Raspberry PI vs The Cloud

Sdílet
Vložit
  • čas přidán 14. 06. 2024
  • Here I'm comparing running the TelAIphone micro-services on a Raspberry Pi 5 8gb and on Digital Ocean. I'd love to hear your thoughts.
    Cheers!
    Check out TelAIphone: https:telAIphone.com
    Check out Digital Ocean and get $200 free:
    m.do.co/c/b2751003260d
  • Věda a technologie

Komentáře • 18

  • @arthurpizza
    @arthurpizza Před 10 dny +2

    I'm cheap and keep finding free Raspberry Pis. That's a combo that keeps me off the cloud.

  • @armstrongskyview2810

    In the uk that game is also called Chinese whispers

  • @steve0steel
    @steve0steel Před 11 dny +3

    Great video, great topic!

  • @samadams557
    @samadams557 Před 8 dny

    I like this video.
    Totally more than w 12 subscriber channel. Guess the counter hasn't updated yet.

  • @N.A._
    @N.A._ Před 9 dny +1

    Did you consider the Raspberry Pi AI Kit, which adds 13 TOPS for $70 ?
    It could reduce those AI tasks to milliseconds

  • @serbansenciuc
    @serbansenciuc Před 9 dny +1

    I was thinking about this as well. An interesting finding was MiniPCs, they are more expensive but more powerful.
    As a side note, if you want to use a Raspberry Pi in a production environment, you should upgrade the storage from an SD card to an M.2 SSD because the SD card has limited write cycles.

    • @jacstrong
      @jacstrong  Před 9 dny +1

      I agree that an SSD would be the way to go. I just wanted to compare stock R Pi to a cloud instance.
      I think however in this use case I would probably use the PCI slot for some type of TPU or accelerator. The entire model and the images it captioned stay in memory so it’s not very disk IO intensive.

  • @Kaze919
    @Kaze919 Před 9 dny +1

    Interesting, do you have the server management experience to know how to set up the fallback server? It’s really intriguing idea and I never thought something like this could be at least comparable in a self hosting scenario.

    • @jacstrong
      @jacstrong  Před 9 dny +1

      Most of my experience is in software development not hardware management. To answer your question no I currently know how to setup a fallback server.
      However, my experience being in software, I built the application to not rely on any one server. The AI jobs are managed by a queuing system, so if any one server goes down then the others will pick up the slack. I currently have 3 servers in two separate locations running these jobs, so chances are one of them will always be online.

  • @Shreyash.guptaa
    @Shreyash.guptaa Před 9 dny +1

    Totally a more than 11 subscriber channel.

  • @MRPtech
    @MRPtech Před 6 dny

    Took me a while to understand why your t-shirt logo is SADIDA :)

    • @jacstrong
      @jacstrong  Před 6 dny

      I didn’t notice either until I was nearly done editing, so I just decided to ignore it 😂

  • @toftul
    @toftul Před 7 dny

    Have you been thinking about AI kit for RPi5? It could potentially boost the workflow with neural networks. I guess it would be MUCH cheaper then renting GPU on the cloud

    • @jacstrong
      @jacstrong  Před 6 dny +1

      I have. Time to get myself one :)

  • @dustsucker4704
    @dustsucker4704 Před 10 dny

    I think it comes down to how you structured your Services and how much latency is tolarated i mean for one of the 16gb 8 core Servers you could buy yourself like 8 raspberry pis and Cluster them up and with a message que you could just make the work a little more distributed. And another tip look at some of the Intel nucs or the zima Board they are cheap to and run x86. Im currently facing a similar Probleme where i have 3 Servers from a cheap hoster and they are great but I spend about 35€ every month for them and that's Kind a expensive for a little side project.

    • @jacstrong
      @jacstrong  Před 10 dny +1

      I totally agree and that’s why I did the experiment. Right now the job runs and the user has to wait for it to finish just because that was the easier way to program it. The job actually has about 6 minutes before it’s actually used again in the game. A little bit of rework and a single Pi could support probably 15 concurrent games. Hopefully that’s a problem I have :)
      I also want to try the Pi with an accelerator, adding a little TPU would be interesting. Zima board would be interesting too.