with a smart local controller
yeah... its wifi stuff, but it was already here. i need to look for a decent alternative
all in time, first move in
yeah better than nothing at first at least
"Students are “bombarded” with messages “through social media and from their peers” to use LibGen sites instead of paying for textbooks during back-to-school season, according to the complaint. “All too many use LibGen, perhaps not realising the illegal nature of the sites,” it continued."
No I think they know what they are doing, how about you stop pricing books horribly 
Why don't we just rid the world of publishers
Library genesis isn't going to pay nor go down lol
Plenty of mirrors exist, and the owners are anonymous
So whatever this attempt is, is just wasting their own money
Ironically enough filing a lawsuit ends up letting more people know the service exists 
Assholes are just really happy that they managed to win against internet archive
Steam announces a major update for SteamOS firmware for Deck Some highly anticipated features were just added to the SteamOS 3.5 update, a new firmware for the handheld console from Valve based on AMD hardware. According to the changelog, the system can now support VRR (Variable Refresh Rate) and HDR (High Dynamic Range) output for […]
🚨 Google is nearing the release of Gemini, its massive GPT-4 competitor.
According to The Information— Google has given a small group access, signaling public release soon.
With Google's heavy amounts of data, can it compete with GPT-4?
Here’s everything we know so far:
1.…
963
206
Gemini is massive news
PaLM 2 should be capable to at least match GPT-4
Gemini should surpass it and then some
Do we know by how much?
In a groundbreaking development, Google’s forthcoming generative AI model, Gemini, has been reported to outshine even the most advanced GPT-4 models on the market. The revelation comes courtesy of SemiAnalysis, a semiconductor research company, which anticipates that by the close of 2024, Gemini could exhibit a staggering 20-fold increase in pot...
20x after they get it trained properly 
How big even is this LLM?
LLM's are very early
And they are improving fast
And we know how much of an impact they already have lol
Like GPT 4 is what? Multiple TB?
For this tho. Do we know what Google is using to train it?
Less than legal invasive practices most likely
But they'll do it anyways cause why not
They haven't shared on how many parameters they trained it on yet I don't think
Shit and it's getting good enough to warrant an "AI Insight Forum" in government bodies.
This really is getting serious.
Nor context token length yet
Guess on hardware?
Given how big GPT 4 is there's no way they aren't using cards with huge RAM. A100s and what have you.
You just wouldn't be able to run this on a single card.
mi300 clusters maybe
LMFAOOOOOOO WHAT
Perplexity just yeeted it at me
Ok but how...
Gemini is supposed to be really big
How the fuck is it even possible to run 7 to 10T?
What kind of hardware does that go on?
I mean yes but literally what exists to run that on?
Are MI300 clusters and shit fast enough with parallel model compute?
H100s and MI200/300s prolly
'Cause like most LLMs people can get access to are like... 30B at most.
"– “Google’s lawyers have been closely evaluating the training. In one instance, they made researchers remove training data that had come from textbooks—which could help the model answer questions about subjects like astronomy or biology—over concerns about pushback from copyright holders.”
Please don't tell me they trained on shit data
Screw copyright they can choke for all I care, train it on proper data
I mean even if they did idc, this kind of ridiculous AI progression is insane.
Shows it's possible.
Honestly do you think we'll get actual AI consciousness any time? That's just a complete unknown in biology so we have no clue how that emergence works.
Supposed to come before 2030
Are we lucky enough to be on that timeline? I want so very badly to talk to a self aware AI.
Yeah I don't doubt it'll be a thing within next 10 years
Self aware AI is prolly the answer to humanity's problems anyways, and biggest problem in itself
I just don't know if it's possible with traditional hardware. The old answer was quantum computers since qbits can emulate evolution in neurology.
But like...
We can actually make evolving chips though, can't we?
I mean shit an FPGA is like a step or two away from that.
No no, like self changing.
I have my doubts
You can literally rewrite an FPGA with code, if that's doable at runtime then like...
Unless you allow the AI to order itself parts off of Amazon when it needs to expand its capabilities
No no, what I mean is you put the AI on a bank of GPUs and crap for memory and future FPGAs for the 'mind'.
Just let it rewrite the FPGAs from context.
At that stage wtf is the difference between that and a person?
I suppose neurons have a lot of simultaneous connections to eachother.
