3 ms·

Am I missing something or is this as big of a disaster as I think? This is terrible news right?

ocdtrekkie 2y ago
Sort of but AI PCs are going to be more expensive than normal PCs so most users won't and will continue to not have a PC that does this. Pushing it out to existing machines would be a very different tier of problem.
- walterbell 2y ago
  > AI PCs are going to be more expensive than normal PCs The Qualcomm Oryon SoC is about half the price of an Intel CPU.
  - ocdtrekkie 2y ago
    Currently an Inspiron 7440 (non-Copilot+) starts at $849 and an Inspiron 7441 (with Copilot+) starts at $1,099. That's a $250 premium to get an "AI PC".
    - walterbell 2y ago
      12-core Oryon dev kit with 32GB RAM is $899, https://www.windowscentral.com/software-apps/windows-11/qualcomm-launches-snapdragon-developer-kit-with-12-core-x-elite-at-build-2024 https://www.windowscentral.com/software-apps/windows-11/qual... It's a brand new device family. OEM price competition should improve as more devices ship in Aug/Sep. Oryon should run Linux with good performance-per-watt. Hopefully the NPU can be disabled to save energy.
      - ocdtrekkie 2y ago
        The part where you are pointing out a dev kit and about how well it runs Linux is really just admitting you don't understand the conversation we are having here. Most users buy a cheap HP from Costco for $300. Businesses will buy the same standard line OptiPlex they bought last year. Very, very few people will magically end up with Windows Recall who didn't intend to.
        walterbell 2y ago
        > how well it runs Linux i.e. PC hardware for AI inference will not be limited to Windows. > Most users buy a cheap HP from Costco for $300. Businesses will buy the same standard line OptiPlex they bought last year. Both Intel and AMD announced upcoming chips with NPUs. Mediatek and Nvidia will likely join the Arm AI PC competition in 2025. Apple's 2024 OS updates are focused on AI features, both on-device and cloud partnership with OpenAI. Intel's Computex tagline a few weeks ago was literally "AI Everywhere". In a few years, silicon for on-device AI inference will likely be pervasive in retail PCs, including Costco, HP and Dell Optiplex. It has been shipping in Apple Silicon Macbooks and iPads since 2020, mostly unused by software until now.
- calgoo 2y ago
  I wonder if MS will then just make it a online processing version, so it just ships all your data "for free!!!!" to MS servers and then processed remotely for you! I also wonder how "knowledge transfer" will happen when you get a new machine in the future? What about backups, do they sit in the cloud already? These all sounds like ways that this "local" AI PC will share data with the MS cloud in one form or another. With Apple doing similar things already on iPhones (I know there are differences, but its still analyzing your data etc), I wonder if Linux might actually become a more mainstream OS in the future. Could also be that people stop using computers as much and just use tablets with docking stations + keyboard & monitors (again, there is work to be done, but its a possibility from a HW level). That would leave us with MacOS & Windows for business desktops (with some Chromebook & Linux sprinkled in there). Education would probably be more Tablet & Chromebook style compute, and gaming is already moving to the cloud (I guess the positive here is that we might finally be able to get rid of AntiCheat software:) ).
  - aleph_minus_one 2y ago
    > I wonder if MS will then just make it a online processing version, so it just ships all your data "for free!!!!" to MS servers and then processed remotely for you! > I also wonder how "knowledge transfer" will happen when you get a new machine in the future? What about backups, do they sit in the cloud already? Honestly, I guess this will become very expensive for Microsoft, and they won't find a good business case what to do with the collected data. So Microsoft is wasting a huge load of money, and additionally their AI spyware causes a huge reputation damage for Microsoft: a lose-lose situation. :-(
    - walterbell 2y ago
      Other use cases for on-device NPU silicon have been demonstrated, e.g. radio sensing of room geometry and human motion, breathing, gestures.
      - aleph_minus_one 2y ago
        Thanks for the clarification of other kinds of spying that this enables. Nevetheless I still have difficulties seeing how this is supposed to make money for Microsoft.
- dmurray 2y ago
  If it works like any other electronics product I've bought in the last ten years, the "AI" version will soon be cheaper thanks to subsidies from advertisers, while the "normal PC" will be twice as expensive and the domain of tinkerers and cranks.
  - __loam 2y ago
    I just want to buy a screen guys.

