knightly the Sneptaur

  • 0 Posts
  • 76 Comments
Joined 3 years ago
cake
Cake day: July 5th, 2023

help-circle
  • I’m confident in my “never” because of the capital economics; the service has to be expensive enough to pay for the infrastructure it requires plus some profit for the shareholders, while simultaneously being cheap enough to offer gamers a better value proposition than buying their own hardware. There’s no margin between those limits, so the only market left for them to appeal to are niches where local rendering performance is limited but network latency and bandwidth are not. Even then, gamers still have the option of streaming from their own hardware using Moonlight rather than paying for a third-party service, so the only customers left are the ones with more money than sense.

    Don’t get me wrong, I love the concept (I even bought an OnLive microconsole back in the day and still regularly use a Steam Link to stream games to the living room TV), but it isn’t nearly convenient or performant enough to justify itself as a subscription service.


  • That’s 100% correct, because the shareholders still haven’t learned their lesson from OnLive, Google Stadia, Amazon Luna, Nvidia GeForce Now, Xbox Game Pass Ultimate, or Playstation Plus Premium.

    Game streaming serices are never going to catch on because the capital needed to build out the infrastructure is ridiculous.

    Without even counting the cost of hosting and bandwidth, a game streaming node already costs as much as a gaming PC, easily $2k. Say it gets 100% utilization with four customers each playing 6 hours a day, and that they’re paying $20/month for the service, then it’d take 25 months for that $2k node to pay for itself. However, by that time the hardware will be old and outdated, in need of replacement.

    That’s zero return on investment for the lifetime of the hardware, absolutely no profit even in this idealized case unless you can charge more than $20. Add in the monthly expenses for colocation space, power, bandwidth, and overhead and you get a product with zero ROI even at $35/month.

    Then, consider that this node can only realistically serve a small geographic area due to network latency limitations, and that using the service requires your customers to already be paying for high-quality broadband, and you’ve got a recipe for a tiny potential customer base and massive capital investment to make nodes available in close enough proximity to all major markets.




  • This makes a lot of sense though, from the description it sounds like they’re trying to build an NPU out of memristors. We’ve been expecting them to show up to do this kind of math for a bit, since they’d cut a lot of redundant computation out of the layered matrix calculus that NPUs are optimized for if we can make them small, fast, and reliable enough.

    And it’s not just for “AI”. A lot of problems, like physics modeling or speech recognition, can be reduced to matrix math. An analog, programmable memristor network can do that kind of calculus almost passively.




  • how much more for the product/service would you be willing to pay for a human operator on the other side or conversely, how cheap would the non-human supported product/service have to be for you to choose it over the more expensive human supported option?

    Better question, how much is a company willing to pay me to use an LLM instead of going to one of their competitors?

    Because if the answer is insultingly small then I’m not patronizing them.





  • I did, I do, and I’m calling this article bullshit for not pointing out that while the protocol might be open-source, they have yet to share the server software that’s required to operate it.

    BlueSky “lets” people host their own profile data because it reduces how much data they have to host. It does not allow them to login and browse the network without going through their centralized servers to do so.

    So, it’s not really decentralized, not really open source, and remains under corporate control until such time as they decide to let anyone compete with them on their own network.







  • It’s real, but the jargon is unintuitive.

    “Teleportation” in the field of quantum mechanics refers to the process by which a quantum state can be copied from one place to another.

    This process is like Shrodinger’s Cat, both alive and dead until you open the box to check. Quantum information simply does not exist until a measurement collapses it into back into classical information, so copying a quantum state literally involves teleporting the information about it from sender to receiver without allowing the box to be opened during the transition.