# NVLink

NVIDIA's high-bandwidth, low-latency GPU-to-GPU interconnect and scale-up networking fabric for multi-GPU systems.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Towards Scale-Up Networking: Challenges and Open Questions

DevFeed: [Towards Scale-Up Networking: Challenges and Open Questions](<https://devfeed.tech/articles/towards-scale-up-networking-challenges-and-open-questions-146131.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.26156>)

Author: Tingting Xu, Zixun Su, Xinyang Huang, Bowen Liu, Kun Tan, Dong Lin, Hao Wang, Yinben Xia, Weifeng Zhang, Binzhang Fu, Shuihai Hu, Kai Chen

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>), [arXiv: Networking](<https://devfeed.tech/sources/arxiv-networking.md>)

Topics: [NVLink](<https://devfeed.tech/topics/nvlink.md>), [Distributed Systems & Parallel Computing](<https://devfeed.tech/topics/distributed-systems-parallel-computing.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [bottleneck](<https://devfeed.tech/tags/bottleneck.md>), [challenges](<https://devfeed.tech/tags/challenges.md>), [collective](<https://devfeed.tech/tags/collective.md>), [communication](<https://devfeed.tech/tags/communication.md>), [compatibility](<https://devfeed.tech/tags/compatibility.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-ni](<https://devfeed.tech/tags/cs-ni.md>), [design](<https://devfeed.tech/tags/design.md>), [device](<https://devfeed.tech/tags/device.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [distributed-training](<https://devfeed.tech/tags/distributed-training.md>), [domain](<https://devfeed.tech/tags/domain.md>), [experts](<https://devfeed.tech/tags/experts.md>), [foundation](<https://devfeed.tech/tags/foundation.md>), [framework](<https://devfeed.tech/tags/framework.md>), [networking](<https://devfeed.tech/tags/networking.md>), [open](<https://devfeed.tech/tags/open.md>), [open-questions](<https://devfeed.tech/tags/open-questions.md>), [scale-up](<https://devfeed.tech/tags/scale-up.md>)

### AI overview

The paper derives scale-up network requirements from critical-path communication in distributed training and inference, focusing on domain size, per-accelerator bandwidth, and collective completion latency. It uses those requirements to compare representative proprietary and open fabrics, describing trade-offs in performance, scalability, compatibility, and implementation cost across their interconnects, topologies, protocols, and communication semantics. It also identifies open challenges in switching and physical interconnects, efficient communication, semantics, resource sharing, and connecting heterogeneous devices. This is a requirement-driven analysis; the supplied excerpt does not provide specific fabric-by-fabric results.

### Source excerpt

arXiv:2611.26156v1 Announce Type: cross Abstract: The rapid growth of AI models and the widespread use of distributed training and inference have made inter-accelerator communication a critical system bottleneck. Scale-up fabrics address this bottleneck by combining high per-accelerator bandwidth with low and predictable latency over tightly coupled accelerator domains. However, the appropriate topology, protocol mechanisms, communication semantics, and system integration remain actively debated. We analyze the critical-path communication generated by parallelism strategies and derive the resulting requirements for domain scale, per-accelerator bandwidth, and collective completion latency. We then use these requirements as a framework to examine representative proprietary and open scale-up fabrics, revealing how their physical interconnects, topologies, protocols, and semantics make different trade-offs among performance, scalability, compatibility, and implementation cost. Moreover, based on this analysis and discussions with industry experts, we identify challenges and open problems in physical interconnects and switching, efficient communication, semantics, resource sharing and heterogeneous device interconnection. This paper provides a requirement-driven foundation for analyzing current scale-up fabrics and guiding the design of future systems.

