from Hacker News

Better performance per watt, but why ARM server market share not grown over x86

by r3tr0 on 3/3/23, 3:53 AM with 6 comments

• by senttoschool on 3/3/23, 4:04 AM

  In 2020, 50% of all new EC2 instances on AWS were using Graviton2, an ARM CPU.[0]

  Today, it would not surprise me if 80% of all new EC2 instances are using Graviton3 and Graviton2.

  [0]https://images.anandtech.com/doci/16640/Neoverse_Intro_10.pn...

  So the rush actually already happened on AWS. Other cloud providers are behind in the ARM transition. Microsoft, for example, has secured ARM server chips from Ampere but does not have an in house ARM server chip. I think this was a mistake for Microsoft. I think they should buy Ampere to accelerate their custom ARM server chips.

  For smaller cloud providers, they will have to wait because the cost to switch to ARM CPUs can be too great and too risky at the moment. So it's safer for them to keep buying AMD Epyc or Intel Xeon chips.

• by rozenmd on 3/3/23, 6:52 AM

  I've rushed to move all my code onto ARM (that I could). The trouble was pesky binaries that don't have ARM builds - I'm not spending the time to build them myself and then verify they work on ARM, when they already work on x86.

• by Am4TIfIsER0ppos on 3/3/23, 10:10 PM

  If you can do 1 thing for 1 power on arm and 1 thing for 2 power on x86 but arm maxes out at 2 things per time and x86 at 3 things per time are you going to spend more time on it in exchange for less power?

• by cable2600 on 3/3/23, 4:38 AM

  Because there is too much legacy code for X86 system servers on the market. It costs a lot to hire developers to port the code. Some of the code is for languages that are so old they don't convert very well.