Hands On With The Most Open-Source, High-Performance System For 2018

[gmturner]

I made some limited experiments a year or so ago (maybe closer to two years) using linux on big blue ~100 thread power9 as a cross compiler and my takeaway was that it was pretty damn sweet. The individual cores were damn slow, so if something serialized like autotools came up there could be some thumb twiddling. But if you found/created lots of parallelism it seemed (subjectively speaking) to scale build performance into the parallelism dimensions more gracefully than consumer-grade x86 (these days I may no longer have an accurate intuition about what datacenter x86 bare-metal feels like but if you put a gun to my head and forced me to guess, I'd say it likely scaled better than that too).

[MaxToTheMax]

I made some limited experiments a year or so ago (maybe closer to two years) using linux on big blue ~100 thread power9 as a cross compiler and my takeaway was that it was pretty damn sweet. The individual cores were damn slow, so if something serialized like autotools came up there could be some thumb twiddling. But if you found/created lots of parallelism it seemed (subjectively speaking) to scale build performance into the parallelism dimensions more gracefully than consumer-grade x86 (these days I may no longer have an accurate intuition about what datacenter x86 bare-metal feels like but if you put a gun to my head and forced me to guess, I'd say it likely scaled better than that too).

Yeah, X86 cross compilation would be an interesting thing to benchmark. I.e. how fast can this machine build an X86 kernel? Something like a Talos II could make a great icecream node.