techhub.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
A hub primarily for passionate technologists, but everyone is welcome

Administered by:

Server stats:

4.6K
active users

#isc25

0 posts0 participants0 posts today

Scott Atchley, who co-keynoted #ISC25, posted a really meaningful response to my ISC25 recap blog post on LinkedIn (linkedin.com/posts/scottatchle). He specifically offered additional perspective on the 20 MW exascale milestone and the pitfalls of Ozaki. It's short but very valuable context.

www.linkedin.comI always enjoy reading Glenn K. | Scott AtchleyI always enjoy reading Glenn K. Lockwood's conference recaps with #ISC25 recap being his latest. First, I was honored that AMD's Mark Papermaster invited me to share some science highlights from OLCF's Frontier. I shared that Frontier has grown slightly in the last year with the integration of the test and development system into the full system. My science highlights included: • GE Aerospace's efforts to reduce noise generated by their RISE engine that will allow GE Aero to bring it to market sooner, • NASA's work to understand how to use retro-propulsion to land humans and their gear on Mars, • Researchers refining the phase diagram for carbon by identifying the narrow region in pressure and temperature that would allow body-centric cubic (BC8) carbon to exist. This material is expected to be 30% harder than diamond, and • Efforts to understand how drug candidates interact with proteins. Unlike AI efforts such as AlphaFold that approximate protein docking, this effort uses Molecular Dynamics to got the two items close together then it switches to quantum mechanics to get an exact docking. This application actually used over 1 exaflop (1 EF) of full precision (FP64) on Frontier. I also highlighted what Frontier's replacement, Discovery, will need to support modeling/simulation as well as artificial intelligence. It will need bandwidth everywhere from processors to scale-up bandwidth between processors to scale-out bandwidth across the whole system in addition to lots of high-precision and low-precision FLOPS. I will reply with more comments. 🧵 #OLCF #Frontier

Looking at the Top500 list from #ISC25, it looks like JUPITER ran HPL with 5,884 nodes--perhaps the full system (advertised as "roughly 6,000" nodes).The run only hit ~33 TF/H200 tho; by comparison, Isambard-AI hit 43 TF/H200. If JUPITER can get up to 43 TF/GPU, it should post at 1.01 EF FP64. #HPC

What an amazing exhibition experience last week was! 😀 ISC High Performance was an adventure, and we are delighted that we were able to welcome visitors not only at our joint GCS stand, but also at our⚡JUPITER booth. 💪 Find out all about our experiences, activities and successes right here: fz-juelich.de/en/ias/jsc/news/
#ISC25! #JUPITER #HPC #Supercomputing #Quantumcomputing #FZJ

www.fz-juelich.de · ISC25: Action-packed days dedicated to JUPITER, quantum power, and networkingSuccessful exhibition performance by JSC at ISC25 with JUPITER and GCS booths
Continued thread

The #isc25 is over and I half-recovered from the weekend, too. Time to continue my thread summing up the #SnakemakeHackathon2025 !

To me, an important contribution was from Michael Jahn from the Charpentier Lab: A complete re-design of the workflow catalogue. Have a look: snakemake.github.io/snakemake- - findability of ready-to-use workflows has greatly improved! Also, the description on how to contribute is now easy to find.

A detailed description has been published in the #researchequals collection researchequals.com/collections under doi.org/10.5281/zenodo.1557464

snakemake.github.ioSnakemake workflow catalog | Snakemake worklow catalog

Wir sind noch ganz im Zeichen der #ISC25 und haben diesen Artikel über die Entwicklung von Rechenkraft gefunden: Selbst mit Beschleunigern und anderen Acceleratoren ist das #MoorschesGesetz nicht mehr zu schaffen. Bis vor einigen Jahren konnte Rechenkraft noch in etwa zwei Jahren verdoppelt werden. Die Architektur fürs Supercomputing muss auch für eine nachhaltige Energienutzung komplexer, heterogener geplant werden. nextplatform.com/2025/06/10/to

Dr. Lin Gan von der chinesischen Tsinghua University zitierte auf der #isc25 in Hamburg Forschungsergebnisse mit Skalierung bis zu 41 Millionen Cores. In den Papers ist von 107520 Nodes die Rede, macht 390 Cores pro Node auf einem (Nach-)Nachfolger des Sunway Taihu Light Supercomputers. Der steht mit rund 93 PFLOPS aus 10,6 Mio. Kernen derzeit auf Platz 21 der Top500.

Continued thread

Interesting take from an #Nvidia engineer I met at #ISC25: "Do you need IEEE-754 compliant #FP64, or do you need digits? Digits we can get you through FP64 emulation."

Not sure what to make of that. Things were an absolute mess before the IEEE-754 standard, and I wouldn't want to ever go back to that. No standards means you cannot at all port software between hardware architectures even from within the same vendor. Having to re-architect software for each and every new chip is not gonna happen.

Continued thread

Returning from the #isc25 I will continue this thread with something applicable everywhere, not just on #HPC clusters:

Workflow runs can crash. There are a number of possible reasons. Snakemake offers a `--rerun-incomple` flag (or short `--ri`) which lets a user resume a workflow.

This contribution from Filipe G. Viera describes a small fix to stabilize the feature. Not only will incomplete files be removed after a crash, now it is ensured that all metadata with them are deleted too, before resuming: zenodo.org/records/15490098

ZenodoMetadata Cleanup for Snakemake

Waiting for my train and going through slides of a BOF from the #isc25 . Apparently, students prefer tutored online courses.

I as a lecturer always prefer onsite courses. Spiced with lots of exercises. Such that I get an immediate impression of proceedings and reception of the material.

Yes, I understand the need-to-travel overhead when living in somewhat remote areas. On the other hand: All those demanding online or even hybrid courses - have you considered that the average participant might learn less? The literature supporting online courses at least is rather underwhelming.

Plus, actually meeting people is way better than chatting online. Same with conferences. I just me people from @SURF and other places whom I only met online before. Such a difference!

Then again, I only travel by train - which cannot expected by everyone ...

#online vs. #onsite #teaching #conferences #AcademicChatter