631
Motherboard sales are now collapsing amid unprecedented shortages fueled by AI (www.tomshardware.com)
submitted 6 months ago by to c/technology@lemmy.world
141 comments fedilink hide all child comments
 

Comments

top 98 comments

sorted by: hot top controversial new old
[–] 317 points 6 months ago (41 children)

AI is an amazing tool for fascists.

Annihilate private access to computing, censor and rewrite all comms, destroy free software and the last remnants of education...

Every single decision made for evil.

And all these vendors who are locking themselves into one customer are about to learn why that's a bad idea.

  • permalink
  • fedilink
  • source
  • hideshow 41 child comments
    •
  • [–] 73 points 6 months ago (2 children)

    The worst thing is that when used for good AI is fantastic! Scientific progress with purpose built AI to find planets, predict the weather, and tons of pattern matching has been in use for decades with positive benefits!

    Even LLMs can be a useful tool in the right situations where looking like words people would say but accuracy is NOT important.

    The problem is trying to use LLMs to do everything and failing while running the tech industry, the environment, and soon the economy into the ground. They took something positive, ruined it and coopted the terminology while shoving it down everyone's throats.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 3 points 6 months ago

    AI is used as propaganda tools to spread it, and it can bee seen on youtube, social media quite readily. plus it sexualizes victims to like csam, and festishized unattainable "women" for conservatives.

    conservaties pretty much buys into/believes in anything that is scammy.

  • permalink
  • fedilink
  • source
  • parent
  • [–] -2 points 6 months ago (32 children)

    I'm not here to argue for or against LLMs in general, but self-hostable AI is a thing. Actually open AI is a thing.

    A blanket statement saying about AI as a whole technology being good for fascism is about as useful as saying "roads are good for fascism" (they're great for troop movement after all).

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 32 child comments
    •
  • [–] 103 points 6 months ago (22 children)

    You can't self host anything when the hardware is no longer affordable.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 22 child comments
    •
  • [–] 10 points 6 months ago (16 children)

    Self hosting an llm ain't the same thing as self hosting nextcloud for your docs and calendar. Yes there are small models but their output is laughable

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 16 child comments
    •
  • [–] 27 points 6 months ago* (15 children)

    Small models are improving and becoming more capable. The quality of local LLMs is basically unbounded. The context size of local LLMs is bounded by hardware. So local LLMs can be very capable for small, self-contained tasks.

    qwen 3.6 35b running locally:

    Write a Python script that can pull weather data from public sources and provide the high and low temperature for the current day in Miami, FL.

    Single shot. No tool/internet use, so it didn't pull this script from elsewhere.

    import requests

def get_miami_weather():
    # Miami, FL coordinates
    LATITUDE = 25.7617
    LONGITUDE = -80.1918

    # Open-Meteo API URL (free, no API key required)
    url = (
        f"https://api.open-meteo.com/v1/forecast?"
        f"latitude={LATITUDE}&longitude={LONGITUDE}"
        f"&daily=temperature_2m_max,temperature_2m_min"
        f"&timezone=auto"
    )

    try:
        response = requests.get(url, timeout=10)
        response.raise_for_status()  # Raises error for 4xx/5xx HTTP status codes
        data = response.json()

        # Index 0 corresponds to the current day
        high_c = data["daily"]["temperature_2m_max"][0]
        low_c = data["daily"]["temperature_2m_min"][0]

        # Convert to Fahrenheit (commonly used in the US)
        high_f = (high_c * 9/5) + 32
        low_f = (low_c * 9/5) + 32

        print("🌤️  Miami, FL Weather for Today:")
        print(f"High: {high_f:.1f}°F ({high_c:.1f}°C)")
        print(f"Low:  {low_f:.1f}°F ({low_c:.1f}°C)")

    except requests.exceptions.HTTPError as http_err:
        print(f"❌ HTTP error occurred: {http_err}")
    except requests.exceptions.ConnectionError:
        print("❌ Error: Could not connect to the weather API.")
    except requests.exceptions.Timeout:
        print("❌ Error: Request timed out.")
    except requests.exceptions.RequestException as err:
        print(f"❌ An error occurred: {err}")
    except KeyError as key_err:
        print(f"❌ Error parsing data: Missing expected key {key_err}")
    except Exception as err:
        print(f"❌ Unexpected error: {err}")

if __name__ == "__main__":
    get_miami_weather()

    Output:

    % python3 ./m_weather.py
🌤️  Miami, FL Weather for Today:
High: 88.0°F (31.1°C)
Low:  73.2°F (22.9°C)

    I tried to keep the size and scope within something that would reasonably fit in a comment. Looks pretty decent to me, but I can't write Python myself. Never learned. I double-checked the LAT & LON of Miami, and it's spot on.

