Community News

The lab outlined its research efforts applying AI across materials science, automated laboratories, and supercomputing systems.

Their SC09 paper showed how adapting sparse matrix data formats to GPU architectures could boost sparse matrix-vector multiplication performance.

The dual-socket MGX nodes pack 88 Olympus cores per chip and will be used to prep software for the upcoming Mission, Vision, and Veritas systems.

Argonne highlighted its computing work, covering the Aurora exascale system, scientific AI, and quantum simulations across its research divisions.

The free online course covers multi-node GPU scaling, NCCL, and NVSHMEM with hands-on access to Cyfronet's Athena cluster.

The free virtual series covers tools like Spack, Flux, and RAJA with hands-on exercises run on provided AWS instances.

His background checkpointing system speeds LLM training by up to 2.2x with near-zero overhead and was integrated into DeepSpeed.

The three-day hybrid program will cover parallel sparse linear solvers for finite element analysis using MPI, OpenMP, and PETSc.

The study surveyed 34 HPC systems across nine Latin American countries, with Argentina and Brazil hosting over two-thirds of the reported capacity.

CEDAR is near capacity ahead of a planned 2.4 PB expansion in mid-October, prompting admins to ask users to pause data writes while reads continue.

Starting in October 2026, the new MLPerf benchmark tests RLVR post-training on Qwen 3.5 397B designed to run in under 1,024 GB300 hours.

Argonne intern projects evaluated using AI for vaccine discovery, weather radar modeling, and generating parallel code for supercomputers.

The €16.9 million Horizon Europe project aims to build generative AI for disease modeling, with MLCommons handling evaluation benchmarks over four years.

The 27-institution initiative will meet at IEEE Quantum Week 2026 to work on open middleware interfaces for running quantum workloads on HPC systems.

The guide offers strategies for research software maintainers to lower onboarding barriers and recognize non-code contributions.

The initiative links clinical and genomic datasets with HPC analytical tools and federated AI learning across Spain's research network.

Researchers say the open-source tools allow AI agents to configure modeling software like PhysiCell and generate tumor models in under 10 minutes.

Purdue will take Bell, Gilbreth, Rossmann, and Rowdy offline for 18 hours on October 3 for datacenter power work.

The supercomputer will go offline for 12 hours on September 29 to upgrade the operating system and OPA400 network stack.

Queues will go offline for two to four hours starting at 9:00 AM CDT on September 22nd for system maintenance.

The lab is combining AI models and mobile robots to plan, execute, and analyze scientific experiments automatically.

Fermilab will expand its open-source QICK control toolkit as part of Microsoft's new quantum research center and hardware makerspace.

The Latin American supercomputing workshop will cover HPC-enabled digital twins, multi-omics pipelines, and AI-driven clinical trials.

PACE cut per-user queued GPU jobs from 500 to 50 and capped concurrent usage at 32 GPU-hours to fix long queue wait times.

The conference is offering NSF-funded travel grants worth $1,300 each to U.S.-based graduate and undergraduate students.

Cineca posted scheduled maintenance windows for its Leonardo and Pitagora supercomputers alongside resolution of recent Leonardo filesystem issues.

Jie Ren, Amit Samanta, and Xuan Wu won fellowships for research in GPU linear algebra, hybrid cloud scheduling, and data compression.

The two-year project will integrate iPronics optical switching into BSC's GPU clusters to evaluate dynamic networking for LLM and MoE workloads.

Job submission ends October 15 after seven years of service, with the file systems remaining accessible for four months for data retrieval.

Global HPC centers met in Cologne and Jülich to discuss accelerated computing, agentic AI workloads, and tour the JUPITER supercomputer.

They scaled the NekRS code across 9,000 nodes on Aurora and Frontier to simulate a liquid-metal fusion breeding blanket.

The researchers modeled 500 protein chains on GPU nodes to trace how staged folding triggers disease-causing fibril formation.

Honorees include Dru Renner for primary simulation modeling and Earl Lawrence for national security AI and machine learning leadership.

The agency is seeking teams to develop manufacturing platforms, PDKs, and full-chip demos for low-dimensional materials targeting HPC and AI.

She will lead work on integrating quantum processors into ALCF's heterogeneous HPC and AI infrastructure for scientific computing.

The initiative is gathering startup requirements for exascale AI services and offering GPU compute credits to survey respondents.

The AIRIS project is porting itwinai-based distributed ML workflows to the JUPITER Booster to scale multimodal biomedical AI across EuroHPC systems.

Georgia Tech is adjusting compute rates on its Phoenix and Firebird clusters starting October 1, 2026, while prorating existing prepaid credits.

The agency is seeking proposals for physical AI and previewed upcoming research topics including computing beyond conventional CMOS limits.

