Retro banner ad: SC26 Field Guide — Chicago, Nov 15–20 — HPC news that scales. Click here.

Fresh catch

223

items this week

In the mix

17 vendor · 57 papers

105 press · 26 events

Comments from the fleshy human

This was a slow week in the news, nothing really major happened, at all. Everything is fine. But, SUPERCOMPUTING CONFERENCE IS HERE!!!! See our SC26 Field Guide; it is mobile friendly and gives you a nice overview of how the week will go. Also includes a party list :) Oh, and I'll bring some stickers to SC26. See you there!

Latest pull

A daily mix, with vendor runs broken up

HPCwirePenguin launched a CXL card that lets AI servers reuse DDR4 memoryPenguin says the eight-DIMM card uses a Marvell CXL controller to let AI servers repurpose older DDR4 memory.
IEEE XploreResearchers analyzed long AI training runs on shared HPC clustersThey found CPU throughput dropped up to 30% from throttling and non-exclusive GPU jobs halved in speed when nodes were shared.
Supercomputing NewsAMD and Cornelis partnered to sell EPYC systems with Omni-Path networkingCornelis launched 800G CN6000 networking, claiming 21% lower MPI latency than Nvidia's XDR InfiniBand.
HPCwireQuiX Quantum secured Dutch funding to build a 100-qubit photonic computerThe consortium will develop a 100+ qubit photonic quantum platform by 2028, targeting control ASICs with latencies below 10 ns.
IEEE XploreEngineers designed a system to self-heal DFT workflows across HPC clustersThey report the system recovered 731 out of 1,677 workflow errors across HPC centers while keeping LLMs out of direct job execution.
Data Center DynamicsMicrosoft signed a deal to deploy Cerebras AI chipsMicrosoft signed on as Cerebras' third hyperscaler customer alongside AWS and Oracle, with full deployment details promised in 2027.
HPCwireAI data center demand sparks a US-China race for nuclear powerA Brookings paper estimates that 188 GW of US data center capacity by 2032 could match the entire US residential sector's annual power use.
IEEE XploreStudy finds open-source foundations rarely fund scientific softwareA survey of DOE HPC software projects found direct funding was the rarest benefit of joining open-source foundations, despite being developers' top concern.
Oak Ridge (OLCF)Researchers ran 5.4-trillion-particle galaxy simulations on Aurora and FrontierThe Gordon Bell finalist team scaled simulations across 10,040 Aurora nodes, reporting 976 petaflops in FP32 on their gravity kernel.
HPCwireHPE launched hybrid Cray storage and a rack cooling controllerHPE says the hybrid E2000 arrays cost 82% less than all-flash while pushing up to 116 GB/s of read throughput per rack.
IEEE XploreResearchers built CGSim to model large-scale distributed grid computingBuilt on SimGrid, the tool simulates up to 50 distributed sites running tens of thousands of concurrent ATLAS PanDA jobs on commodity hardware.
LAMMPSLAMMPS released patch_13Nov2026Classical molecular dynamics engine for materials — atoms, polymers, metals and granular systems at very large scale. Release notes for patch_13Nov2026 are published upstream.
HPCwireLANL moved seven HPC clusters to OpenCHAMI managementLos Alamos moved seven production clusters with over 11,000 compute nodes to the open-source OpenCHAMI systems-management framework.
IEEE XploreResearchers built a workflow to incrementally port HPC climate models to GPUsThey tested the approach on PALM-4U's pressure solver, achieving a 65x kernel speedup and 10x gain over multi-core CPUs while verifying accuracy.
Supercomputing NewsGoogle paired TPU 8tx accelerators with AMD EPYC CPUs for confidential computeGoogle swapped its Axion CPUs for AMD EPYC chips to add confidential computing to TPU 8tx, claiming up to 2x lower latency than TPU 8t.
HPCwireBSC outlined its technical lineup and research sessions for SCResearchers from Barcelona Supercomputing Center are presenting work on SmartNICs, RISC-V, memory systems, and quantum computing at the conference.

The Daily

FLOPBETA

HPC news that scales.

HPC, Scientific Computing, and AI News

Saturday, November 14, 2026

Flop around, find out.

COMMENTS FROM THE FLESHY HUMAN

This was a slow week in the news, nothing really major happened, at all. Everything is fine.

But, SUPERCOMPUTING CONFERENCE IS HERE!!!! See our SC26 Field Guide; it is mobile friendly and gives you a nice overview of how the week will go. Also includes a party list :) Oh, and I'll bring some stickers to SC26. See you there!

Fresh catch

last 7 days

17
vendor announcements: 105
news articles: 57
papers: 30
releases: 26
events this week: 223
items indexed

Top 5 things

Data Center DynamicsJensen Huang resigned as Nvidia CEO

Argonne (ALCF)RIKEN team modeled 5.4 trillion particles on Aurora and Frontier IEEE XploreResearchers built adaptive data transport for continuum workflow engines

SupermicroSupermicro ships H15 servers with AMD EPYC 9006 CPUs

HPCwireQuiX Quantum secured Dutch funding to build a 100-qubit photonic computer

Security risks

Upcoming events

Full calendar →

Hardware announcements

AMD and Cornelis partnered to sell EPYC systems with Omni-Path networking Cornelis launched 800G CN6000 networking, claiming 21% lower MPI latency than Nvidia's XDR InfiniBand.
Syntys started building a 20MW data center in Qatar Syntys says the carrier-neutral facility will support high-density configurations for GPU-accelerated AI and HPC workloads.
Microsoft signed a deal to deploy Cerebras AI chips Microsoft signed on as Cerebras' third hyperscaler customer alongside AWS and Oracle, with full deployment details promised in 2027.
Penguin launched a CXL card that lets AI servers reuse DDR4 memory Penguin says the eight-DIMM card uses a Marvell CXL controller to let AI servers repurpose older DDR4 memory.
Marvell showed optical and CXL interconnects for AI clusters They demonstrated optical fabrics, CXL memory expansion, and 1.6T active electrical cables to connect multi-rack AI systems.

Top stories

Latest pull

HPC, Scientific Computing, and AI News

Fresh catch

Top 5 things

Security risks

Upcoming events

Hardware announcements

The latest pull