    It did take 47 seconds, while a cloud LLM would probably take 5 or less.

    All I'm saying is local LLM isn't garbage and it is getting better all the time.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 15 child comments
    •
  • [–] 6 points 6 months ago (8 children)

    That's interesting.

    How much ram did it use while running?

    If you used a GPU, how much does it cost in today's prices?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 8 child comments
    •
  • [–] 11 points 6 months ago* (last edited 6 months ago) (7 children)

    It's a MacBook Pro. 36GB of ram. I am sure Macs have some kind of gpu and I understand it somehow combines GPU ram with system ram, but I don't really know Mac hardware very well.

    It's beefy for a laptop, but the desktop I built for myself several years ago had 32 GB of ram and a GTX 1660, so I'm guessing they are similar in capability. I gave that to my daughter, so I can't run a comparison right now.

    EDIT: After doing just a bit of research, I've learned the unified memory architecture that Macs use, while not ideal for many purposes, is actually a big advantage for running larger inference models. So it's possible that this particular model wouldn't run at all on my Linux box or would run much slower because the full model wouldn't fit in the 6GB of VRAM and create a lot of memory thrashing.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 7 child comments
    •
  • [–] 4 points 6 months ago (4 children)

    You can use something like KoboldCPP on Linux, which allows both RAM and VRAM combined to run a model. O'course, not as fast when compared to pure VRAM or the Mac approach, but it is an option. I use my 128gb RAM with some GPUs for running models.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 1 point 6 months ago (3 children)

    Ollama and llama.cpp allow it too but it's super slow in my experience.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 1 point 6 months ago (2 children)

    Speed depends on how much of the model is on VRAM, and the dense/MoE architecture of that model. The RAM's benefit is more about having the ability to run the model in the first place. In any case, a dense Qwen3.6 27b would take up about 27-33gb-ish of memory, plus whatever context size you set.

    Upcoming implementation of MTP will increase the size of models, but in exchange, they will also run faster. About a 30%ish boost for dense models, a bit less for Mixture of Expert varieties, from the looks of it.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 1 point 6 months ago (1 child)

    When I've tried running a ~14 gigabyte distillation of whatever model it is I was trying to run, it would come out super slow at I believe 50/50 GPU to CPU. It gets so slow it was just more bearable to run a 7 or 8 b model that would actually fit entirely in VRAM and run entirely on GPU. Also made the rest of computer usage more bearable.

    To be fair I do only have a 6 core 6 thread CPU though. It shot up to 600% usage so even the DDR4 memory wasn't really bottlenecking it. I suspect a 9950X would fare a lot better.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 1 point 6 months ago

    I am using a 5950x, with 128gb of DDR4 3600 memory. The GPUs are a 3060 and 4090, totaling 36gb of VRAM. IMO, being bottlenecked by the CPU is definitely a thing, it just comes third after the VRAM and RAM considerations.

    With a 35b+3a MoE at Q8 with KV8, I get...

    [11:54:32] CtxLimit:18858/262144, Init:0.18s, Processed:17294 in 7.66s (2259.18T/s), Generated:1564/32768 in 29.01s (53.91T/s), Total:36.85s

  • permalink
  • fedilink
  • source
  • parent
  • [–] 3 points 6 months ago

    Yup, you want memory accessible to the GPU for local AI. AMD Strix Point and Mac devices are popular options. CPU can run LLMs but very slowly. I've got 32 GB of RAM and 8 VRAM and it's borderline useless for models that don't fit in the VRAM.

  • permalink
  • fedilink
  • source
  • parent
  • [–] 1 point 6 months ago* (1 child)

    qwen 3.6 is awesome, but 48-64gb is still real money these days. (though 32gb on dedicated separate machine is also more money). Sonnet 3.5 to opus 4.5 level benchmarks. and the online cost metrics for 27b and 35b are way off considering the overall usefulness of a 48-64gb machine (inclusive of gpu vram for 35b) which even in single, non batching, use could displace $5-$7/day of use.