CUNY is using NSF ACCESS allocations on Indiana University's Jetstream2 system to host online cryptography homework for 150 students.

OSC is draining and rebooting nodes across all three clusters to upgrade NFS mounts to NFSv4.2 and fix login stability.

The system will feature AMD CPUs, accelerators, and an all-NVMe file system when it enters production in 2027.

Engineers modified NVIDIA FLARE to run on Frontier's AMD GPUs and trained ResNet across sites using data quantization to manage latency.

The annual meeting will feature tutorials, user presentations, and technical sessions for researchers using TACC compute systems.

NCSA's Healthcare Innovation Program Office bridges clinical researchers and supercomputing teams to build AI workflows and data pipelines for health tools.

The PACT 2026 co-located meeting aims to address data silos stranded between experimental instruments, AI pipelines, and computing platforms.

The IEEE TCSC SCALE Challenge at CCGrid 2027 is now accepting submissions demonstrating scalability across HPC and scientific computing.

The 2027–2028 groups will advise the board on data sovereignty, health data, semantic interoperability, FAIR data and AI, and long-term retention.

CSCS is deploying an IBM Quantum System Two with Nighthawk chips in the same facility as Alps, aiming for operation by the end of 2026.

The three-year project will install a Nighthawk-powered quantum machine alongside the Alps supercomputer at CSCS in Lugano.

IQM will deliver a 150-qubit system to Kajaani in 2027, upgrading it through 2029 to enable hybrid HPC and error-corrected quantum workflows.

The project estimates Earth's total installed compute at 82 zettaflops of peak dense BF16, equivalent to nearly 83 million Nvidia H100 GPUs.

The team combined quantum and classical computing to simulate a 12,635-atom biomolecule.

The summer internship program embedded students at centers like NCSA and Purdue to train them in HPC operations, networking, and user support.

The €68.5 million HPE system delivers over 60 petaflops and ranked 42nd on the June 2026 TOP500 list.

Admins are working to fix login and queue submission issues across Stampede3, Vista, Lonestar6, and the Analysis Portal.

JSC runs around 30 hands-on training courses each year covering simulation, data analytics, and AI on HPC systems.

About 300 researchers gathered in Tokyo to discuss formal reasoning, using AI to solve math problems, and applying math theory to AI reliability.

Volume 18 of the CASC newsletter highlights recent advancements in asynchronous communication for HPC workloads and the application of large language models to boost compiler optimization.

The toolkit brings hybrid quantum algorithms to Fugaku and the ROQUO GPU cluster to connect them directly to quantum systems.

Researchers used SDSC Expanse, PSC Bridges-2, and TACC Stampede3 to match quantum simulations to laser experiments on helium.

The lab highlighted science and engineering workloads running on supercomputers with tens of thousands of GPUs.

Purdue will decommission Bell in October 2026 and migrate users to Gautschi, which features AMD Genoa CPUs and NVIDIA H100 GPUs.

Users must manually move their files to the new Ranch hardware before staff remove symbolic links to the old system on November 1, 2026.

The internship offers students funded research slots across three units covering scientific networking, data infrastructure, and physics simulations.

The lab will head six projects and join 17 others using AI and supercomputing to speed up scientific modeling and national security research.

They claim replacing conventional equations with neural networks scales quantum many-body calculations from a dozen interacting particles up to 100.

Organizers must submit in-cooperation requests four weeks before CFPs or apply for full sponsorship at least a year in advance.

AMD says physical simulations and validation will keep FP64 precision essential even as AI workloads expand.

The talk covers how physical AI models on 3D meshes and point clouds require different cloud compute and memory architectures than LLMs.

The software connects quantum systems to Fugaku and the 540-Blackwell-GPU ROQUO cluster to run hybrid algorithms.

ALCF shared details on AskALCF, a tool using agentic AI to handle support requests autonomously.

Berkeley Lab and USC researchers used BerkeleyGW on GPU supercomputers to run many-body excited-state calculations on systems with 10,000 atoms.

After 15 years, the UK organization has renamed itself the Institute for Research Software to focus on sustaining scientific code and RSE careers.

Researchers mapped QAOA to a spin-boson system to evaluate large quantum circuit performance on ALCF and NERSC systems without running full circuits.

They say the 500-node system will pack over 4,000 MI355X GPUs with 1.1 PB of HBM to deliver 317 PFLOPS of double-precision performance.

The cluster and related services will be completely inaccessible during the maintenance window while updates are completed.

The five-year, $35 million award will fund a unified national portal to coordinate AI compute and data access across academic and industry providers.

The center will provide researchers with shared access to large-scale AI compute, datasets, and specialized tools.