Loading...
> >

Nvidia shelves 2026 new card releases

TorquinoxTorquinox Posts: 4,877
in The Commons

The good news keeps coming. Apparently NVidia shelved every new gpu card release for 2026. https://tech-insider.org/nvidia-skips-2026-gaming-gpus/ 
 

Okay, so it's September, but the 24Gb 5080 Super expected in December is apparently also shelved. IDK really what it means except that, coupled with spiraling hardware price increases, folks are going to have a harder time upgrading or replacing systems due to cost. 
 

I've read a lot of articles speculating about what will happen if NVidia abandons the consumer gpu card market because it represents a small portion of their total income. Some believe NVidia will provide GPU as a subscription service. I don't like that idea, but I could see that happening.

 

Nvidia Is not alone in pulling back from the consumer market. Micron also ended their Crucial line of consumer ram kits because data center build out is far more profitable. Interesting times!

Comments

  • 3DIO3DIO Posts: 653
    edited September 13

    The Solution:  Avoid using cloud services at all costs!
    This will cause their market abuses and investments to backfire in such a catastrophic way, they will never repeat those practices again.

     

    Post edited by 3DIO on
  • PadonePadone Posts: 4,326
    edited September 13

    The data centers are always payd by us, the consumers. It is us who have control of the market. If Micron abandons consumers for data centers it means consumers pay more for data centers than for new hardware. Personally I'm good with old hardware until I'm forced to change it by the applications I use. I can also stay outdated with applications until I'm forced to update them for some new features I need. Actually my old 1060 card works fine with DS4 and I see no reason to change that, especially with the actual market prices.

    @3DIO Small correction, avoid -paying- for cloud services, we can use them for free there's no angle for them then.

    p.s. However, AI services are truly impressive, apart too much censorship that is really blocking creativity. I wasn't able to get a girl wearing a bikini, or a night gown, really nothing NSFW but the AI decided so.

    Post edited by Padone on
  • WendyLuvsCatzWendyLuvsCatz Posts: 41,565
    edited September 13

    while I use AI, it is locally

    cloud based is a big no for me

    I suspect that is actually the case with most people

    and the other uses for data centres such as monitoring people are not popular

    it will bite them in the bum, might take a while for the crash, but when it happens it will be spectacular devil

    Post edited by WendyLuvsCatz on
  • PadonePadone Posts: 4,326

    As I understand it, using AI locally requires at least 16G vram and 64G ram, a powerful cpu, better if it's a nvidia gpu. Just to have basic results not comparable with those data centers can provide. Not something everyone can afford with the actual market prices.

    Please correct me if I'm wrong, I'd like to.

  • WendyLuvsCatzWendyLuvsCatz Posts: 41,565
    edited September 13

    I use low VRAM versions in Pinokio

    8GB VRAM and 32GB RAM versions 

    it's not fast but faster than 3D rendering 

    Post edited by WendyLuvsCatz on
  • PadonePadone Posts: 4,326

    Thank you for letting me know how it goes on your side. That can be a reasonable upgrade for my hardware. But are the results good ? I hear local AI only provides basic results, not comparable with data centers and barely useful.

  • TorquinoxTorquinox Posts: 4,877
    edited September 13

    I believe the vast amounts of money going into the ai industry is short-circuiting how the market usually works. While the whole ai build out is probably a market bubble, it's not clear if or when that will pop.
     

    Meanwhile, there is this wholesale shift going on in the marketplace. It seems like a brute force re-engineering of how the computer market works, where the biggest players move all the necessary hardware out of our hands and shunt it into paywall-regulated data centers.


    I think people have already accepted the computing cloud as a fact of life. Those of us who want to keep everything local are outliers.

    Post edited by Torquinox on
  • ValiskaValiska Posts: 178
    edited September 13

    Padone said:

    As I understand it, using AI locally requires at least 16G vram and 64G ram, a powerful cpu, better if it's a nvidia gpu. Just to have basic results not comparable with those data centers can provide. Not something everyone can afford with the actual market prices.

    Please correct me if I'm wrong, I'd like to.

    I have 48G RAM and 8G VRAM. I can run moderately capable LLMs in LM Studio or Ollama.

    Useful stuff so far: 

    • They've managed to tell me about some recent science I wasn't aware of (for writing science fiction), and have managed to usefully summarize some stuff I fed them.
    • They sometimes can partially make up for the growing gross deficits of Google search, in that they might be able to identify information on the basis of a description. That's helpful if you can describe what you want but don't know the right terms of art to call it up.
    • I've managed to get one satisfactory science fictional image out of the image generators at Nightcafe (because I haven't gotten around to setting up any local image generation, and probably won't try that for a while yet).
    • I got a handy summary of one batch of literary information I fed it.

    Not so useful:

    • I haven't yet managed to get one to take over some text transfer and editing (too bad, it's a nuisance job that mostly involves putting back together English-and-tables jigsaws from some badly constructed PDFs. Time-consuming but not very interesting).
    • Neither LLMs or image generators actually understand how the world works, or what anything means. Online image generators, last time I tried, could not be made to make a successful image of science fictional objects rarely represented in existing art.
    • Processing time and heat generation wasted flattering the user.

