Andreas Herten

@andih
317 Followers
271 Following
177 Posts
My web is my castle • High-Performance Computing with GPUs at Jülich Supercomputing Centre
Lab Bloghttps://x-dev.pages.jsc.fz-juelich.de/
Personal Websitehttps://www.andreasherten.de
Not sure I even have German followers here, but: Zusammen mit @bne86 habe in einem Podcast über #JUPITER berichtet: https://codeforthought.buzzsprout.com/1326658/episodes/18342781-de-jupiter-hpc-fur-die-wissenschaft-mit-a-herten-und-b-von-st-vieth
(Follow-Up-Fragen beantworte ich gerne, z.B. hier.) #HPC #exa_JUPITER
[DE] Jupiter HPC für die Wissenschaft - mit A Herten und B von St Vieth - Code for Thought

Deutsche Ausgabe: Anfang September 2025 wurde der neue Hochleistungsrechner Jupiter in Betrieb genommen, mit viel Fanfaren und Prominenz. Zurecht, denn Jupiter ist der schnellste Rechner in Europa. Aber wie kriegt man so eine Maschine am Lauf...

Buzzsprout
@rotnroll666 was he at least proud to share? Humbled by it?
@hattom yes! Good results coming out ☺️
@hattom yes! The machine is pretty good, we are good results coming out. At the moment, we are moving to acceptance.
@hattom it was a few weeks ago. A Friday morning, but not sure which anymore 😬
Finally, #JUPITER crossed the 1 ExaFLOP/s threshold today. The list is lying to you, though, it's not like it's 1000 PFLOP/s exactly, it's 1000.184 PFLOP/s; the rest got lost to rounding.
The 184 TFLOP/s are pretty much exactly the same as the previous #JSC supercomputer JUGENE from 2008. Incredible, that this gets lost in the margins only 17 years later.
https://mastodon.social/deck/@fzj_jsc@social.fz-juelich.de/115566616685186256
#HPC #exa_JUPITER
@filipesmg I like it; teaffee. But you’re right of course.
Has anyone already determined the average coffee density at #SC25? Need to normalize to ICS* to align my coffee intake. #HPC**
*: International Coffee Standard
**: The C is for Coffee, of course
@glennklockwood It slightly incentivizes on the wrong thing, though. In terms of power-per-compute ratio, you can get these sweet Watts most efficiently by making a rack of hairdryers. 28 x 2 kW = 56 kW needed for an adder! Instant Gigascale!
But at least Watts are easily understandable by basically anyone… FLOP/s is not only inflated but also nothing from everyone’s day-to-day business.
There a nonsensical performance metrics in computing – and then there’s “Gigawatts of Compute”. #HPC