## NVIDIA Vera Rubin NVL72 on CoreWeave: 10x More Tokens Per Megawatt Than Blackwell

DevFeed: [NVIDIA Vera Rubin NVL72 on CoreWeave: 10x More Tokens Per Megawatt Than Blackwell](<https://devfeed.tech/articles/nvidia-vera-rubin-nvl72-on-coreweave-10x-more-tokens-per-megawatt-than-blackwell-144651.md>)

Original publisher: [Read original article](<https://wf.coreweave.com/blog/nvidia-vera-rubin-nvl72-on-coreweave-10x-more-tokens-per-megawatt-than-blackwell>)

Author: CoreWeave

Published: 2026-11-19T22:40:29Z

Content type: news

Language: en

Sources: [Coreweave Blog Posts](<https://devfeed.tech/sources/coreweave-blog-posts.md>)

Topics: [Vera Rubin NVL72](<https://devfeed.tech/topics/vera-rubin-nvl72.md>), [NVIDIA Vera Rubin](<https://devfeed.tech/topics/nvidia-vera-rubin.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [TensorRT-LLM](<https://devfeed.tech/topics/tensorrt-llm.md>), [NVFP4](<https://devfeed.tech/topics/nvfp4.md>), [NVIDIA Blackwell](<https://devfeed.tech/topics/nvidia-blackwell.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [GB200](<https://devfeed.tech/topics/gb200.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [agentic-security](<https://devfeed.tech/tags/agentic-security.md>), [agentic-workloads](<https://devfeed.tech/tags/agentic-workloads.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [blackwell](<https://devfeed.tech/tags/blackwell.md>), [blog](<https://devfeed.tech/tags/blog.md>), [coreweave](<https://devfeed.tech/tags/coreweave.md>), [first](<https://devfeed.tech/tags/first.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [megawatt](<https://devfeed.tech/tags/megawatt.md>), [milestone](<https://devfeed.tech/tags/milestone.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-vera-rubin](<https://devfeed.tech/tags/nvidia-vera-rubin.md>), [nvl72](<https://devfeed.tech/tags/nvl72.md>), [performance](<https://devfeed.tech/tags/performance.md>), [silicon](<https://devfeed.tech/tags/silicon.md>), [tokens](<https://devfeed.tech/tags/tokens.md>), [vera-rubin-nvl72](<https://devfeed.tech/tags/vera-rubin-nvl72.md>)

### AI overview

In CoreWeave's DeepSeek R1 evaluation, Vera Rubin NVL72 delivered 10x the tokens per second per megawatt of NVIDIA GB200 NVL72 at a matched interactivity target (TPS/user). Both systems used inference optimizations including expert parallelism, NVFP4, multi-token prediction, and disaggregated prefill and decode with TensorRT-LLM and Dynamo. CoreWeave attributes the result's relevance to reasoning workloads' high token demand and describes Vera Rubin's 72 Rubin GPUs, 36 Vera CPUs, 260 TB/s NVLink 6 fabric, and native NVFP4 support. The measurement is an initial result; CoreWeave expects further gains from tuning and says it will publish updated numbers. Customer benefits in the excerpt are expectations, not reported measured outcomes.

### Source excerpt

CoreWeave publishes first-ever measured silicon performance of NVIDIA Vera Rubin NVL72, marking a new milestone in AI infrastructure.

## D-Matrix bets stacked DRAM can solve inference's memory bandwidth problem

DevFeed: [D-Matrix bets stacked DRAM can solve inference's memory bandwidth problem](<https://devfeed.tech/articles/d-matrix-bets-stacked-dram-can-solve-inference-s-memory-bandwidth-problem-144257.md>)

Original publisher: [Read original article](<https://siliconangle.com/2026/11/19/raptor-chip-stacks-3d-dram-tackle-inference-memory-limits-sc26/>)

Author: Kelly Knight

Published: 2026-11-19T19:21:48Z

Content type: news

Language: en

Sources: [SiliconANGLE](<https://devfeed.tech/sources/siliconangle.md>)

Topics: [d-matrix](<https://devfeed.tech/topics/d-matrix.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [AI Factory](<https://devfeed.tech/topics/ai-factory.md>)

