🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

In this paper is presented Swing, a new algorithm to improve allreduce performance on torus networks which performs a logarithmic number of steps and transmits the minimal number of bytes while reducing the distance between communicating nodes. #HPC arxiv.org/pdf/2401.09356…

Check out how to accelerate allreduce and other collectives on multidimensional torus networks and HammingMesh using Swing. Up to 3x performance improvement on different topologies! Daniele De Sensi will present it at USENIX Association #NSDI24. arxiv.org/abs/2401.09356

Are you passionate about #HPC, #GPU programming, and #numerical algorithms? Do you want tohelp us develop a new generation of computational tools to democratize access to massively parallel computing systems? A #PhD #position is now open at VU Amsterdam! workingat.vu.nl/vacancies/phd-…

We have open PhD/PostDoc positions👩‍🎓on workload acceleration using SmartNICs. You can find more details at hipeac.net/jobs/14559/phd… Submission deadline on March 16th, 2024. Feel free to drop me a message✉️if you need more information.

Passionate about #HPC? 📍The 2nd CINI Summer School on High Performance Computing and Emerging Technologies #HPC2024 will be held on 17-21 June 2024 Università di Trento, Italy ℹ hpc-summer-school-24.disi.unitn.it/home @hicrestlab Consorzio CINI Cineca SuperComputing ICSC E4 Computer Engineering

Happening right now at #NSDI👉 Nils Blach presenting "A High-Performance Design, Implementation, Deployment, and Evaluation of The Slim Fly Network" and Daniele De Sensi presenting his paper "Swing: Short-cutting Rings for Higher Bandwidth Allreduce" Torsten Hoefler 🇨🇭 ETH CS Department #HPC

How to model performance and accelerate reduction collectives for the Cerebras Wafer-Scale Engine? Learn more from Piotr Luczynski's talk at HPDC 2025! Friday 7th June, at 2 PM. Paper: arxiv.org/abs/2404.15888

Yutong Lu gave a great talk about #HPC interconnect optimization in Cetraro. She shows optimized allreduce for Dragonfly, now onto Slim Fly 😉. Also consider topology-aware collectives like Daniele's cool Swing algorithm 😎. Closing with extended notifications for #RMA.

HammingMesh - our optimized failure-tolerant network topology for #DeepLearning enables embedding multi-dimensional rings now Communications of the ACM research highlight! cacm.acm.org/research-highl… Low cost, high scheduling flexibility, and remarkable similarity with the later TPUv4 network. #HPC