    Local costs are much lower than online costs in linked chart, but if online, there are better models

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [+] -9 points 6 months ago (2 children)

    You see hot that's tangential to what you're replying to?

    Ai is evil

    LOCAL AI is not all evil

    Computers are expensive

    Your point is completely valid, but in another discussion.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 30 points 6 months ago* (last edited 6 months ago)

    Sorry, but I think the point about local AI not necessarily being evil is the tangent here.

    The OP is about motherboard shortages, which is being driven by the big AI companies and is making hardware unaffordable for normal users

    The top level reply to that is about how that's bad because it removes the ability for people to be in control of their own computing

    Then someone comes in, saying "yeah, but you can host your own AI so that it's not evil so not all AI is bad"

    Then someone points out that you can only host your AI if you can afford the hardware to do so which, as the OP and the comment you replied to pointed out, is getting really hard to do.

  • permalink
  • fedilink
  • source
  • parent
  • [–] 41 points 6 months ago (2 children)

    if you did not understand the comment from above it's fine but splitting hairs like you are doing is silly (everybody knows it's not 100 % of AI is 100% evil)...

    your comment is exactly the same as when people say "guns don't kill people, people kill people"... yes, we all know guns are not autonomously killing people, the point is that guns, as a tool, are remarkably good at doing something we do not want, which is to kill people

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 9 points 6 months ago (1 child)

    Not to go on a separate tangent, but that's the entire point of guns. They are supposed to kill. That's not meant to be some crazy conservative defense of them or opposition to regulating them. Just pointing out something that seems to get lost in conversations.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 3 points 6 months ago

    Correct... so when I tell you "guns DON'T kill people, people kill people" you are right to assume and I am just an idiot trying to jingle keys in front of you to distract you from the fact that guns do in fact kill people.

  • permalink
  • fedilink
  • source
  • parent
  • [–] 4 points 6 months ago

    The US government is already setting down the legal framework to make self hostable AI ilegal so good luck with that. Also self hostable AI is still being trained on stolen material so still fascist.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 42 points 6 months ago (1 child)

    Who knew cartels were bad for markets? I am sure economists and regulators were screaming at legislators for years but I guess they couldn't hear them over the sound of the bags of dirty lobbyist money landing on their desks.

    And then there are the financial "irregularities" funding the AI boom which are also not getting any attention.

  • permalink
  • fedilink
  • source
  • hideshow 1 child comment
    •
  • [–] 2 points 6 months ago

    Who knew cartels were bad for markets? I am sure economists and regulators were screaming at legislators for years

    Unfortunately there's plenty of economists that still believe in hardcore libertarianism, which is what led to our current situation

  • permalink
  • fedilink
  • source
  • parent
    • [–] 34 points 6 months ago (8 children)

    The wet dream for big tech has been to get people to pay subscription fees for compute, just like businesses do for cloud hosting. They tried with Stadia to get people to play games hosted in the cloud, but that was never going to fly.

    With the compute demands of AI (which is comparable to a AAA game except for the largest models), they dont want to make the same mistake and let you have the compute. They see this as an oppurtunity for subscription fees for the earth.

    The fact that we cant get hardware for a reasonable price is an added bonus to this plan.

    All of this only works of everyone subscribes to this shit. Businesses will, because its just easier to manage it. Consumers though should not give in. If you want to run an agent, use a small local model.

    The best thing that can be done is to make local open source agents and models approachable for regular users. Right now, they arent.

  • permalink
  • fedilink
  • source
  • hideshow 8 child comments
    •
  • [–] 6 points 6 months ago (1 child)

    The wet dream for big tech has been to get people to pay subscription fees for compute, just like businesses do for cloud hosting.

    Imagine the mental health benefits when AI datacentres make computers unaffordable, so we all have to go outside more, and then the AI datacentres shrivel because they have no customers, because we can't access anything with no computers. So the AI companies die off.

    I can dream, ok?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 2 points 6 months ago

    I would love this to be an unintended outcome from all this. However, I don't think that's where we're headed.

    I, for one, think there's a lot of slop in and around the engineering of phones. We might see a lot more software, storage, and overall activity crunched, compressed, and crammed into our portable devices instead. And with more stuff in the cloud/SaaS realm, they can also become (even) thinner clients at the same time. :(

    It's "heavier" gear like laptops and desktops that'll probably get pushed into the pro and "prosumer" market.