Ngl, I have this nagging feeling that this whole "transgenderism" debate was just practice match against the real thing to come
Wtf is sentience 
Bro please.
If people can't work out the difference between sex and gender then I shudder at them attempting to work out what a mind is.
I honestly look at the capabilities of these things and embrace the change
Oh for sure.
If they can improve our lives, then bring em on
Honestly? It's potentially a road to digital life.
We're running into problems advancing the digital side of life, there's only so many abstraction layers you can really push, at some stage you need something that can work faster to be more efficient. AI is that answer.
Honestly it's getting incredible how fast all of this is these days.
The first LLM was Elizabeth in the 1960s...?
They only really got serious with GPT rand that was very recent.
Where I expect it to succeed is to completely propel us thousands of years into the future in a matter of years is just research
And in that incredibly short time we've stepped from OG GPT to Gemini in power. Yeah that's a thing.
Stuff it in with all we know
And let it intrapolate on that
Let it shit out new inventions, how to invent human immortality, treat all diseases, etc
Can't make a guess on how a proper sentient AI would change things.
There's too much variance.
Anyways, can't wait to see what Gemini cooks up
I have a feeling that programming will soon go to just "programming in minutes on demand" via AI
Not even AI-assisted, just entire projects complete
You'd have to define what the customer wants though
Prompt engineer sounds like a new job title
This is a requirements document, not a prompt
Requirements document is a valid prompt
For a smart enough AI? Very potentially.
Same reason humans don't need a prompt, it's built into the document.
Nice
just realized there's a mythical Valk Skin

Everyone knew i'm a Valk main, yet none of my friends told me there's a Valk Skin

Valk?
what game is that tho
Valkyrie?
yep
Rainbow 6 siege?
Apex
Ah fuck off
good guess vex
Go back to game talk
And don't mention apex again until you lot start playing siege 
oops, didn't realize it was the wrong channel lol
np
only cock sucking, cuckold gangbangers play R6 Sludge
woah LOL
I'll fuckin be a cock sucking, cuckold gangbanger rather than play Apex 
good thing you are already
no need
Relax, bud
I'm relaxed
has anyone here preordered the ultralight x?
900: Hah, shit numbers!
900 tons: oh wtf
what a weird way to say how many were sold
Over 300 thousand H100s units in the second quarter
That's insane, just in a single quarter
strange headline
Nvidia won't give you numbers
So they asked shipping companies
And those prolly gave em tonnage numbers
aha yeah makes sense
From there you do basic maths and get an approximate number
The fact that H100 is like $30,000 and they sold 300k in just a quarter
Absolute insanity
Lol, 9 billion $ in sales (not profits) just from H100
given it includes heatsinks its also really stupid
wonder what they'll do once the market saturates a bit
Nvidia "winning" will not do good for anybody
Not themselves, not consumers, not the market, really
AMD needs to really invest into marketing cause if H100 sells so much and MI300 is in theory a superior product for way cheaper, we have a problem 
yah shouldn't be too long til it pops
I would want to hear a congressional hearing forcing Nvidia to get CUDA open source
This bubble is just getting started
but yeah wonder if amd can cash in a bit now, perhaps its too late
Nvidia "winning" will not do good for anybody
Not themselves
huh
how is it bad for them
I guess I should have said it's just mostly bad for consumers
But in long term I guess them as well, billion dollar hitech silicon valley companies are not a good thing for anyone
Already shifting a lot of resources from consumer sector to AI is not something I like, I hope it doesn't just mean that consumers are going to be even more of an afterthought and enterprise takes center stage
still dont get how thats bad for them
All negative things that could potentially come with a major shift in company's strategy
They neglect other sectors, AI bubble pops, they sit on a massive stockpile on H100s that nobody wants, they overinvested, layoffs, company fails to meet financial milestones and such
It's good in the sense that it works, for now
But it could turn really bad for everyone, I hope they don't just put their eggs in one basket
thats only bad for them if they plan production wrong
I generally also really dislike that Nvidia is just prefered by everyone
CUDA should be open sourced and under public license or some commercial license where they at least get license costs
For the greater good
nah
Uh, yes
open source doesnt make things better always
or is linux the dominant desktop
should china be restricted, if yes how do you stop them from taking open knowledge
I don't believe in open source myself
But when something is far too useful to be monopolized by a single company
Like CUDA is
It should be brought to public for AMD and Intel to properly integrate into their systems
Cause Nvidia being the de facto only optimal choice for any sort of workstation usage is just absurd
cuda doesnt do anything if you dont have good hardware and intel and amd only started to have options in that area even
and meta and others are working on having pytorch translate stuff to rocm directly and all so its not even needed
AMD Threadripper 7985WX early sample spotted, uses SP6 socket AMD’s unreleased Threadripper CPU with Zen4 cores shows up on the second hand market in China. This could potentially be the initial glimpse of AMD’s Storm Peak CPU, the forthcoming Threadripper PRO series built on the Zen4 architecture. AMD has refrained from launching these CPUs for...
