Vendor Updates
Vendor Updates
Everything the vendors have put out. Pick a vendor to catch up on their run of announcements.
Nov 20, 20261
- Dell Technologies · 2d agoFirebird picked Dell to expand its AI compute cluster in Armenia
The buildout follows phase one in Hrazdan, which uses Dell PowerEdge systems with Nvidia Blackwell GPUs for training and inference.
Read originaldell.com/en-us/newsroom/announcements/firebird-selects-dell-technologies-to-support-next-phase-of-ai-infrastructure-expansion-in-armeniaDell Technologies · 2d ago
Firebird picked Dell to expand its AI compute cluster in Armenia
The buildout follows phase one in Hrazdan, which uses Dell PowerEdge systems with Nvidia Blackwell GPUs for training and inference.
Read original https://dell.com/en-us/newsroom/announcements/firebird-selects-dell-technologies-to-support-next-phase-of-ai-infrastructure-expansion-in-armeniaserverdatacenterai-infrastructureserverdatacenterai-infrastructure
Nov 18, 20263
- Supermicro · 4d agoSupermicro began shipping all-flash storage servers with PCIe Gen 6
Powered by AMD EPYC 9006 CPUs, the 2U systems pack up to 3.93 PB and claim double the sequential read bandwidth of PCIe Gen 5.
Read originalsupermicro.com/en/pressreleases/supermicro-ships-first-all-flash-storage-servers-delivering-pcie-gen-6-performanceSupermicro · 4d ago
Supermicro began shipping all-flash storage servers with PCIe Gen 6
Powered by AMD EPYC 9006 CPUs, the 2U systems pack up to 3.93 PB and claim double the sequential read bandwidth of PCIe Gen 5.
Read original https://supermicro.com/en/pressreleases/supermicro-ships-first-all-flash-storage-servers-delivering-pcie-gen-6-performancestoragepcieflashserverstoragepcieflashserver - Intel · 4d agoMicrosoft picked Intel EMIB packaging and Diamond Rapids chips for Azure
Microsoft will package custom Azure silicon with Intel EMIB and secure Diamond Rapids CPUs, claiming up to 2.3x performance gains with CXL.
Read originalintel.com/content/www/us/en/newsroom/news/corporate/microsoft-azure-infrastructure-collaboration.htmlIntel · 4d ago
Microsoft picked Intel EMIB packaging and Diamond Rapids chips for Azure
Microsoft will package custom Azure silicon with Intel EMIB and secure Diamond Rapids CPUs, claiming up to 2.3x performance gains with CXL.
Read original https://intel.com/content/www/us/en/newsroom/news/corporate/microsoft-azure-infrastructure-collaboration.htmlno tagsno tags - SambaNova · 4d agoTACC triples its SambaNova inference system with a $5M NSF award
Supported by a $5 million NSF award, TACC is tripling its SambaNova deployment to expand its Sagebrush cluster for scientific AI inference.
Read originalsambanova.ai/press/sambanova-expands-tacc-partnership-tripling-sambastack-deployment-to-accelerate-scientific-aiSambaNova · 4d ago
TACC triples its SambaNova inference system with a $5M NSF award
Supported by a $5 million NSF award, TACC is tripling its SambaNova deployment to expand its Sagebrush cluster for scientific AI inference.
Read original https://sambanova.ai/press/sambanova-expands-tacc-partnership-tripling-sambastack-deployment-to-accelerate-scientific-aihpcdeploymentacceleratoraihpcdeploymentacceleratorai
Nov 17, 20265
- NVIDIA · 5d agoNvidia expanded packaging and substrate supply deals for AI chips
Nvidia announced multi-billion-dollar deals with Kinsus, SPIL, and Amkor to expand substrate and advanced packaging capacity for its accelerator platforms.
Read originalnvidianews.nvidia.com/news/nvidia-expands-commitments-with-supply-chain-partners-to-support-ai-infrastructure-buildoutNVIDIA · 5d ago
Nvidia expanded packaging and substrate supply deals for AI chips
Nvidia announced multi-billion-dollar deals with Kinsus, SPIL, and Amkor to expand substrate and advanced packaging capacity for its accelerator platforms.
Read original https://nvidianews.nvidia.com/news/nvidia-expands-commitments-with-supply-chain-partners-to-support-ai-infrastructure-buildoutno tagsno tags - Micron · 4d agoMicron details LPDDR6 memory features for data center servers
Micron says LPDDR6 brings dual-bank refresh and enhanced on-die ECC to lower power draw and improve reliability in data centers.
Read originalmicron.com/about/blog/memory/dram/lpddr6-enabling-scalable-power-efficient-memory-systems-for-data-centersMicron · 4d ago
Micron details LPDDR6 memory features for data center servers
Micron says LPDDR6 brings dual-bank refresh and enhanced on-die ECC to lower power draw and improve reliability in data centers.
Read original https://micron.com/about/blog/memory/dram/lpddr6-enabling-scalable-power-efficient-memory-systems-for-data-centersno tagsno tags - Qualcomm · 5d agoQualcomm put its Dragonfly AI200 inference systems on Cirrascale Cloud
They claim the liquid-cooled cards pack 768 GB of LPDDR memory each to serve trillion-parameter AI models.
Read originalqualcomm.com/news/releases/2026/11/qualcomm-and-cirrascale-collaborate-to-advance-next-generation--Qualcomm · 5d ago
Qualcomm put its Dragonfly AI200 inference systems on Cirrascale Cloud
They claim the liquid-cooled cards pack 768 GB of LPDDR memory each to serve trillion-parameter AI models.
Read original https://qualcomm.com/news/releases/2026/11/qualcomm-and-cirrascale-collaborate-to-advance-next-generation--no tagsno tags - NVIDIA · 5d agoNVIDIA created a board committee to oversee export controls
The standing committee will oversee global trade compliance and bring in outside experts on national security and export rules.
Read originalnvidianews.nvidia.com/news/nvidia-board-of-directors-establishes-national-security-and-export-control-oversight-committeeNVIDIA · 5d ago
NVIDIA created a board committee to oversee export controls
The standing committee will oversee global trade compliance and bring in outside experts on national security and export rules.
