[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
▼ Settings Mobile Home
4chan
/pol/ - Politically Incorrect


Thread archived.
You cannot reply anymore.




File: stonking.jpg (282 KB, 1634x1130)
282 KB JPG
computer part manufacturers themselves are losing stock profits

AI's are predominantly software organzations

but this is hardware:
CPUs
SSD storage

its all going down
>>
don't question investors. They'll just run AI in their imaginations
>>
>>545246950
The stock price going down doesnt mean the component price will
>>
When the bubble pops, I'm building a PC with 1Tb of RAM and 10Tb of SSD storage, just for the lols.
>>
>>545246950
Components are at all time high.
>>
Delusional inbred incels
Stay poor forever
>>
Two more datacenters and they're profitable.
>>
>>545247398
Building your own PC is not coming back. Modern integrated architectures have high-bandwidth memory on chip.
Maybe they make a build-a-bear like workshop for PC boomers where you can choose the color of the housing or something.
>>
>>545247649
>Modem integrated
Like what? You know Intel and AMD servers aren't running integrated shit they are running xenon's and epyc
>>
Fuck em

I hope their hubris costs them a lot and the bubble keeps bursting

These fucking datacenter prices are obnoxious
>>
>>545247741
Using separate RAM for CPU and GPU is just wasteful and it's going away.
>>
The last big AI value added I read Starbucks is rolling out AI customer identification to tag everyone who enters their stores and change menus based on who you are. Something no one wants.

It's a bubble.
>>
>>545247843
GPU and CPU memory is different, unified memory works on slow ass GPUs and is bottlenecking its performance.
>>
>>545247649
>delusional begging
Fucking ipad faggots.
>>
>>545246950
Will this make storage and ram cheapers?
>>
>>545247398
thats not how it works. data centers do not use the memory you use. they use the same manufacturing plants. so even if those data centers dump their hardware, none of it will be of any use to you whatsoever. it's not even DDR5. it's HBM3E and other types of high-bandwidth memory that's soldered directly onto the package next to the GPU core. it can't be repurposed and it's utterly useless for normie use.
>>
>>545248234
>you can't kludge a server into a PC
Interesting. You appear to be disabled.
>>
Trump may lose a lot of support in this election. Investors aren't sure what Trump will do, and they aren't sure what the new majority will do. And so they're uncertain. Not surprising the market is dipping.
>>
>>545248234
Okay, but once AI kikes stop buying the whole stock of chips the supply should still increase.
>>
>>545248150
>unified memory works on slow ass GPUs and is bottlenecking its performance.
Tell this to apple and nvidia. They need to hear about this before producing more slow ass, bottlenecked computers.
>>
>>545246950
>>545246989
>>545247398
>computer part manufacturers themselves are losing stock profits
Their profits are actually up another 25% this quarter based on the earnings report they just released, with a projection for profits to double in 2027 and 2028. 1.5 trillion in AI spending up from 850 billion in 2026.

The selloff is because the market is totally irrationally bearish while the companies themselves are profiting massively.

You will never afford your gay little computer and that's a good thing. Get a real hobby, /v/turd.
>>
>>545247398
Prices are never going down. Just like GPU prices never came down after mining boom.
>>
>>545246950
Jeetu Patel is on it he has the solution
>>
>>545248346
Okay, let's ask Nvidia.
Dedicat
>By using explicit prefetching (cudaMemPrefetchAsync), speeds double to 10.9 GB/s, putting it nearly on par with standard cudaMemcpyAsync (11.4 GB/s).
Unified
In integrated platforms like the Microsoft Surface Laptop Ultra featuring RTX Spark, this unified LPDDR5X memory architecture shares up to 300 GB/s of bandwidth across a 128 GB memory pool.

Oh look, Nvidia agrees with me, who could've thought
>>
>>545248738
So you gonna desolder those RAM chips for your server?
>>
>>545249334
Servers use dedicated GPUs
>>
>>545246950
it's going down because cxmt and other chinese companies are ramping up production and make a fortune.
>>
>>545247649
>Modern integrated architectures have high-bandwidth memory on chip.
you mean... a cache?