  • permalink
  • fedilink
  • source
  • parent
  • [–] 6 points 6 months ago (2 children)

    The wet dream for big tech has been to get people to pay subscription fees for compute, just like businesses do for cloud hosting.

    Thankfully there's a growing number of businesses that have been burned by this, and it seems like companies are starting to try bringing their critical systems back in-house again

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 4 points 6 months ago (1 child)

    There is this tipping point where it becomes more cost effective to bring it inhouse, even with the staffing requirements. For small to medium sized buinesses though cloud all the way.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 1 point 6 months ago

    where it becomes more cost effective

    Reliability and risk are also factors. What do you do when a vendor tries to lock you into a walled garden before cranking up prices? What about storage of sensitive information? Sometimes the additional cost of doing it in-house pays off in ways that are difficult to track

  • permalink
  • fedilink
  • source
  • parent
  • [–] 4 points 6 months ago (2 children)

    I work for a large retailer that you've definitely heard of. We are pulling away from our cloud hosted presence and are building out a self-managed virtual data center in one of our own physical data centers.

    Even enterprise knows that paying a monthly uncontrolled cost is shit.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 1 point 6 months ago (1 child)

    So you guys are bascially building out your own cloud? Where does one even start with something like that?

    We are fully cloud where I am, but i have this dream where we self host our inferrence. Ever since i learned more about that 40k acre data cneter in Utah offering capacity for the big guys (AWS, etc), im very skeptical about how safe our data is when sending it to a model. Ethical issues aside

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
    • [–] 32 points 6 months ago (7 children)

    Motherboards were already way too goddamn expensive, anyway.

    About a year ago I was considering upgrading my AM4 PC to AM5. The rock-bottom cheapest motherboards were only slightly cheaper than the relatively high-end one I got 5-ish years ago. I decided to stick with my current PC.

  • permalink
  • fedilink
  • source
  • hideshow 7 child comments
    •
  • [–] [S] 20 points 6 months ago (2 children)

    There's going to be a lot of us running 2019-vintage PCs indefinitely.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 6 points 6 months ago*

    Just repasted both of mine with phase change materials and thermal putty, including GPU dies. Looking at getting better/more fans and a smarter hub soon. Gonna make this 5800X3D last as long as possible; I'm in this for the long haul lol

  • permalink
  • fedilink
  • source
  • parent
  • [–] -3 points 6 months ago (2 children)

    I think motherboards have been pretty aptly priced. What do you think a complex piece of a computer like that should cost? The most important part of your computer that ties every single part together?

    150USD? You can get a decent board for that.

    Under 100? First of all, that’s insane, and second of all, you can get a budget board for under 100.

    350 is too much? You’re looking at a high end board. You’re paying high end board prices.

    Is your complaint that 500 is too much for a mobo? Why are you even looking at 500 dollar mobos?

    All of my boards I’ve gotten in the last five builds have been 200-300 and those are amazing machines that are either newer and excellent or older and still hold up over a decade later.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 2 points 6 months ago* (1 child)

    On the first computer I ever built with parts bought with my own money (around 2006 or 2007) I chose one of the best mainboards available at the time (ASUS M2N32-SLI Deluxe lol).

    It cost about 160€ back then. That is about 250€ in todays money. There was a slightly more expensive "crosshair" version of this board, but really nothing more expensive than that.

    The last mainboard I bought was "low-end" one that cost about 200€ (around two years ago).

    Back then (~2006) you could get a "okay" mainboard for 50€ (~90€ today).

    To me all but the lowest-end stuff seems a lot more expensive today.

    For my first PC I bought an Nvidia Geforce 8800 GTS, which was the second-best GPU at the time. The GTX model cost around 500€ (~770€ today) if I remember correctly.

    My first self-bought computer cost me about 1300€ and was pretty close to top of the line.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 1 point 6 months ago

    Your experience is pretty similar to mine! I went with two 7900GT in SLI for my first machine tho—an 8800GTS would have been a better pick. That mobo cost me around 240 and it was fairly high end.

    High end nowadays consists of a lot more stuff, though, and prices have not largely increased. Consider the difference between the 8800 series when it came out to the cost of a 3090/4090/5090! When my store got the 8800 series in we all laughed, that was way too much for a video card, nobody’s gonna spend over 600 dollars on a single video card! … D:

    For motherboards though, I’d say 180-300 today covers pretty much all mid/high end boards, save for a few outliers. I’d be curious to know what an average user would get out of a board that costs more than 350.