Read original https://nvidianews.nvidia.com/news/nvidia-board-of-directors-establishes-national-security-and-export-control-oversight-committeeexport-controlgovernancehpcai-infrastructureexport-controlgovernancehpcai-infrastructure - DDN · 5d agoDDN will let customers run EXAScaler on third-party hardware
DDN announced a software-only version of EXAScaler 7 to run on qualified standard servers, with a beta planned for early 2027.
Read originalddn.com/press-releases/ddn-introduces-software-defined-exascaler-deployment-expanding-hardware-choice-for-enterprise-ai-infrastructureDDN · 5d ago
DDN will let customers run EXAScaler on third-party hardware
DDN announced a software-only version of EXAScaler 7 to run on qualified standard servers, with a beta planned for early 2027.
Read original https://ddn.com/press-releases/ddn-introduces-software-defined-exascaler-deployment-expanding-hardware-choice-for-enterprise-ai-infrastructurestorageparallel-filesystemexascalerstorageparallel-filesystemexascaler
Nov 16, 20267
- NVIDIA · 6d agoNvidia announced a Vera CPU rack platform for HPC
They claim the liquid-cooled 192-CPU rack delivers up to 2.4x the performance per watt of x86 systems on CFD simulations.
Read originalnvidianews.nvidia.com/news/nvidia-expands-vera-cpu-platform-to-accelerate-scientific-discoveryNVIDIA · 6d ago
Nvidia announced a Vera CPU rack platform for HPC
They claim the liquid-cooled 192-CPU rack delivers up to 2.4x the performance per watt of x86 systems on CFD simulations.
Read original https://nvidianews.nvidia.com/news/nvidia-expands-vera-cpu-platform-to-accelerate-scientific-discoverycpuhpcverasupercomputingcpuhpcverasupercomputing - AMD · 6d agoOak Ridge opened AMD-powered Lux supercomputer for research
Built by HPE with AMD MI355X GPUs and EPYC chips, the cluster ranks 14th on the TOP500 list and is now open for DOE scientific workloads.
Read originalir.amd.com/news-events/press-releases/detail/1307/amd-powered-lux-supercomputer-opens-to-scientific-users-at-oak-ridge-becoming-first-ai-factory-for-does-genesis-missionAMD · 6d ago
Oak Ridge opened AMD-powered Lux supercomputer for research
Built by HPE with AMD MI355X GPUs and EPYC chips, the cluster ranks 14th on the TOP500 list and is now open for DOE scientific workloads.
Read original https://ir.amd.com/news-events/press-releases/detail/1307/amd-powered-lux-supercomputer-opens-to-scientific-users-at-oak-ridge-becoming-first-ai-factory-for-does-genesis-missionsupercomputerhpcepycinstinctsupercomputerhpcepycinstinct - Dell Technologies · 6d agoDell added AMD EPYC servers and Instinct MI455X rack systems
Dell claims its new Helios rack-scale system delivers up to 2.9 exaflops using 72 AMD Instinct MI455X GPUs.
Read originaldell.com/en-us/blog/dell-technologies-and-amd-offer-more-ways-to-deploy-ai-from-first-node-to-full-rackDell Technologies · 6d ago
Dell added AMD EPYC servers and Instinct MI455X rack systems
Dell claims its new Helios rack-scale system delivers up to 2.9 exaflops using 72 AMD Instinct MI455X GPUs.
Read original https://dell.com/en-us/blog/dell-technologies-and-amd-offer-more-ways-to-deploy-ai-from-first-node-to-full-rackserverai-infrastructureacceleratoramdserverai-infrastructureacceleratoramd - NVIDIA · 6d agoNVIDIA, IBM, and Fermilab build quantum-classical codesign platform
They will link DGX B200 systems to IBM quantum hardware and claim a new CUDA-Q algorithm speeds up error-correction decoding by up to 13,000x.
Read originalnvidianews.nvidia.com/news/nvidia-announces-collaboration-with-ibm-and-fermi-national-accelerator-laboratory-to-launch-quantum-classical-codesign-platform-plus-error-correction-breakthrough-to-advance-practical-quantum-computingNVIDIA · 6d ago
NVIDIA, IBM, and Fermilab build quantum-classical codesign platform
They will link DGX B200 systems to IBM quantum hardware and claim a new CUDA-Q algorithm speeds up error-correction decoding by up to 13,000x.
Read original https://nvidianews.nvidia.com/news/nvidia-announces-collaboration-with-ibm-and-fermi-national-accelerator-laboratory-to-launch-quantum-classical-codesign-platform-plus-error-correction-breakthrough-to-advance-practical-quantum-computingquantum-computinghpccuda-qquantum-computinghpccuda-q - AMD · 6d agoAMD finished acquiring spatial AI startup World Labs
The acquisition brings 3D environment model researchers to AMD, with founder Fei-Fei Li joining as chief scientist.
Read originalir.amd.com/news-events/press-releases/detail/1306/amd-completes-acquisition-of-world-labsAMD · 6d ago
AMD finished acquiring spatial AI startup World Labs
The acquisition brings 3D environment model researchers to AMD, with founder Fei-Fei Li joining as chief scientist.
Read original https://ir.amd.com/news-events/press-releases/detail/1306/amd-completes-acquisition-of-world-labsaiacquisitionhpcaiacquisitionhpc - Intel · 6d agoIntel pitched its D270 data center GPU for HPC workloads
They claim the 350W PCIe card packs 160 GB of memory and delivers up to 5.9x higher throughput than Nvidia's RTX Pro 6000 in finance codes.
Read originalintel.com/content/www/us/en/newsroom/news/data-center/why-intel-gpu-d270-belongs-in-the-hpc-conversation.htmlIntel · 6d ago
Intel pitched its D270 data center GPU for HPC workloads
They claim the 350W PCIe card packs 160 GB of memory and delivers up to 5.9x higher throughput than Nvidia's RTX Pro 6000 in finance codes.
Read original https://intel.com/content/www/us/en/newsroom/news/data-center/why-intel-gpu-d270-belongs-in-the-hpc-conversation.htmlgpuhpcacceleratorgpuhpcaccelerator - hgpu.org · 6d agoResearchers benchmarked Nvidia Vera against AMD Turin and Intel Granite Rapids
Bristol researchers found Vera leads across most scientific workloads due to high memory bandwidth, but trails Intel and AMD in compute-bound tasks.
