• 0 Posts
  • 20 Comments
Joined 1 year ago
cake
Cake day: August 16th, 2023

help-circle


  • The info graphic suggests that they use the different cooling rate of the first and second layer to lock in the applied magnetic field of heat assisted magnetic recording.

    They beat both layers

    They apply a magnetic field to save a “1” bit

    Both layers are magnetized to a “1” bit

    The first layer cools down and locks that bit into place permanently.

    They apply the magnetic field to save a “0” bit

    While the second layer is still hot and accepts the “0” orientation of the magnetic field, the first layer is already too cold and will not change its magnetization.

    The second layer cools down and locks that bit into place.

    Neat!













  • Edit: whoops, got a little bit sidetracked and didn’t talk about cloudflare at all. I’ll leave it up nonetheless as it contains info.

    The reverse proxy only listens on port 80 and 443, so yes, all your services will be accessible through just one/two ports.

    The reverse proxy will parse the http request headers and ask the appropriate upstream service (e.g. jellyfin) on localhost:12345 what it should send as a reply. Yes, this means that you need to have a http header so that the reverse proxy can differentiate the services. You don’t need to buy a domain for that, you can use iPhone to make your made up domain map to a local IP address, but you need to call the reverse proxy as sub.domain.com. 192.168.0.123:80 won’t work, because the proxy has no idea which service you want to reach.

    I found it really easy to set up with docker compose and caddy as a reverse proxy. Docker services on the same network automatically resolve their names so the configuration file for caddy (the reverse proxy) is literally just sub.mydomain.com { reverse_proxy jellyfin:12345 }. This will expose the jellyfin docker, which is listening on port 12345, as sub.mydomain.com on port 80.





  • No, it’s a datum - about how people feel

    Performance numbers are easy to find. The prices have not been great and the 4060 is held back by its reduced memory speed, but it’s a performance increase nevertheless. The flagship product, the one that shows what is currently possible in terms of GPU power, did show remarkable improvement in top performance.

    I’m more salty about AMD not supporting ai workloads on their consumer gpus. Yes, ROCm exists and it will work on quite a few cards, but officially it’s not supported. This is a major reason why Nvidia is still the only serious player in town.