Post B4dJFtHXBI0HhBqpoe by khm@hj.9fs.net
 (DIR) More posts by khm@hj.9fs.net
 (DIR) Post #B4dJFtHXBI0HhBqpoe by khm@hj.9fs.net
       0 likes, 0 repeats
       
       The way I did it was by replacing a 500-teraflop computer with another 500-teraflop computer which was a tenth the size of the old one, instead of assuming I needed to fill the datacenter again.  That's not a common approach in HPC, but for stable workloads it can be done!CC: @wim_v12e@merveilles.town @csepp@merveilles.town
       
 (DIR) Post #B4dJQiRJpfWjFqacj2 by csepp@merveilles.town
       0 likes, 0 repeats
       
       Is there such a thing as high-performance #permacomputing ? :flan_think: On one hand, HPC implies centralization.  On the other hand, some important tasks like weather prediction or build farms benefit everyone and require centralized infrastructure.
       
 (DIR) Post #B4dJQigutg0I2Dt52O by skyfaller@jawns.club
       0 likes, 0 repeats
       
       @csepp Do they require centralized infrastructure? Remember Seti @ home and folding @ home?https://en.wikipedia.org/wiki/SETI%40homehttps://en.wikipedia.org/wiki/Folding%40home
       
 (DIR) Post #B4dJQiscCBMScVMQGu by khm@hj.9fs.net
       0 likes, 0 repeats
       
       HPC and distributed computing serve different needs.  HPC is for when your massively-parallel job is tightly coupled.  Distribution works better with embarassingly parallel work.CC: @csepp@merveilles.town