Read originalhgpu.org/?p=31483hgpu.org · 6d ago
Researchers benchmarked Nvidia Vera against AMD Turin and Intel Granite Rapids
Bristol researchers found Vera leads across most scientific workloads due to high memory bandwidth, but trails Intel and AMD in compute-bound tasks.
Read original https://hgpu.org/?p=31483cpuhpcbenchmarkmicroarchitecturecpuhpcbenchmarkmicroarchitecture
Nov 15, 20261
- SambaNova · Nov 15, 2026Intel and SambaNova built an air-cooled system for split AI inference
The setup pairs Intel D-Series GPUs for prefill with SambaNova SN50 RDUs for decode to run AI inference inside air-cooled data centers.
Read originalsambanova.ai/blog/the-air-cooled-token-factory-intel-gpu-d-series-sambarack-sn50SambaNova · Nov 15, 2026
Intel and SambaNova built an air-cooled system for split AI inference
The setup pairs Intel D-Series GPUs for prefill with SambaNova SN50 RDUs for decode to run AI inference inside air-cooled data centers.
Read original https://sambanova.ai/blog/the-air-cooled-token-factory-intel-gpu-d-series-sambarack-sn50serverai-hardwarebenchmarkdatacenterserverai-hardwarebenchmarkdatacenter
Nov 14, 20261
- Supercomputing News · Nov 14, 2026AMD 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.
Read originalsupercomputing.news/hpc/amd-cornelis-joint-supercomputing-program-omni-path-epycSupercomputing News · Nov 14, 2026
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.
Read original https://supercomputing.news/hpc/amd-cornelis-joint-supercomputing-program-omni-path-epycinterconnectepycomni-pathhpcinterconnectepycomni-pathhpc
Nov 13, 20261
- Google Cloud HPC · Nov 13, 2026Google Cloud added agent skills and AlphaEvolve to its HPC toolkit
They paired Cluster Toolkit with reusable agent skills and added an AlphaEvolve solution on Cloud Batch to automate algorithm optimization.
Read originalcloud.google.com/blog/topics/hpc/whats-new-in-hpc-and-ai-at-sc26Google Cloud HPC · Nov 13, 2026
Google Cloud added agent skills and AlphaEvolve to its HPC toolkit
They paired Cluster Toolkit with reusable agent skills and added an AlphaEvolve solution on Cloud Batch to automate algorithm optimization.
Read original https://cloud.google.com/blog/topics/hpc/whats-new-in-hpc-and-ai-at-sc26cloud-hpcai-infrastructuresupercomputingcloud-hpcai-infrastructuresupercomputing
Nov 12, 20262
- NVIDIA · Nov 12, 2026Jensen Huang stepped down as NVIDIA CEO
Jensen Huang has resigned as NVIDIA CEO effective immediately, with CFO Colette Kress taking over as interim chief executive while the board searches for a successor.
Read originalnvidianews.nvidia.com/news/nvidia-announces-ceo-transitionNVIDIA · Nov 12, 2026
Jensen Huang stepped down as NVIDIA CEO
Jensen Huang has resigned as NVIDIA CEO effective immediately, with CFO Colette Kress taking over as interim chief executive while the board searches for a successor.
Read original https://nvidianews.nvidia.com/news/nvidia-announces-ceo-transitiongovernancehpcai-infrastructuregovernancehpcai-infrastructure - DDN · Nov 12, 2026DDN joined the OpenFold Consortium for AI protein prediction
DDN has joined the group behind OpenFold, pitching its Infinia object and POSIX storage platform for AI protein-folding pipelines.
Read originalddn.com/blog/what-ai-protein-structure-prediction-demands-from-data-infrastructureDDN · Nov 12, 2026
DDN joined the OpenFold Consortium for AI protein prediction
DDN has joined the group behind OpenFold, pitching its Infinia object and POSIX storage platform for AI protein-folding pipelines.
Read original https://ddn.com/blog/what-ai-protein-structure-prediction-demands-from-data-infrastructureno tagsno tags
Nov 11, 20264
- Supercomputing News · Nov 11, 2026Google paired TPU 8tx accelerators with AMD EPYC CPUs for confidential compute
Google swapped its Axion CPUs for AMD EPYC chips to add confidential computing to TPU 8tx, claiming up to 2x lower latency than TPU 8t.
Read originalsupercomputing.news/ai/google-tpu-8tx-confidential-computing-root-of-trustSupercomputing News · Nov 11, 2026
Google paired TPU 8tx accelerators with AMD EPYC CPUs for confidential compute
Google swapped its Axion CPUs for AMD EPYC chips to add confidential computing to TPU 8tx, claiming up to 2x lower latency than TPU 8t.
Read original https://supercomputing.news/ai/google-tpu-8tx-confidential-computing-root-of-trusttpuacceleratorconfidential-computinghpctpuacceleratorconfidential-computinghpc - Dell Technologies · Nov 11, 2026University of Alabama is building a statewide HPC center with Dell
The system pairs Dell PowerEdge servers with Nvidia H200 NVL and RTX Pro 6000 GPUs over Quantum-2 InfiniBand for statewide research.
Read originaldell.com/en-us/newsroom/announcements/university-of-alabama-builds-statewide-high-performance-computing-center-with-dell-technologiesDell Technologies · Nov 11, 2026
University of Alabama is building a statewide HPC center with Dell
The system pairs Dell PowerEdge servers with Nvidia H200 NVL and RTX Pro 6000 GPUs over Quantum-2 InfiniBand for statewide research.
Read original https://dell.com/en-us/newsroom/announcements/university-of-alabama-builds-statewide-high-performance-computing-center-with-dell-technologieshpcsupercomputerclusterhpcsupercomputercluster - Supermicro · Nov 11, 2026Supermicro ships H15 servers with AMD EPYC 9006 CPUs
Supermicro claims the new systems deliver up to 33% more cores, 2x PCIe bandwidth, and 1.6 TB/s per socket memory bandwidth.
Read originalsupermicro.com/en/pressreleases/supermicro-now-shipping-next-generation-servers-powered-6th-gen-amd-epyc-9006Supermicro · Nov 11, 2026
Supermicro ships H15 servers with AMD EPYC 9006 CPUs
Supermicro claims the new systems deliver up to 33% more cores, 2x PCIe bandwidth, and 1.6 TB/s per socket memory bandwidth.
