Conversation

Notices

1. kaniini (kaniini@mastodon.dereferenced.org)'s status on Monday, 02-Oct-2017 23:00:41 EDT kaniini

   my thoughts on ARM:

   pricing ranges from:

   - cheap and decently powered for the price point (raspberry pi)

   - moderately more expensive (~$250) and slightly underpowered with non-free blobs required to boot the bloody thing (nvidia jetson and the like)

   - very expensive and a complete performance disappointment in the real world due to broken memory coherency models (thunderx, appliedx-gene, etc.)

   ARM isn't going to make it in the datacentre if they can't compete on performance.

   1. kaniini (kaniini@mastodon.dereferenced.org)'s status on Monday, 02-Oct-2017 23:03:30 EDT kaniini

      in reply to

      which is why ARM is back to being a ghetto in basically every distro ever: people believed the hype that ARM bigiron would be great for a while, but the performance numbers never got close to lining up with the hype.

      so people who were willing to pay for ARM support quit paying for it and went back to x86 and other stuff like OpenPOWER which actually deliver today.

      qualcomm is making a server cpu with Centriq line but it requires non-free blobs to boot, so why would anybody care?

      1. kaniini (kaniini@mastodon.dereferenced.org)'s status on Monday, 02-Oct-2017 23:08:27 EDT kaniini

         in reply to

         mainline distributions (Debian, Fedora, Alpine, etc.) do not really have a strong ARM interest because the main target of ARM is embedded.

         embedded community have their own options with yocto and buildroot.

         so ARM support basically slides along barely making release qualification in the various distributions while in reality it is rotting at it's core in the distributions because nobody is really hacking on it.

         1. kaniini (kaniini@mastodon.dereferenced.org)'s status on Monday, 02-Oct-2017 23:10:27 EDT kaniini

            in reply to

            and nobody is hacking on it because:

            - there's very little commercial interest in ARM bigiron until the performance numbers actually line up with what is being hyped through benchmarks (more on this later)

            - you can't scale beyond 2GB RAM with commonly available hardware, you usually can't boot from SATA disks, etc.

            - as a result of the scalability wall, ARM isn't really viable in the "home server" role anymore, and stuff like NUCs have come along and crushed them there anyway.

            1. kaniini (kaniini@mastodon.dereferenced.org)'s status on Monday, 02-Oct-2017 23:12:53 EDT kaniini

               in reply to

               so really what ARM Holdings should do is come up with something that is in between the raspberry pi (and clones) and the bigiron offerings.

               if there were ARM machines you could buy that could support up to 16GB to 32GB DDR4 RAM in SO-DIMMs, had SATA and supported either UEFI or SBSA, and had a pricepoint without RAM of around $500 to $600 (the cost of a decent x86 board and cpu), ARM would be in a much better place and it would be an interesting target.

               1. kaniini (kaniini@mastodon.dereferenced.org)'s status on Monday, 02-Oct-2017 23:44:57 EDT kaniini

                  in reply to

                  in essence something like this would cause the same type of transformation at ARM that OpenPOWER did with POWER architecture.

                  competition and the ability to create libre hardware will spur demand if it is actually real.