or did you mean to say "on die"?
>>
File: no_ram_sticks.jpg (103 KB, 1294x942)
103 KB JPG
>>545249388
>>545249519
Please tell me where to find the removable RAM sticks.
>>
>>545246950
stockeja these nuts
>>
>>545247649
Even Intel's system on a chip still has separate memory.
>>
File: 1000017408.jpg (144 KB, 1600x1491)
144 KB JPG
>year 5 of ai craze
>ai still only good for local hentai image gens
>>
>>545249435
Not nearly enough unfortunately. They'd need to 4x their projected wafers to really make a dent in the next couple years.
>>
>>545246950
The entire system was found to be self-injecting money for profit like bitcoin scammers did. Eventually that funneling stops working eithout fresh money, and AI has failed to generate anything but 4 faggots writing prompts for free.
Their last beach was “you need to buy AI to protect you from AI” but that only made people see the obvious retardness of that statement.
All that is left is to sit down and enjoy America self-imploding as mutts will bailout them all with taxpayers money and start another recession when that bailoit does not return because both the lack of profit and recession.
And the funniest part…you faggots did this before with .net. You never fucking learn.
>>
>>545248425
>Profits are up because of artificial scarcity
>Reinvest it all in the thing that generated none
Nothing is more retard than a mutt investor.
>>
>>545247649
Personal computing with custom parts isn't gong anywhere.
Modern industries rely too heavily on it.
Imagine having to connect to a datacenter every time you want to edit some video or make some 3D asset.
It would be incredibly insecure for one and introduce another point of failure and latency. No industry would willingly adopt it, and if they do people would just migrate to solutions that offer them more control over their hardware and software.
It's like forcing trains on a country of people used to owning personal vehicles.
>>
>>545249738
>Even Intel
ok boomer

>>545249952
>Modern industries rely too heavily on it.
No they don't. We have seen consumer brands closing. It's all AI now - hence why people talk about an AI bubble.

You can get a complete GB10 chip system with 128GB of unified memory for 5000 bucks.
There's nothing comparable you can build yourself for that price.
>>
>>545248346
They're building slow bottlenecked computers because they want you to be paying them for the entire thing. They dont like the idea of you buying RAM from someone that isnt NVIDIA so now they are putting 200$ of RAM inside the GPU and charging you an extra 500$ for it, and its all or nothing.
And by "you" I mean infinite money fountain AI retards who are buying these chips by the thousand without looking twice at them. Actual consumers would reject the trend but right now we are in the computer equivalent of the southern US Real Estate market where they're just pushing million dollar mcmansions that are missing basic shit like forgetting to blueprint a house with sinks in the bathrooms, because they are selling them sight unseen to people from across the country,
>>
>>545249667
My guy, that is a dedicated ram on GPU and CPU this isn't unified memory. Are you retarded?
>>
>>545248425
Their valuations are absurd even with increased profits and despite what ever they say the profits will not keep increasing because AI hyperscalers are running out of money.
>>
>>545246950
The AI shitfest best bullshits are;
- You will be replaced but no worry everyone will live like a millionaire ... because and just.
- The mass will have no income to buy the products but that's no problem because the corps never wanted to sell to the public.
- Profit will be high and cemented even if it's openly admitted they don't care with the consumers and the demands.
- It's sustainable business model even if all of them on government infusions.
- Etc
>>
>>545250313
>that is a dedicated ram on GPU and CPU
yes but you still didn't tell me how you want to use that RAM to build your own server.
>>545250313
>this isn't unified memory
It is retard. Both CPU and GPU can access the same memory without making copies.
>>
>>545250742
Unified means you have one RAM for GPU and CPU.
That is HBM ram on GPU and DDR on CPU.
Basically you don't understand what you are talking about and are arguing over soldered/unsoldered not understanding that soldered == unified
>>
>>545248425
>market is totally irrational
kek. the things baggies say
>>
>>545250896
Alright if logically unified memory is not enough for you then show me the RAM sticks you are going to use for your server on this physically unified architecture.
>>
>>545251033
You keep using words but the words you use don't mean what you think they mean. That's where your confusion arises.
>>
>>545251033
>It functions as a dedicated GPU accelerator rather than a hybrid CPU+GPU APU, so it does not merge CPU and GPU domains into a single unified system memory pool in the same manner as the MI300A, though it provides a massive local HBM3 capacity for large AI models.