Read original https://supermicro.com/en/pressreleases/supermicro-now-shipping-next-generation-servers-powered-6th-gen-amd-epyc-9006servercpuhpcamdservercpuhpcamd - CoreWeave · Nov 11, 2026CoreWeave shares MLPerf benchmarks and cloud infrastructure updates
The blog feed highlights multi-rack Nvidia Vera Rubin MLPerf training results and Blackwell inference benchmark data.
Read originalcoreweave.com/resources/blogsCoreWeave · Nov 11, 2026
CoreWeave shares MLPerf benchmarks and cloud infrastructure updates
The blog feed highlights multi-rack Nvidia Vera Rubin MLPerf training results and Blackwell inference benchmark data.
Read original https://coreweave.com/resources/blogsno tagsno tags
Nov 10, 20263
- NVIDIA · Nov 10, 2026NVIDIA donated OpenShell to CNCF and added tools for GPU Kubernetes
NVIDIA moved its OpenShell agent runtime to CNCF and previewed new open-source tools to manage Kubernetes on GPU clusters.
Read originalblogs.nvidia.com/blog/open-source-openshell-infrastructureNVIDIA · Nov 10, 2026
NVIDIA donated OpenShell to CNCF and added tools for GPU Kubernetes
NVIDIA moved its OpenShell agent runtime to CNCF and previewed new open-source tools to manage Kubernetes on GPU clusters.
Read original https://blogs.nvidia.com/blog/open-source-openshell-infrastructureno tagsno tags - AMD · Nov 10, 2026AMD launched 6th Gen EPYC server processors based on Zen 6
AMD claims its new 256-core EPYC 9996 delivers up to 2.2x the performance and 2x the performance per dollar of Intel's Xeon 6980P.
Read originalir.amd.com/news-events/press-releases/detail/1305/amd-showcases-6th-gen-amd-epyc-processors-to-lead-next-era-of-ai-and-enterprise-computingAMD · Nov 10, 2026
AMD launched 6th Gen EPYC server processors based on Zen 6
AMD claims its new 256-core EPYC 9996 delivers up to 2.2x the performance and 2x the performance per dollar of Intel's Xeon 6980P.
Read original https://ir.amd.com/news-events/press-releases/detail/1305/amd-showcases-6th-gen-amd-epyc-processors-to-lead-next-era-of-ai-and-enterprise-computingcpuepycservercpuepycserver - Supercomputing News · Nov 10, 2026AI agents port Fortran supercomputing codes while teams verify physics
Teams used AI agents to modernize major Fortran codes like GAMESS and FESOM2, but researchers say verifying physics correctness remains the slow part.
Read originalsupercomputing.news/hpc/fortran-hpc-ai-agents-verificationSupercomputing News · Nov 10, 2026
AI agents port Fortran supercomputing codes while teams verify physics
Teams used AI agents to modernize major Fortran codes like GAMESS and FESOM2, but researchers say verifying physics correctness remains the slow part.
Read original https://supercomputing.news/hpc/fortran-hpc-ai-agents-verificationfortrancode-migrationhpcfortrancode-migrationhpc
Nov 9, 20262
- Samsung Semiconductor · Nov 9, 2026Samsung and Nvidia team up on AI physics for chip simulation
Samsung says it is using Nvidia AI physics tools and CUDA libraries to speed up thermal, mechanical, and materials simulations for chip design.
Read originalsemiconductor.samsung.com/news-events/tech-blog/samsung-showcases-ai-driven-semiconductor-engineering-research-and-collaboration-with-nvidia-at-nvidia-ai-day-seoul-2026Samsung Semiconductor · Nov 9, 2026
Samsung and Nvidia team up on AI physics for chip simulation
Samsung says it is using Nvidia AI physics tools and CUDA libraries to speed up thermal, mechanical, and materials simulations for chip design.
Read original https://semiconductor.samsung.com/news-events/tech-blog/samsung-showcases-ai-driven-semiconductor-engineering-research-and-collaboration-with-nvidia-at-nvidia-ai-day-seoul-2026no tagsno tags - NVIDIA · Nov 9, 2026NVIDIA published first MLPerf benchmark results for Vera Rubin
They claim the Vera Rubin NVL72 delivers up to 2x more training performance per GPU than GB300 systems and 10x throughput per watt on agentic workloads.
Read originalblogs.nvidia.com/blog/mlperf-training-endpoints-vera-rubinNVIDIA · Nov 9, 2026
NVIDIA published first MLPerf benchmark results for Vera Rubin
They claim the Vera Rubin NVL72 delivers up to 2x more training performance per GPU than GB300 systems and 10x throughput per watt on agentic workloads.
Read original https://blogs.nvidia.com/blog/mlperf-training-endpoints-vera-rubinbenchmarkgpucpumlperfbenchmarkgpucpumlperf
Nov 8, 20262
- NVIDIA · Nov 8, 2026Rebellions will connect its R200 NPUs to Nvidia systems with NVLink Fusion
Rebellions plans to use NVLink Fusion to hook its R200 inference chips into Nvidia MGX racks for deployment in SK Telecom data centers.
Read originalnvidianews.nvidia.com/news/nvidia-rebellions-and-sk-telecom-announce-collaboration-to-advance-sovereign-ai-infrastructure-with-nvidia-nvlink-fusionNVIDIA · Nov 8, 2026
Rebellions will connect its R200 NPUs to Nvidia systems with NVLink Fusion
Rebellions plans to use NVLink Fusion to hook its R200 inference chips into Nvidia MGX racks for deployment in SK Telecom data centers.
Read original https://nvidianews.nvidia.com/news/nvidia-rebellions-and-sk-telecom-announce-collaboration-to-advance-sovereign-ai-infrastructure-with-nvidia-nvlink-fusionnvlinkinterconnectai-infrastructurenvlinkinterconnectai-infrastructure - hgpu.org · Nov 8, 2026Researchers benchmark AMD MI300A capability gains against MI250X
FP8 on the MI300A ran GEMMs 3.95x faster than MI250X FP16, but unfused quantization caused per-layer slowdowns during LLM inference.
Read originalhgpu.org/?p=31469hgpu.org · Nov 8, 2026
Researchers benchmark AMD MI300A capability gains against MI250X
FP8 on the MI300A ran GEMMs 3.95x faster than MI250X FP16, but unfused quantization caused per-layer slowdowns during LLM inference.