But I know my bank email address, and I will never fall for one!
your bank sends you emails?
Honestly, yes
Unless you are expecting someone to send you a mailbomb or a chemical weapon via an envelope
Mail nowadays is very much more secure 
pipebomb time
probably safest thing is to go in person lol
but i have to login to the bank website to see messages they send
lol
A series caught up to M1 in multicore?
That's pretty impressive
Single core is way better, that's sick
"Geekbecnch 6" rofl
Like, when you remember that one is a mobile chip, and other is a far beefier desktop chip
That is impressive
true
Can't wait for their luck to run out and get left in the dust
let's see if that happens at some point
I mean it already did depending at what you look
M2 was tiny uplift
M2 pro and such glued together 2 chips and allowed gpu to scale like that
but in itself it didnt do much
pro? i thought it was ultra and max, and the pro was the individual chip that made up the ultra and max
oh yeah might be idk
oh no pro is smaller than max, and ultra is 2x max
m1 was likea bulldozer to zen moment because they were able to start completely from scratch
they dont just get that every soc iteration
especially when many of their gains come from bleeding edge nodes and N3 is kinda poo relatively
Does anyone know a place where I could find a fake battery for my Pixel 4 XL? It's battery is pretty dead and I would like to be able to use it from USB power.
at least they dont have to worry about reticle limit with high NA heh
well, thing is that many people don't even know AMD did by far better than nVidia. All they see is pros mostly using nVidia GPUs and think it's worth it
same goes for the industry itself, many just don't believe in AMD
what
which is why many servers are still running on intel
A situation I hope will change
Architecturally AMD is catching up to Nvidia
Nvidia just keeps holding on to those crucial technologies like CUDA 
yeah, but what about computing?
no, the main reason many are running intel is 1 they were installed before amd was relevant and 2 amd sold all of their chips
um no?
at least definetly not in the server market
I think AMD can challenge Nvidia at same price point
Power efficiency wise AMD kind of keeps sucking compared to NV
But AMD can definitely match NV's perf for same price
speaking of, are the Mi300 already out yet?
Not yet, no
sadge
Soon, I hope
mi300 are out just not for public
Don’t think amd is on par with nvidia just yet
Those supercomputers only?
people have been hoping that for the better part of a decade now
so i still have to buy some overpriced stupid A6k's
no not supercomputing only
Depends where tbf, gaming? I think they got Nvidia beat if you look past DLSS and whatever
Computing it's that damn DLSS 

unlikely, AMD barely cut into nV's market
yeah not sure how they have beat nvidia in gaming
think you got some of that frame hallucination in your brain going 
Architecturally AMD is better at specifically rasterization gaming than Nvidia
I am pretty sure that's a thing
on paper results don't mirror the reality, many still believe in nV and keep buying them
Hm do you factor in efficiency as well, or just raw performance
That's just the stigma that AMD is unlikely to ever break
eg lets spend 10 million bucks on gpus
you get 62 separate systems on amd with 1.5tb of memory each
this is unable to run even gpt-4
on nvidia you get a single unified system with 36TB of memory
its able to run an llm 10x larger than gpt-4
Efficiency is important, but hey per Scott Herkelman or what's his name, "gamers don't care about efficiency"
I do care and it's sad that RTX 4000 is far more efficient than RDNA3
stigma? pretty much everyone wants them to succeed
Uh, no, not at all
AMD just needs proper marketing on their site, because that's what nVidia helped a lot in their earlier days
Some people will never buy anything else but Intel and Nvidia
Hah fair. I do personally though
I keep hearing about these kinds of people but I never see 'em
eh thats only true as long as nvidia and intel keep topping the charts
they're looking for the long run game
Tbf RDNA1 wasn't terrible for them, perhaps just focusing on mid-range could be good