Yeah it's GPU, it does have unified memory but it's design is not to be used on CPU
>>
>>545251178
>it's design is not to be used on CPU
Because it literally is a CPU.
>>
>>545251278
>AMD CDNA™ (Compute DNA) is a datacenter-focused GPU microarchitecture built by AMD for high-performance computing (HPC) and artificial intelligence (AI) workloads
Brother, just use Google.
>>
>>545251419
Idk what you want from me.
>>
>>545251460
Literally says it's CPU and GPU unified and the design dock says it's designed to be used as GPU rather than CPU and CPU is for support work.
Not that ducking hard to grasp is it?
>>
>>545247301
It will because gamers are unrionically their lifeboat
>>
>>545251460
Unified memory for inference not for training models. My company has the same problem we make NPUs but can't train models on them. Worthless little fuckers
>>
>>545251536
You're arguing with an idiot. If you stop now, your life will improve.
>>
>>545251572
there are no games in 2020s

(except old ones)
>>
>>545248150
No. Maybe in the high end segment. For 1080p medium settings an integrated GPU, like 780m or the 8060s is running fine with ddr5 6400. Low to entry medium rigs gonna be 10x10 cm big gaming PCs and running every emu for any console. Use Linux and youre fine with 32 GB RAM
>>
And low Power consumption, 50-100watts
>>
>>545247649
Yes, it is. Their dumb terminal model will never work. They've gone all in on AI far too early, and it's about to blow up in their (our) faces. The public perception of AI before any major disasters is already bad, and it won't get any better when the disasters occur.
>>
>>545247398
Same, and with a 5070. Just to launch youtube and idgaf.
>>
>>545251536
>>545251731
I'm sure your server with outdated RAM sticks will be a great electric heater. Good luck with that.
>>
>>545251851
My server is running on unified memory, literally just using the CPU, GPU is only used to output nothing else. Different servers require different hardware. I know, a revolutionary idea.
>>
>>545248346
>Tell this to apple and nvidia. They need to hear about this before producing more slow ass, bottlenecked computers.
gotta admit the M series SoC procs Macs agree
>>
>>545246950
Thus is more /g/ stuff but I'll allow it,
>>
>>545247398
Anon,

This is the old timey computer crash. AI is just how it is done now, and you turned into an old man.
>>
>>545252062
Share some details about your setup, anon.
>>
File: Hubris.mp4 (409 KB, 1034x738)
409 KB
409 KB MP4
An awful lot of people in this thread alone seem to be pretty confident they know what's going to happen.
However I think this is just hubris.
None of you know shit.
>AI is just how its done now
Blow it out your ass
We don't have the power production necessary for everything to run this way
Not even close
>Just buy more power
When you think this way, that's how I know you're retarded
I've seen this exact argument countless times
None of you have a fucking clue
>>
>>545246950
AMD and Nvidia are bricking existing GPU with their stupid patches.
>>
>>545257395
>bricking existing GPU with their stupid patches
said by the stupid outsourced DEI "programmer" who AI shitted out those patches.
>>
>>545246950
They literally withheld production to maximize prices since they knew datacenters and statefunded techniggers will buy them fucking the consumer market almost beyond all repair suffer no dipshit ceo cries.
>>
>>545248234
It will actually be good for people like me but I'm an edge case homelabber who will run a mini data center in my garage
>>
>>545246950
All of those stocks are still up on the year lmao. They were tanked because they’re filtering into AI and announced lower hardware sales. I predict a rebound Monday, per usual, because redposters are never right.
>>
>>545248425
You subhuman israeli rats are really pissy about humans resistance to AI. Kill yourself.
>>
>>545257697
It’s called a correction. It is when the price went up too much before and now you have to lower it to a more sensible one.
>>
>>545258175
Because of an earnings call that the market will forget in a week. First time?
>>
>>545253996
It's just a low profile Fujitsu esprimo working as NAS and local network manager, nothing too fancy.
>>
>>545246950
>AI invests 1 trillion dollars only in America in 2026
>somehow this still happens after the botched IPOs

This is catastrophic.

It is running the script of the railway crash of the 1870s.

Also,

>>545248425
Dummy, they are using capital to sell tech to AI companies rather than produce something to sell it to customers.
BTC
bc1qtm4rep7kyp2p28hlkfktpnd0j7uew47vnt2mev
LTC
ltc1q9xlaysse0mpfwf025esd69kt9zu3md22gh23du
ETH
0x6abA18D203520c8F90ca63eE7426a7bC345D98e6
SOL
E5ekoiauCR8yiNp451dmuPzhUxo84P5x3m9WJBK5kxZe
XRP
r3jYXNdMk2Vz5hXhKCKE9re8uz93JQFG1r
DOGE
DSBq5jRWkiwmEh1G7zBwmQDgRkvRr61fmb
XMR
85CTwq5wGjWZw5GgzgwZPrHyZxpLXMK5oZSSSk5j4x9w1bpqzisaWAUZYaAPfh6rP2MZJcaFqF1XmEzkw6YCqJSS2Txu6Z8




Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.