Read original https://hgpu.org/?p=31469gpuamdisaarchitecturegpuamdisaarchitecture
Nov 6, 20261
- Crusoe · Nov 6, 2026Google bought Crusoe's data center site in Wyoming
Crusoe finished pre-development and permitting for the site, which uses a hybrid power plan with fuel cells, grid power, and gas turbines.
Read originalcrusoe.ai/resources/newsroom/crusoe-wyoming-data-center-campus-updateCrusoe · Nov 6, 2026
Google bought Crusoe's data center site in Wyoming
Crusoe finished pre-development and permitting for the site, which uses a hybrid power plan with fuel cells, grid power, and gas turbines.
Read original https://crusoe.ai/resources/newsroom/crusoe-wyoming-data-center-campus-updatedatacentercloudacquisitiondatacentercloudacquisition
Nov 5, 20263
- DDN · Nov 5, 2026Google Cloud and DDN detailed Managed Lustre for cloud HPC
Salesforce claims using the DDN-backed service sped up model training by 1.5x and cut storage costs by 42%.
Read originalddn.com/blog/one-file-system-for-the-whole-ai-and-hpc-lifecycleDDN · Nov 5, 2026
Google Cloud and DDN detailed Managed Lustre for cloud HPC
Salesforce claims using the DDN-backed service sped up model training by 1.5x and cut storage costs by 42%.
Read original https://ddn.com/blog/one-file-system-for-the-whole-ai-and-hpc-lifecycleno tagsno tags - Supermicro · Nov 5, 2026Supermicro partnered with IBM and Nvidia on a shared KV cache storage system
They claim the shared setup delivers up to 56x faster time-to-first-token for 130,000-token prompts running on a 120-billion-parameter model.
Read originalsupermicro.com/en/pressreleases/supermicro-launches-kv-cache-solution-ibm-and-nvidiaSupermicro · Nov 5, 2026
Supermicro partnered with IBM and Nvidia on a shared KV cache storage system
They claim the shared setup delivers up to 56x faster time-to-first-token for 130,000-token prompts running on a 120-billion-parameter model.
Read original https://supermicro.com/en/pressreleases/supermicro-launches-kv-cache-solution-ibm-and-nvidiastorageinferenceserveraistorageinferenceserverai - Cerebras · Nov 5, 2026Cerebras pushed back on claims that its wafer-scale chips wear out in two years
They claim their wafers run at an average 65°C junction temperature and will last decades rather than wearing out in two years.
Read originalcerebras.ai/blog/what-it-took-to-build-wafer-scale-and-why-it-lastsCerebras · Nov 5, 2026
Cerebras pushed back on claims that its wafer-scale chips wear out in two years
They claim their wafers run at an average 65°C junction temperature and will last decades rather than wearing out in two years.
Read original https://cerebras.ai/blog/what-it-took-to-build-wafer-scale-and-why-it-lastsno tagsno tags
Nov 4, 20263
- Lambda · Nov 4, 2026Lambda added role-based access control to its GPU cloud
The GPU cloud provider added service identities, groups, and granular roles so automation and team members no longer require full admin API keys.
Read originallambda.ai/blog/lambda-cloud-iam-rbacLambda · Nov 4, 2026
Lambda added role-based access control to its GPU cloud
The GPU cloud provider added service identities, groups, and granular roles so automation and team members no longer require full admin API keys.
Read original https://lambda.ai/blog/lambda-cloud-iam-rbacgpucloudsecuritygpucloudsecurity - Supercomputing News · Nov 4, 2026ETH Zurich researchers mapped network bottlenecks in AI training
Tracing runs up to 4,096 GPUs, they found Llama 70B is bandwidth-bound while Grok-1 is compute-driven and sensitive to latency above 50 µs.
Read originalsupercomputing.news/ai/sc26-ai-network-provisioning-trace-study-breakpointsSupercomputing News · Nov 4, 2026
ETH Zurich researchers mapped network bottlenecks in AI training
Tracing runs up to 4,096 GPUs, they found Llama 70B is bandwidth-bound while Grok-1 is compute-driven and sensitive to latency above 50 µs.
Read original https://supercomputing.news/ai/sc26-ai-network-provisioning-trace-study-breakpointsinterconnectnetworkingbenchmarkai-traininginterconnectnetworkingbenchmarkai-training - Lambda · Nov 4, 2026Lambda plans a $33 billion data center campus in Alabama
Lambda announced plans to invest $33.1 billion in a new data center campus in Lowndes County, Alabama, with construction slated to start in early 2027.
Read originallambda.ai/blog/lambda-to-invest-33-billion-in-alabama-with-new-data-centerLambda · Nov 4, 2026
Lambda plans a $33 billion data center campus in Alabama
Lambda announced plans to invest $33.1 billion in a new data center campus in Lowndes County, Alabama, with construction slated to start in early 2027.
Read original https://lambda.ai/blog/lambda-to-invest-33-billion-in-alabama-with-new-data-centergpuclouddatacenterai-infrastructuregpuclouddatacenterai-infrastructure
Nov 3, 20264
- Solidigm · Nov 3, 2026Solidigm detailed its floating-gate NAND roadmap for enterprise SSDs
They say their floating-gate NAND development aims to increase storage density and lower costs for data center SSDs.
Read originalnews.solidigm.com/en-WWSolidigm · Nov 3, 2026
Solidigm detailed its floating-gate NAND roadmap for enterprise SSDs
They say their floating-gate NAND development aims to increase storage density and lower costs for data center SSDs.
Read original https://news.solidigm.com/en-WWno tagsno tags - DDN · Nov 3, 2026DDN and Cisco validated Infinia storage on UCS servers
DDN validated its Infinia software storage on Cisco UCS servers, claiming it ran 64 billion risk simulations in two minutes.
Read originalddn.com/blog/the-machine-that-feeds-the-machine-ddn-infinia-on-cisco-ucs-connected-by-nexus-oneDDN · Nov 3, 2026
DDN and Cisco validated Infinia storage on UCS servers
DDN validated its Infinia software storage on Cisco UCS servers, claiming it ran 64 billion risk simulations in two minutes.