  • permalink
  • fedilink
  • source
  • parent
    • load more comments (-1 replies)
    [–] 29 points 6 months ago (4 children)

    Good! Motherboard prices have been wildly out of control for a while now. Asus selling 1k motherboards when that same tier of motherboard used to top out at 400 bucks max. Let these greedy fucks have their lunch.

  • permalink
  • fedilink
  • source
  • hideshow 4 child comments
    •
  • [–] 7 points 6 months ago (3 children)

    Don't forget gutting the connectivity on lower end motherboards. Apparently PS/2 ports and LED error lights are premium features now.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
    • [–] 41 points 6 months ago (10 children)

    How has this whole saga not been an obvious indictment of 'the free market?'

    Big players shouldn't be allowed to gobble up all the resources needed by small ones. How is it not obvious that they need to wait until production increases to meet their needs before embarking on their little project?

  • permalink
  • fedilink
  • source
  • hideshow 10 child comments
    •
  • [–] 38 points 6 months ago (6 children)

    No this is exactly what the free market is. Regulations that make things fairer for small people is communism.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 6 child comments
    •
  • [–] 11 points 6 months ago (1 child)

    Then give me communism bc this free market shiz aint it. Giving us the scraps that AI doesn't want is just nuts if we get anything at all. Esp since no one beyond billionaires want this ai crap

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 2 points 6 months ago (3 children)

    What about regulations that make things fairer for tall people?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
    • [–] 36 points 6 months ago (8 children)

    I keep thinking maybe things will good by 2030 and remember that's 4 years away. Game devs please target the Steam Deck and Switch 2 as the baseline. Mid range and high end is just too premium for most people. Even entry level enthusiast gaming hardware is too expensive because of memory and storage. Steam Deck and Switch 2 are good low power draw integrated graphics level. That's not terrible for pricing

  • permalink
  • fedilink
  • source
  • hideshow 8 child comments
    •
  • [–] 23 points 6 months ago (6 children)

    Even having high-end enthusiast hardware, I want those devices as the baseline too. Whatever optimizations they do still apply over the whole hardware spectrum.

    Also, you can technically say 2030 is less than 4 years away if you want to traumatize old people. Lol.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 6 child comments
    •
  • [–] 10 points 6 months ago (4 children)

    I like watching sci-fi made in the 80s and 90s where they thought that 2015 was a really advanced distant prospect.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 4 points 6 months ago

    According to valves hardware survey over half of all gamers have hardware less powerful than the Steam Machine so provided valve don't go mad with the pricing that may actually end up happening.

    I think the steam deck and the switch 2 are probably too low power to be reasonable targets.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 45 points 6 months ago (9 children)

    If we can't make a time machine to go backwards, can we at least pause time? The future absolutely fucking sucks, let's just avoid it altogether lol

  • permalink
  • fedilink
  • source
  • hideshow 9 child comments
    •
  • [–] 5 points 6 months ago (4 children)

    Seed debris in the orbit to destroy current satellites and then prevent new ones from being launched for several decades

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 9 points 6 months ago (3 children)

    Honestly I quite like space.

    Better solution is to just kill like six people. Because that's all the AI industry is, it's six really annoying rich guys who are having the most extravagant pissing contest in history.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 4 points 6 months ago

    Six? There's Jensen Huang, Lisa Su, Alex Karp, Peter Thiel, Sam Altman, Dario Amodei, Mark Zuckerberg, Elon Musk, Satya Nadella and Sundar Pichai at a minimum to start with, but there are many others too. I'm including the CEOs of Nvidia and AMD because they've been pushing AI for years to get more valuable customers than we ever were to them.

    Just in case anyone needs a list.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 15 points 6 months ago

    I have a home server waiting around for an SSD for a year now. I have the money, but I don't like feeling like I'm getting scammed. So I'd rather wait for this market to collapse than give them my money.

  • permalink
  • fedilink
  • source
    •
    load more comments
    view more: next ›
    13 comments in c/technology@lemmy.zip view that post
    7 comments in c/hackernews@lemmy.bestiver.se view that post
    2 comments in c/pcgaming@lemmy.ca view that post