The microblog: 2022.07.01 09:00:04

2022.07.01 09:00:04 (1542764961281110016) from Daniel J. Bernstein, replying to "Erik Tews (@e_tews)" (1542700264024670208):

This is the multicore era. The baseline is a system with, say, 8 cores. Instead invest in 12 cores, and then recoup the investment through the power savings of running at lower speed. This generally produces better speeds for lower cost. Also, the hardware tends to last longer.


2022.06.30 20:26:58 (1542575436265316353) from "Aleksey Shipilëv (@shipilev)":

Reminder: if your goal is to conserve energy per job (heavy-duty workstations, build servers running at 100% CPU, 24/7) and peak power (home servers with silent air cooling), do not let your CPUs boost up. The perf-per-watt efficiency peak is at lower frequencies.

2022.06.30 20:26:58 (1542575438739939329) from "Aleksey Shipilëv (@shipilev)", replying to "Aleksey Shipilëv (@shipilev)" (1542575436265316353):

Case in point, TR 3970X, OpenJDK build + tier1 testing: - default: ~24.5 minutes, 420W peak power, 540 kJ per job - CoreBoost off: ~26 mins, 337W peak, 410 kJ - CB off, PPT=125W: ~28.5 mins, 210W peak, 340 kJ So: - 1.06x slower, 1.31x cheaper - 1.16x slower, 1.59x cheaper

2022.07.01 04:42:59 (1542700264024670208) from "Erik Tews (@e_tews)", replying to "Aleksey Shipilëv (@shipilev)" (1542575438739939329):

Assume I run a ci/cd pipeline for such builds on my own hardware. When would getting two systems in "low power" mode pay off compared to a single one in "high power" mode?