Read original https://ddn.com/blog/the-machine-that-feeds-the-machine-ddn-infinia-on-cisco-ucs-connected-by-nexus-oneno tagsno tags - Lambda · Nov 3, 2026Lambda benchmarked NVIDIA Nemotron 3.5 Super VL on Blackwell GPUs
Lambda claims running the 120B model in NVFP4 on one Blackwell GPU reached 1,860 tokens per second, beating two GPUs running BF16.
Read originallambda.ai/inference-models/nvidia/nvidia-nemotron-3.5-super-vl-120b-a12bLambda · Nov 3, 2026
Lambda benchmarked NVIDIA Nemotron 3.5 Super VL on Blackwell GPUs
Lambda claims running the 120B model in NVFP4 on one Blackwell GPU reached 1,860 tokens per second, beating two GPUs running BF16.
Read original https://lambda.ai/inference-models/nvidia/nvidia-nemotron-3.5-super-vl-120b-a12bno tagsno tags - SK hynix · Nov 3, 2026SK hynix met with Japanese suppliers to align on AI memory manufacturing
SK hynix met with over 20 Japanese equipment and materials partners in Chiba to share production roadmaps for AI memory chips.
Read originalnews.skhynix.com/en/2026-partners-day-japanSK hynix · Nov 3, 2026
SK hynix met with Japanese suppliers to align on AI memory manufacturing
SK hynix met with over 20 Japanese equipment and materials partners in Chiba to share production roadmaps for AI memory chips.
Read original https://news.skhynix.com/en/2026-partners-day-japanno tagsno tags
Nov 2, 20266
- Arm · Nov 2, 2026Arm detailed its FP8 architecture and register design
Arm added E4M3 and E5M2 formats to Armv9 with a dedicated control register so software can switch operand types per tensor without pipeline stalls.
Read originaldeveloper.arm.com/community/arm-community-blogs/b/architectures-and-processors-blog/posts/part-1-introduction-to-arm-fp8-8-bit-floating-point-formatsArm · Nov 2, 2026
Arm detailed its FP8 architecture and register design
Arm added E4M3 and E5M2 formats to Armv9 with a dedicated control register so software can switch operand types per tensor without pipeline stalls.
Read original https://developer.arm.com/community/arm-community-blogs/b/architectures-and-processors-blog/posts/part-1-introduction-to-arm-fp8-8-bit-floating-point-formatsfp8architecturearmfloating-pointfp8architecturearmfloating-point - Solidigm · Nov 2, 2026Solidigm updated its SSD firmware support matrix
The table lists the latest generic firmware versions across Solidigm's data center and client solid-state drive lineups.
Read originalsolidigm.com/support-page/drivers-downloads/ka-00099.htmlSolidigm · Nov 2, 2026
Solidigm updated its SSD firmware support matrix
The table lists the latest generic firmware versions across Solidigm's data center and client solid-state drive lineups.
Read original https://solidigm.com/support-page/drivers-downloads/ka-00099.htmlno tagsno tags - Dell Technologies · Nov 2, 2026Dell and CrowdStrike added Falcon security to Nvidia AI systems
Dell integrated CrowdStrike's Falcon platform into its Nvidia AI Factory infrastructure to secure agent activity and compute workloads.
Read originaldell.com/en-us/dt/corporate/newsroom/announcements/detailpage.press-releases~usa~2026~10~crowdstrike-and-dell-technologies-announce-integrated-ai-security-for-the-dell-ai-factory-with-nvidia.htmDell Technologies · Nov 2, 2026
Dell and CrowdStrike added Falcon security to Nvidia AI systems
Dell integrated CrowdStrike's Falcon platform into its Nvidia AI Factory infrastructure to secure agent activity and compute workloads.
Read original https://dell.com/en-us/dt/corporate/newsroom/announcements/detailpage.press-releases~usa~2026~10~crowdstrike-and-dell-technologies-announce-integrated-ai-security-for-the-dell-ai-factory-with-nvidia.htmno tagsno tags - Arm · Nov 2, 2026Arm detailed FP8 instructions for SIMD, SVE, and SME
The ISA additions add FP8 conversion, exponent-based scaling, and widening matrix-accumulate operations across Arm's vector extensions.
Read originaldeveloper.arm.com/community/arm-community-blogs/b/architectures-and-processors-blog/posts/part-2-fp8-instructions-for-arm-advanced-simd-sve-and-smeArm · Nov 2, 2026
Arm detailed FP8 instructions for SIMD, SVE, and SME
The ISA additions add FP8 conversion, exponent-based scaling, and widening matrix-accumulate operations across Arm's vector extensions.
Read original https://developer.arm.com/community/arm-community-blogs/b/architectures-and-processors-blog/posts/part-2-fp8-instructions-for-arm-advanced-simd-sve-and-smesvesmeisafp8svesmeisafp8 - Qualcomm · Nov 2, 2026Qualcomm validated test silicon for near-memory AI compute
They say the test silicon proves their near-memory architecture works, and a datacenter chip with bonded DRAM is already in bring-up.
Read originalqualcomm.com/news/onq/2026/11/qualcomm-hbc-validated-silicon-ai-inferenceQualcomm · Nov 2, 2026
Qualcomm validated test silicon for near-memory AI compute
They say the test silicon proves their near-memory architecture works, and a datacenter chip with bonded DRAM is already in bring-up.
Read original https://qualcomm.com/news/onq/2026/11/qualcomm-hbc-validated-silicon-ai-inferenceno tagsno tags - Samsung Semiconductor · Nov 2, 2026Samsung seeks university proposals for AI system architecture
Samsung's AGI Computing Lab opened a call for academic research into system performance analysis and architecture for AI workloads.
Read originalsemiconductor.samsung.com/about-us/global-network/us-rnd-labs/memory-labs/university-collaborationSamsung Semiconductor · Nov 2, 2026
Samsung seeks university proposals for AI system architecture
Samsung's AGI Computing Lab opened a call for academic research into system performance analysis and architecture for AI workloads.
Read original https://semiconductor.samsung.com/about-us/global-network/us-rnd-labs/memory-labs/university-collaborationno tagsno tags
Nov 1, 20261
- Crusoe · Nov 1, 2026Crusoe unveiled liquid-cooled modular data centers for dense GPU clusters
They claim the liquid-cooled Spark modules offer 5x the compute density of air-cooled units and can house Nvidia Rubin NVL72 racks.