Tags: [3d](<https://devfeed.tech/tags/3d.md>), [3d-dram](<https://devfeed.tech/tags/3d-dram.md>), [accelerated-computing](<https://devfeed.tech/tags/accelerated-computing.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-data-centers](<https://devfeed.tech/tags/ai-data-centers.md>), [ai-inference](<https://devfeed.tech/tags/ai-inference.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [can](<https://devfeed.tech/tags/can.md>), [chip](<https://devfeed.tech/tags/chip.md>), [compute](<https://devfeed.tech/tags/compute.md>), [corsair](<https://devfeed.tech/tags/corsair.md>), [cube-event-coverage](<https://devfeed.tech/tags/cube-event-coverage.md>), [d-matrix](<https://devfeed.tech/tags/d-matrix.md>), [designs](<https://devfeed.tech/tags/designs.md>), [dram](<https://devfeed.tech/tags/dram.md>), [emerging-tech](<https://devfeed.tech/tags/emerging-tech.md>), [first](<https://devfeed.tech/tags/first.md>), [game](<https://devfeed.tech/tags/game.md>), [generation](<https://devfeed.tech/tags/generation.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [high-bandwidth-memory](<https://devfeed.tech/tags/high-bandwidth-memory.md>), [high-performance-computing](<https://devfeed.tech/tags/high-performance-computing.md>), [high-speed-networking](<https://devfeed.tech/tags/high-speed-networking.md>), [hpc](<https://devfeed.tech/tags/hpc.md>), [hpc-infrastructure](<https://devfeed.tech/tags/hpc-infrastructure.md>), [inference](<https://devfeed.tech/tags/inference.md>), [john-furrier](<https://devfeed.tech/tags/john-furrier.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-bandwidth](<https://devfeed.tech/tags/memory-bandwidth.md>), [nasdaq-nvda](<https://devfeed.tech/tags/nasdaq-nvda.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvlink-fusion](<https://devfeed.tech/tags/nvlink-fusion.md>), [raptor](<https://devfeed.tech/tags/raptor.md>), [satyam-srivastava](<https://devfeed.tech/tags/satyam-srivastava.md>), [sc26](<https://devfeed.tech/tags/sc26.md>), [sc26eventpage](<https://devfeed.tech/tags/sc26eventpage.md>), [scientific-simulation](<https://devfeed.tech/tags/scientific-simulation.md>), [solve](<https://devfeed.tech/tags/solve.md>), [supercomputing](<https://devfeed.tech/tags/supercomputing.md>), [sustainable-computing](<https://devfeed.tech/tags/sustainable-computing.md>), [thecube](<https://devfeed.tech/tags/thecube.md>)

### AI overview

D-Matrix says Raptor, its planned second-generation chip, will stack 3D DRAM beneath compute logic to address inference's memory-bandwidth demands. The Pavehawk test chip is validating the approach: memory sits on a separate layer, close to logic, without crossing an interposer. The company places this memory tier between SRAM and HBM in capacity and claims 10 times HBM's bandwidth and energy efficiency. Srivastava also says pairing Corsair-based servers with Hopper GPUs can raise utilization from about 20% to as much as 80%. D-Matrix's adoption of Nvidia NVLink Fusion is intended to connect its accelerators to Nvidia AI factories, supporting systems that coordinate GPUs, CPUs and accelerators according to workload. These performance figures and utilization gains are company-reported claims.

### Source excerpt

As artificial intelligence workloads shift to inference, memory bandwidth decides whether hardware generates tokens quickly or sits idle. D-Matrix Corp. is answering with Raptor, a second-generation chip stacking 3D DRAM beneath its compute logic. AI infrastructure has become a systems game, and winners combine memory, interconnect and software into cohesive designs, according to Satyam Srivastava [...] The post D-Matrix bets stacked DRAM can solve inference's memory bandwidth problem appeared first on SiliconANGLE.