    The LLMs' ability to write grammatical English is impressive.

    The most capable LLMs I can run start losing track of the logic of a conversation at somewhere between 5-8 pages, give or take. That means LLMs often don't turn out to be a great timesaver, because if you're poking at something of much complexity, you have to keep stopping to make sure you have summaries that will allow the LLM to keep going a little farther. Comes a point where it just loses track of what it's supposed to be trying to do.

    Sometimes it never lands in the right quadrant and then it doesn't do much good even in short form.

    So far for me it's an interesting toy, but mostly not yet a particularly efficient tool for many purposes. Maybe it will soon become so; I don't think it would need too much improvement to get over the hump, for me. From what I've seen so far, I am expecting it to be pretty interesting for post-processing.

    The only kind of question I'll ask a publically accessible AI is a question I don't care if everyone, including radically untrustworthy giant entities, knows I asked.

    Post edited by Valiska on
  • ValiskaValiska Posts: 178

    Torquinox said:

    I believe the vast amounts of money going into the ai industry is short-circuiting how the market usually works. While the whole ai build out is probably a market bubble, it's not clear if or when that will pop.
     

    Meanwhile, there is this wholesale shift going on in the marketplace. It seems like a brute force re-engineering of how the computer market works, where the biggest players move all the necessary hardware out of our hands and shunt it into paywall-regulated data centers.


    I think people have already accepted the computing cloud as a fact of life. Those of us who want to keep everything local are outliers.

    If I can't afford the next gaming laptop for computer graphics, I'm buying something lower-end, switching to Linux, coding to amuse myself, and reverting to traditional artistic media. I can use the existing Daz/Poser library for quick line renderings, to save some time in preliminary sketches. I will largely drop offline, rather than move farther online.

  • jmucchiellojmucchiello Posts: 2,136

    I recommend reading The Tech Report youtube channel. They constantly talk about the inevitable AI bubble and how we'll all end up paying for it. There are six or so companies trying to ship 10 tons of canaries in a 5 ton cargo hold by having half the canaries in flight at all times. Many of the RAM foundaries have commitments through 2028 that are exclusively AI data center commitments. And since they are so lucritive, they aren't spending on consumer grade anything in the near term. This is why NVIDIA isn't releasing newer video cards. They cost too much to manufacture (well, assemble technically) given the ton of cash being dumped into AI.

  • SnowSultanSnowSultan Posts: 3,838
    edited September 13

    I have what's probably considered a decent system for running AI locally (nvidia 4090, 128 GB RAM, but an older CPU). I can tell you a bit of my experiences with using generative/image-related AI (I don't run LLMs on mine, I use ChatGPT for that).

    Useful stuff:

    • If you can figure out ComfyUI - which is a whole other frustrating ordeal - it is excellent for making 3D renders look realistic. You can turn a test render into a photo with little trouble.
    • Makes environments easily, no need for huge meshes, skyboxes, or scattering systems.
    • You can replace clothing, hair, and other things, as well as making effects that would be difficult in 3D, like realistic water, smoke, clothing folds, and soft skin dynamics.
    • Depending on the models used, you can restyle your image easily into cartoon or anime (with mixed results), or cinematic styles like 1950s, noir, vintage photos, vector, or pixel art.
    • It's free if you do it at home and uncensored (which virtually no online service is nowadays).

     

    What it doesn't do well:

    • Generate highly accurate objects from scratch with no references. You'll often get very odd results when asking for objects that are not common or that are creative, like a demon tail. This is where using 3D in conjunction with AI is important; the 3D can be the reference to help guide the AI (it will still often mess it up, which is why it's good that you're not paying for each attempt).
    • Consistency with characters or environments can be tricky to achieve, it can be done but it's a bit random. You can get much better consistency using a online service like OpenArt or Higgsfield.
    • Local video generation doesn't come close to what can be done online. Almost every cool cinematic video with lots of action that you see done with AI are done online, likely using expensive video models like Seedance. You can make videos at home, but they can take a long time, run your graphics card hard, and aren't going to give you Hollywood results.

     

    I have a lot of examples of both local render enhancements and Seedance fight scene videos on my deviantArt page, as well as ramblings in my journals about my daily tests if you want to see any examples. Hope that can be of some help.  :)

    Post edited by SnowSultan on
  • 3DIO3DIO Posts: 653
    edited September 13

    @Torquinox
    I think you underestimate how awake the public are these days.  Personally, I'm 100% confident that any and all attempts to enslave the population will fail.
    Their failure is guaranteed, and  HERE  is why that is the case  :-D
     

    They prefer to call it :
    "Reticulum - Unstoppable Networks For The People.".

    I personally prefer to call it :
    "Salvation Technology For The Human Race.".

    As you can see from the languages supported on Their Website, it has already established itself as a globally-recognised technology of salvation  heart
    Maybe watch a few Reticulum videos on YouTube, and spread the word to your fellow humans whenever and wherever possible, and appropriate!
     