Read originalcrusoe.ai/resources/blog/crusoe-spark-liquid-managed-rl-mlflowCrusoe · Nov 1, 2026
Crusoe unveiled liquid-cooled modular data centers for dense GPU clusters
They claim the liquid-cooled Spark modules offer 5x the compute density of air-cooled units and can house Nvidia Rubin NVL72 racks.
Read original https://crusoe.ai/resources/blog/crusoe-spark-liquid-managed-rl-mlflowno tagsno tags
Oct 29, 20262
- CIQ · Oct 29, 2026CIQ launched Stratum Kubernetes Engine for cluster management
The tool packages core Kubernetes components into a single signed file for deploying and upgrading clusters on Enterprise Linux.
Read originalciq.com/blog/introducing-ciq-stratum-proCIQ · Oct 29, 2026
CIQ launched Stratum Kubernetes Engine for cluster management
The tool packages core Kubernetes components into a single signed file for deploying and upgrading clusters on Enterprise Linux.
Read original https://ciq.com/blog/introducing-ciq-stratum-prono tagsno tags - Kioxia · Oct 29, 2026Kioxia posts article collection on SSDs and data center storage for AI
The company published resources on storage for GPU workloads, including efforts aimed at reaching 100 million IOPS.
Read originalamericas.kioxia.com/en-us/insights.htmlKioxia · Oct 29, 2026
Kioxia posts article collection on SSDs and data center storage for AI
The company published resources on storage for GPU workloads, including efforts aimed at reaching 100 million IOPS.
Read original https://americas.kioxia.com/en-us/insights.htmlno tagsno tags
Oct 28, 20266
- hgpu.org · Oct 28, 2026Engineers ported million-token Qwen3 serving to AMD ROCm
Researchers cut per-rank transient memory from 91.6 GiB to 10.6 GiB to run million-token Qwen3 models on AMD ROCm with vLLM.
Read originalhgpu.org/?p=31446hgpu.org · Oct 28, 2026
Engineers ported million-token Qwen3 serving to AMD ROCm
Researchers cut per-rank transient memory from 91.6 GiB to 10.6 GiB to run million-token Qwen3 models on AMD ROCm with vLLM.
Read original https://hgpu.org/?p=31446gpurocmamdinferencegpurocmamdinference - Solidigm · Oct 28, 2026Solidigm published attestation manifests for D7-PS10xx SSDs
The files let users verify device authenticity on PS1010 and PS1030 data center SSDs using the SPDM protocol.
Read originalsolidigm.com/support-page/drivers-downloads/ka-01851.htmlSolidigm · Oct 28, 2026
Solidigm published attestation manifests for D7-PS10xx SSDs
The files let users verify device authenticity on PS1010 and PS1030 data center SSDs using the SPDM protocol.
Read original https://solidigm.com/support-page/drivers-downloads/ka-01851.htmlno tagsno tags - VAST Data · Oct 28, 2026CoreWeave outlines datacenter-scale architecture for GPU clusters
CoreWeave says rising GPU density is forcing operators to manage power, cooling, and telemetry across the whole datacenter rather than single racks.
Read originalvastdata.com/blog/coreweave-on-building-resilient-ai-infrastructure-at-massive-scaleVAST Data · Oct 28, 2026
CoreWeave outlines datacenter-scale architecture for GPU clusters
CoreWeave says rising GPU density is forcing operators to manage power, cooling, and telemetry across the whole datacenter rather than single racks.
Read original https://vastdata.com/blog/coreweave-on-building-resilient-ai-infrastructure-at-massive-scaleno tagsno tags - hgpu.org · Oct 28, 2026Meta optimized recommendation model training on AMD Instinct GPUs
Meta claims their co-design methods delivered up to 1.5x higher training throughput on AMD MI350X hardware compared to directly ported code.
Read originalhgpu.org/?p=31445hgpu.org · Oct 28, 2026
Meta optimized recommendation model training on AMD Instinct GPUs
Meta claims their co-design methods delivered up to 1.5x higher training throughput on AMD MI350X hardware compared to directly ported code.
Read original https://hgpu.org/?p=31445amdrocmgpuhpcamdrocmgpuhpc - Samsung Semiconductor · Oct 28, 2026Samsung details zHBM to stack memory directly on AI chips
Samsung says directly stacking memory onto processors with hybrid copper bonding cuts data transfer distance and lowers energy use.
Read originalsemiconductor.samsung.com/news-events/tech-blog/part-2-zhbm-next-generation-memory-for-the-data-movement-eraSamsung Semiconductor · Oct 28, 2026
Samsung details zHBM to stack memory directly on AI chips
Samsung says directly stacking memory onto processors with hybrid copper bonding cuts data transfer distance and lowers energy use.
Read original https://semiconductor.samsung.com/news-events/tech-blog/part-2-zhbm-next-generation-memory-for-the-data-movement-erano tagsno tags - hgpu.org · Oct 28, 2026Researcher created a clean-room CUDA toolchain for Nvidia GB10
They report running a 35B parameter model on a Grace Blackwell GPU with zero Nvidia userspace libraries mapped.
Read originalhgpu.org/?p=31447hgpu.org · Oct 28, 2026
Researcher created a clean-room CUDA toolchain for Nvidia GB10
They report running a 35B parameter model on a Grace Blackwell GPU with zero Nvidia userspace libraries mapped.
Read original https://hgpu.org/?p=31447cudanvidiacompilergpucudanvidiacompilergpu
Oct 27, 20266
- Cerebras · Oct 27, 2026Oracle Cloud Infrastructure plans to offer Cerebras AI inference
Oracle and Cerebras announced plans to make wafer-scale AI inference available on OCI in 2027.
Read originalcerebras.ai/press-release/cerebras-and-oracle-team-up-to-bring-cerebras-ai-inference-to-oracle-cloud-inf-3389429Cerebras · Oct 27, 2026
Oracle Cloud Infrastructure plans to offer Cerebras AI inference
Oracle and Cerebras announced plans to make wafer-scale AI inference available on OCI in 2027.
Read original https://cerebras.ai/press-release/cerebras-and-oracle-team-up-to-bring-cerebras-ai-inference-to-oracle-cloud-inf-3389429ai-inferencecloudwafer-scaleacceleratorai-inferencecloudwafer-scaleaccelerator - DDN · Oct 27, 2026UNSW Sydney picked DDN to upgrade its research data storage
DDN and Arcitecta will replace 11 legacy storage systems with EXAScaler and Mediaflux to streamline research computing data access.