## Nvidia expands AI infrastructure push into high-performance computing

DevFeed: [Nvidia expands AI infrastructure push into high-performance computing](<https://devfeed.tech/articles/nvidia-expands-ai-infrastructure-push-into-high-performance-computing-141282.md>)

Original publisher: [Read original article](<https://www.networkworld.com/article/4250303/nvidia-expands-ai-infrastructure-push-into-high-performance-computing.html>)

Author: Andy Patrizio

Published: 2026-11-18T17:10:21Z

Content type: news

Language: en

Sources: [Network World](<https://devfeed.tech/sources/network-world.md>)

Topics: [High-Performance Computing](<https://devfeed.tech/topics/high-performance-computing.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>), [Supercomputing](<https://devfeed.tech/topics/supercomputing.md>), [CUDA-X](<https://devfeed.tech/topics/cuda-x.md>), [systems](<https://devfeed.tech/topics/systems.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [Nemotron](<https://devfeed.tech/topics/nemotron.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [AI Networking](<https://devfeed.tech/topics/ai-networking.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-models](<https://devfeed.tech/tags/ai-models.md>), [ai-workloads](<https://devfeed.tech/tags/ai-workloads.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [artificial-intelligence-cpus-and-processors-industry-markets-networking-technology-industry](<https://devfeed.tech/tags/artificial-intelligence-cpus-and-processors-industry-markets-networking-technology-industry.md>), [computing](<https://devfeed.tech/tags/computing.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cpus](<https://devfeed.tech/tags/cpus.md>), [cpus-and-processors](<https://devfeed.tech/tags/cpus-and-processors.md>), [cuda](<https://devfeed.tech/tags/cuda.md>), [data](<https://devfeed.tech/tags/data.md>), [data-processing](<https://devfeed.tech/tags/data-processing.md>), [edition-us](<https://devfeed.tech/tags/edition-us.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [high-performance](<https://devfeed.tech/tags/high-performance.md>), [high-performance-computing](<https://devfeed.tech/tags/high-performance-computing.md>), [industry](<https://devfeed.tech/tags/industry.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [language-en](<https://devfeed.tech/tags/language-en.md>), [markets](<https://devfeed.tech/tags/markets.md>), [networking](<https://devfeed.tech/tags/networking.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [performance](<https://devfeed.tech/tags/performance.md>), [storytype-news](<https://devfeed.tech/tags/storytype-news.md>), [technology-industry](<https://devfeed.tech/tags/technology-industry.md>)

### AI overview

Nvidia's SC26 announcements extend its push from GPU sales toward full HPC and AI infrastructure. Rubin HPC, expected in 2028, targets scientific simulations, engineering, and data processing with more than 50,000 CUDA cores, seventh-generation NVLink, and sparse-data optimizations. A rack-scale Vera CPU system combines 192 CPUs per rack with Quantum-X800 InfiniBand and SHARP, using NVL72 chassis, power, and cooling components. Nemotron for Science offers open models, datasets, evaluation tools, and post-training recipes for scientific AI agents that can interpret research, generate code, and use computational tools. CUDA-X updates include claimed speedups ranging from 3x to 900x for specific scientific tasks, with comparisons varying by library and baseline. Lambda is deploying Nvidia Photonics co-packaged optics switches to connect GB300 NVL72 racks containing over 10,000 GPUs; Nvidia presents lower power use, reliability, and scalability as intended benefits.

### Source excerpt

Perhaps anxious to talk about technology instead of its CEO drama, Nvidia unveiled a raft of HPC-related news at the Supercomputing '26 show taking place in Chicago. Nvidia's announcements at SC26 continue the company's efforts to expand beyond selling GPUs into supplying complete AI and high-performance computing infrastructure. Rubin HPC: A new GPU for scientific computing The news covers next-generation GPU and CPU platforms, networking and interconnect technology, accelerated scientific computing, and systems designed for large-scale AI workloads. Here is a breakdown of the major developments. Nvidia announced Rubin HPC, a new GPU within its Vera Rubin platform designed specifically for high-performance computing applications. Unlike GPUs primarily optimized for AI training and inference, Rubin HPC is designed to boost scientific simulations, engineering applications, and large-scale data processing. Rubin HPC has more than 50,000 CUDA cores on a single GPU. The core count for regular Roman GPUs has not been disclosed. It includes seventh-generation NVLink for accelerated GPU-to-GPU communication, sparse-data acceleration, and optimizations for sparse matrix operations, stencil computations, and sorting and reduction operations. Rubin HPC is expected to ship in 2028. Vera CPU platform expands into HPC Nvidia also announced a rack-scale version of its Vera CPU platform specifically designed for HPC. When it first introduced its own CPU, Nvidia positioned it as an alternative to traditional server processors from Intel and AMD. Now Nvidia is positioning its CPU for a specific market. The new configuration combines 192 Vera CPUs per rack plus Quantum-X800 InfiniBand networking, and SHARP in-network computing technology. It uses existing NVL72 rack components, including chassis, power distribution, and cooling infrastructure. Nemotron for Science: AI agents for scientific research Nemotron is a family of AI models developed by Nvidia, designed to power AI applicatio

