1. 36
  1. 9

    Do people even understand what implications this has on server space? I think Amazon and larger cloud players are gonna start investing into ARM space for their hosted databases. This is both power and speed efficient, even if we start saving a penny per hour imagine the impact this will have. We are going back to the world of fragmented CPUs and x86 monopoly is coming to an end?

    1. 25

      There are multiple synergetic trends leading this way:

      1. Intel has wasted the last decade
      2. AMD is back in the game, the Ryzen series are very competitive
      3. Apple M1 also seems to be capable
      4. NVidia attempting to buy ARM makes many vendors consider alternatives (SPARC, RISC-V, MIPS)
      5. USA embargo against China won’t go away, makes Chinese companies, also other smaller powers refresh their search for alternative independent computing solutions
      6. Privacy concerns about mandatory “security co-processors” and state and ad-industry surveillance make many of us wait (wish?) for commercial availability of completely open solutions like RISC-V, despite their uncompetitive performance in the short-to-mid term.
      1. 12

        Next week is the two-year anniversary of ARM instances on EC2; if they weren’t researching RDS or other services on ARM long before that public release that’s severe managerial malpractice on their part.

        I ran Postgres on an a1.2xlarge for a few days over the summer, replicating from an on-prem x86-64 Postgres. No perf numbers, but it worked great.

        1. 3

          I wish Amazon was selling laptops with ARM processors running Linux.

        2. 4

          I hope! This would also help reduce the impact of computing on ecology.

          1. 3

            ARM vendors (apart from Apple) so far have been focused on high-volume low-end cheap CPUs. Many startups trying to make ARM servers didn’t make it past a buggy demo board. I’m surprised that AWS managed to pull of Graviton.

            But maybe it will get better. Lack of a good developer machine has been a barrier as well. The choice only was between Pi-like toy computers and ridiculously overpriced custom boards that weren’t that much faster anyway.

            1. 1

              That’s not quite true. Some very expensive ARM cores power several supercomputers. e.g. the A64FX at ~$20,000 a go. The EU Funded Human Brain Project built several huge ARM machines, too.

            2. 2

              AWS moved ElastiCache instances to arm recently.

            3. 3

              Would be interesting to see performance compared while taking into account temperatures. Both Intel and AMD chips are heavily dependent on temperature for reaching and maintaining boost clocks.

              This article compares the M1 against previous generation MacBooks, which optimise for compactness and quietness over performance. An Intel chip in a properly cooled system would perform better. They also include Ryzen desktop CPUs into the mix, which had to have been cooled using conventional desktop parts.

              These numbers are useful for people choosing between MacBooks but in terms of actual performance they could be misleading.

              1. 3

                As far as I can tell, the article does not discuss the most important component here: the secondary storage itself. Some active benchmarking is needed here before any conclusions can be drawn.

                The new M1 apparently has an SSD that is nearly twice as fast as a previous gen mac, but its numbers are typical for an NVMe drive. I rather suspect that normalizing for storage performance would render the M1 CPU itself uninteresting.

                1. 2

                  The new M1 apparently has an SSD that is nearly twice as fast as a previous gen Mac

                  only on the MacBook Air which previously had an SSD that was 2x sower than what all other Macs had

                  1. 1

                    Fair enough. Point stands though.