Read originalddn.com/press-releases/ddn-selected-to-modernize-research-data-infrastructure-at-unsw-sydneyDDN · Oct 27, 2026
UNSW Sydney picked DDN to upgrade its research data storage
DDN and Arcitecta will replace 11 legacy storage systems with EXAScaler and Mediaflux to streamline research computing data access.
Read original https://ddn.com/press-releases/ddn-selected-to-modernize-research-data-infrastructure-at-unsw-sydneyno tagsno tags - NVIDIA · Oct 27, 2026Oracle launched Nvidia Vera Rubin NVL72 systems on OCI
Oracle made Nvidia's Vera Rubin NVL72 and Vera CPUs available on OCI, saying clusters will scale to multi-gigawatt capacity.
Read originalnvidianews.nvidia.com/news/nvidia-and-oracle-expand-nvidia-vera-rubin-to-multi-gigawatt-scale-and-bring-nvidia-vera-cpu-to-oracle-cloud-infrastructureNVIDIA · Oct 27, 2026
Oracle launched Nvidia Vera Rubin NVL72 systems on OCI
Oracle made Nvidia's Vera Rubin NVL72 and Vera CPUs available on OCI, saying clusters will scale to multi-gigawatt capacity.
Read original https://nvidianews.nvidia.com/news/nvidia-and-oracle-expand-nvidia-vera-rubin-to-multi-gigawatt-scale-and-bring-nvidia-vera-cpu-to-oracle-cloud-infrastructuregpucpusupercomputercloudgpucpusupercomputercloud - AMD · Oct 27, 2026AMD and Meta expand AI infrastructure partnership with custom GPUs and Helios racks
Meta is testing AMD's Helios rackscale systems and plans to deploy custom Instinct GPUs alongside EPYC CPUs for its AI infrastructure.
Read originalir.amd.com/news-events/press-releases/detail/1302/amd-and-meta-advance-strategic-partnership-to-power-metas-superintelligence-infrastructureAMD · Oct 27, 2026
AMD and Meta expand AI infrastructure partnership with custom GPUs and Helios racks
Meta is testing AMD's Helios rackscale systems and plans to deploy custom Instinct GPUs alongside EPYC CPUs for its AI infrastructure.
Read original https://ir.amd.com/news-events/press-releases/detail/1302/amd-and-meta-advance-strategic-partnership-to-power-metas-superintelligence-infrastructuregpuepycinstinctdatacentergpuepycinstinctdatacenter - Dell Technologies · Oct 27, 2026Dell launched security and backup tools for its AI and storage systems
The updates add security hardening for Dell AI Factory infrastructure and integrate snapshot orchestration across PowerProtect and PowerScale.
Read originaldell.com/en-us/dt/corporate/newsroom/announcements/detailpage.press-releases~usa~2026~10~dell-technologies-strengthens-cybersecurity-and-resilience-across-ai-environments-identity-and-infrastructure.htmDell Technologies · Oct 27, 2026
Dell launched security and backup tools for its AI and storage systems
The updates add security hardening for Dell AI Factory infrastructure and integrate snapshot orchestration across PowerProtect and PowerScale.
Read original https://dell.com/en-us/dt/corporate/newsroom/announcements/detailpage.press-releases~usa~2026~10~dell-technologies-strengthens-cybersecurity-and-resilience-across-ai-environments-identity-and-infrastructure.htmno tagsno tags - CoreWeave · Oct 27, 2026CoreWeave launched Forge for AI model serving and fine-tuning
They say the tool connects model serving, observability, and fine-tuning into a single development environment to speed up updates.
Read originalcoreweave.com/blog/coreweave-forge-turn-ai-iteration-into-compounding-improvementCoreWeave · Oct 27, 2026
CoreWeave launched Forge for AI model serving and fine-tuning
They say the tool connects model serving, observability, and fine-tuning into a single development environment to speed up updates.
Read original https://coreweave.com/blog/coreweave-forge-turn-ai-iteration-into-compounding-improvementno tagsno tags
Oct 26, 20263
- Intel · Oct 26, 2026Intel agrees to acquire quantum software maker Classiq
Intel says the deal will combine its silicon spin-qubit hardware with Classiq's circuit synthesis software into a full quantum stack by 2027.
Read originalintel.com/content/www/us/en/newsroom/news/corporate/intel-classiq-accelerate-scalable-quantum-computing.htmlIntel · Oct 26, 2026
Intel agrees to acquire quantum software maker Classiq
Intel says the deal will combine its silicon spin-qubit hardware with Classiq's circuit synthesis software into a full quantum stack by 2027.
Read original https://intel.com/content/www/us/en/newsroom/news/corporate/intel-classiq-accelerate-scalable-quantum-computing.htmlno tagsno tags - The Register · Oct 26, 2026Cadence and Nvidia released specialized AI models for chip design
They claim the Nemotron-based Maestro models deliver up to 30 percent higher output quality and 5x better token efficiency for EDA tools.
Read originaltheregister.com/hpc/2026/10/26/cadence-nvidia-unveil-new-ai-models-to-put-eda-on-autopilot/5305228The Register · Oct 26, 2026
Cadence and Nvidia released specialized AI models for chip design
They claim the Nemotron-based Maestro models deliver up to 30 percent higher output quality and 5x better token efficiency for EDA tools.
Read original https://theregister.com/hpc/2026/10/26/cadence-nvidia-unveil-new-ai-models-to-put-eda-on-autopilot/5305228edaaihardware-designnvidiaedaaihardware-designnvidia - CIQ · Oct 26, 2026CIQ tests find disabling Gemma 4 reasoning cuts inference costs up to 8x
Turning off reasoning mode in llama.cpp raised Gemma 4 pass rates from around 52% to 100% on short tasks while cutting power and token costs up to 8x.
Read originalciq.com/blog/strix-halo-gemma-4-reasoning-offCIQ · Oct 26, 2026
CIQ tests find disabling Gemma 4 reasoning cuts inference costs up to 8x
Turning off reasoning mode in llama.cpp raised Gemma 4 pass rates from around 52% to 100% on short tasks while cutting power and token costs up to 8x.
Read original https://ciq.com/blog/strix-halo-gemma-4-reasoning-offno tagsno tags