     

    reticulum_logo_512.png
    512 x 512 - 85K
    Post edited by 3DIO on
  • TorquinoxTorquinox Posts: 4,877

    @3DIO I hope you're right. That's an area where I'm happy to be wrong.

  • 3DIO3DIO Posts: 653
    edited September 13

    Sure, and check this out my weary friend!
     

    "Reticulum is the cryptography-based networking stack for building local and wide-area networks with readily available hardware. Reticulum can continue to operate even in adverse conditions with very high latency and extremely low bandwidth.

    The vision of Reticulum is to allow anyone to operate their own sovereign communication networks, and to make it cheap and easy to cover vast areas with a myriad of independent, interconnectable and autonomous networks. Reticulum is Unstoppable Networks for The People.

    Reticulum is not one network. It is a tool for building thousands of networks. Networks without kill-switches, surveillance, censorship and control. Networks that can freely interoperate, associate and disassociate with each other. Reticulum is Networks for Human Beings.

    From a users perspective, Reticulum allows the creation of applications that respect and empower the autonomy and sovereignty of communities and individuals. Reticulum provides secure digital communication that cannot be subjected to outside control, manipulation or censorship.

    Reticulum enables the construction of both small and potentially planetary-scale networks, without any need for hierarchical or beaureucratic structures to control or manage them, while ensuring individuals and communities full sovereignty over their own network segments."
     

    Post edited by 3DIO on
  • PadonePadone Posts: 4,326

    @Valiska @SnowSultan Thank you for sharing your experience, at least this gives us some info about usage and limitations, even with high-end hardware. I think I'll try it myself starting with my actual limited hardware, just to see how it goes. I understand Comfy UI can also use optimized models that are good for low vram.

  • kyoto kidkyoto kid Posts: 42,415
    edited 1:47AM

    ...this is worse than the crypto craze of a decade ago.  Back then it was just GPUs which was bad enough, now it is pretty much everything,

    The AM5 upgrade is now 3.400 USD. may as well buy a lotto ticket.

    Even backstepping to an AM4 Ryzen 9 5900XT and 64 GB DDR4 3200 is ridiculous.

     

     

    Post edited by kyoto kid at
  • TorquinoxTorquinox Posts: 4,877
    edited 1:47AM

    jmucchiello said:

    I recommend reading The Tech Report youtube channel. They constantly talk about the inevitable AI bubble and how we'll all end up paying for it. There are six or so companies trying to ship 10 tons of canaries in a 5 ton cargo hold by having half the canaries in flight at all times. Many of the RAM foundaries have commitments through 2028 that are exclusively AI data center commitments. And since they are so lucritive, they aren't spending on consumer grade anything in the near term. This is why NVIDIA isn't releasing newer video cards. They cost too much to manufacture (well, assemble technically) given the ton of cash being dumped into AI.

    The pattern of available information supports what you're saying, too. I'll look into the tech report. Thanks. 

    Post edited by Torquinox at
  • donniekeidicdonniekeidic Posts: 147

    I'm actually kind of surprised I haven't heard of organized crime rings breaking into data centers. Or maybe they are all heavily guarded.

  • jmucchiellojmucchiello Posts: 2,136

    donniekeidic said:

    I'm actually kind of surprised I haven't heard of organized crime rings breaking into data centers. Or maybe they are all heavily guarded.

    You would have to know what you're doing to rob a data center. As soon as you take the plug out of the wall, a dozen or more people probably are immediately alerted. And that assumes you got into the building undetected.

    Next you have to move a bunch of rackmount equipment because a smash n grab approach can't possibly yield much.

    It would have to be an inside job.

    Maybe it could the next Ocean's movie.

  • TorquinoxTorquinox Posts: 4,877

    kyoto kid said:

    ...this is worse than the crypto craze of a decade ago.  Back then it was just GPUs which was bad enough, now it is pretty much everything,

    The AM5 upgrade is now 3.400 USD. may as well buy a lotto ticket.

    Even backstepping to an AM4 Ryzen 9 5900XT and 64 GB DDR4 3200 is ridiculous.

    I agree! Ridiculous!

  • jmucchiellojmucchiello Posts: 2,136

    Torquinox said:

    I agree! Ridiculous!

    No, it's greed. The RAM makers have the tech industry over a barrel and are making criminal margins. And they could ramp up production to lower prices. But what incentive is there to lower prices. Making more production costs money and what happens when they don't need the extra capacity? So they're just sitting back and rolling around in gold like Scrooge McDuck.

  • kyoto kidkyoto kid Posts: 42,415

    ...+1

  • DareshiranuDareshiranu Posts: 334

    They are going to discover that consumers will not be satisfied with thin clients and sub-standard services provided to them (for enormous recurring fees, of course) by the inevitable cloud services which themselves will demand ever increasing amounts of hardware and thus increasing fees irrespective of the greed of corporations. It will become the inverse of the economy of scale.

Sign In or Register to comment.