Sort of but AI PCs are going to be more expensive than normal PCs so most users won't and will continue to not have a PC that does this. Pushing it out to existing machines would be a very different tier of problem.

walterbell 2y ago
> AI PCs are going to be more expensive than normal PCs The Qualcomm Oryon SoC is about half the price of an Intel CPU.
- ocdtrekkie 2y ago
  Currently an Inspiron 7440 (non-Copilot+) starts at $849 and an Inspiron 7441 (with Copilot+) starts at $1,099. That's a $250 premium to get an "AI PC".
  - walterbell 2y ago
    12-core Oryon dev kit with 32GB RAM is $899, https://www.windowscentral.com/software-apps/windows-11/qualcomm-launches-snapdragon-developer-kit-with-12-core-x-elite-at-build-2024 https://www.windowscentral.com/software-apps/windows-11/qual... It's a brand new device family. OEM price competition should improve as more devices ship in Aug/Sep. Oryon should run Linux with good performance-per-watt. Hopefully the NPU can be disabled to save energy.
    - ocdtrekkie 2y ago
      The part where you are pointing out a dev kit and about how well it runs Linux is really just admitting you don't understand the conversation we are having here. Most users buy a cheap HP from Costco for $300. Businesses will buy the same standard line OptiPlex they bought last year. Very, very few people will magically end up with Windows Recall who didn't intend to.
      - walterbell 2y ago
        > how well it runs Linux i.e. PC hardware for AI inference will not be limited to Windows. > Most users buy a cheap HP from Costco for $300. Businesses will buy the same standard line OptiPlex they bought last year. Both Intel and AMD announced upcoming chips with NPUs. Mediatek and Nvidia will likely join the Arm AI PC competition in 2025. Apple's 2024 OS updates are focused on AI features, both on-device and cloud partnership with OpenAI. Intel's Computex tagline a few weeks ago was literally "AI Everywhere". In a few years, silicon for on-device AI inference will likely be pervasive in retail PCs, including Costco, HP and Dell Optiplex. It has been shipping in Apple Silicon Macbooks and iPads since 2020, mostly unused by software until now.
calgoo 2y ago
I wonder if MS will then just make it a online processing version, so it just ships all your data "for free!!!!" to MS servers and then processed remotely for you! I also wonder how "knowledge transfer" will happen when you get a new machine in the future? What about backups, do they sit in the cloud already? These all sounds like ways that this "local" AI PC will share data with the MS cloud in one form or another. With Apple doing similar things already on iPhones (I know there are differences, but its still analyzing your data etc), I wonder if Linux might actually become a more mainstream OS in the future. Could also be that people stop using computers as much and just use tablets with docking stations + keyboard & monitors (again, there is work to be done, but its a possibility from a HW level). That would leave us with MacOS & Windows for business desktops (with some Chromebook & Linux sprinkled in there). Education would probably be more Tablet & Chromebook style compute, and gaming is already moving to the cloud (I guess the positive here is that we might finally be able to get rid of AntiCheat software:) ).
- aleph_minus_one 2y ago
  > I wonder if MS will then just make it a online processing version, so it just ships all your data "for free!!!!" to MS servers and then processed remotely for you! > I also wonder how "knowledge transfer" will happen when you get a new machine in the future? What about backups, do they sit in the cloud already? Honestly, I guess this will become very expensive for Microsoft, and they won't find a good business case what to do with the collected data. So Microsoft is wasting a huge load of money, and additionally their AI spyware causes a huge reputation damage for Microsoft: a lose-lose situation. :-(
  - walterbell 2y ago
    Other use cases for on-device NPU silicon have been demonstrated, e.g. radio sensing of room geometry and human motion, breathing, gestures.
    - aleph_minus_one 2y ago
      Thanks for the clarification of other kinds of spying that this enables. Nevetheless I still have difficulties seeing how this is supposed to make money for Microsoft.
dmurray 2y ago
If it works like any other electronics product I've bought in the last ten years, the "AI" version will soon be cheaper thanks to subsidies from advertisers, while the "normal PC" will be twice as expensive and the domain of tinkerers and cranks.
- __loam 2y ago
  I just want to buy a screen guys.