## Nvidia reveals Rubin HPC, a GPU with 50,000+ CUDA cores for scientific simulation

DevFeed: [Nvidia reveals Rubin HPC, a GPU with 50,000+ CUDA cores for scientific simulation](<https://devfeed.tech/articles/nvidia-reveals-rubin-hpc-a-gpu-with-50-000-cuda-cores-for-scientific-simulation-137868.md>)

Original publisher: [Read original article](<https://www.techspot.com/news/114583-nvidia-reveals-rubin-hpc-gpu-50000-cuda-cores.html>)

Author: Skye Jacobs

Published: 2026-11-17T15:46:00Z

Content type: news

Language: en

Sources: [TechSpot](<https://devfeed.tech/sources/techspot.md>)

Topics: [High-Performance Computing](<https://devfeed.tech/topics/high-performance-computing.md>), [CUDA-X](<https://devfeed.tech/topics/cuda-x.md>), [NVIDIA Vera Rubin](<https://devfeed.tech/topics/nvidia-vera-rubin.md>), [CUDA](<https://devfeed.tech/topics/cuda.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>)

Tags: [accelerated-computing](<https://devfeed.tech/tags/accelerated-computing.md>), [acceleration](<https://devfeed.tech/tags/acceleration.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-workflows](<https://devfeed.tech/tags/agentic-workflows.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [algorithms](<https://devfeed.tech/tags/algorithms.md>), [amd](<https://devfeed.tech/tags/amd.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [analytics](<https://devfeed.tech/tags/analytics.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [cuda](<https://devfeed.tech/tags/cuda.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [hpc](<https://devfeed.tech/tags/hpc.md>), [multi-core](<https://devfeed.tech/tags/multi-core.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [rubin](<https://devfeed.tech/tags/rubin.md>), [rubin-gpu](<https://devfeed.tech/tags/rubin-gpu.md>), [rubin-hpc](<https://devfeed.tech/tags/rubin-hpc.md>), [science](<https://devfeed.tech/tags/science.md>), [simulation](<https://devfeed.tech/tags/simulation.md>)

### AI overview

Nvidia says its Rubin HPC GPU, announced for the Vera Rubin platform and expected in 2028, will have more than 50,000 CUDA cores and target scientific and engineering simulations used to test ideas generated by AI agents. The company says its "balanced design" scales data movement with compute, using sparse data acceleration for recurring routines such as sparse matrix operations, stencil computations, sorting and reductions. GPUs will connect through seventh-generation NVLink, and agents can use CUDA-X libraries with tuned algorithms. Key HPC specifications remain undisclosed: memory capacity and bandwidth, double-precision capability, interconnect bandwidth, and whether the chip uses multiple dies. Nvidia has not confirmed its physical layout.

### Source excerpt

The core count offers the first clue to what Nvidia is building. The current Rubin GPU has 224 streaming multiprocessors and 28,672 CUDA cores, so getting past 50,000 would take at least 74% more cores. Read Entire Article

## NVIDIA Announces Rubin HPC: Science-Focused Server GPU to Arrive in 2028

DevFeed: [NVIDIA Announces Rubin HPC: Science-Focused Server GPU to Arrive in 2028](<https://devfeed.tech/articles/nvidia-announces-rubin-hpc-science-focused-server-gpu-to-arrive-in-2028-140184.md>)

Original publisher: [Read original article](<https://www.servethehome.com/nvidia-announces-rubin-hpc-science-focused-server-gpu-to-arrive-in-2028/>)

Author: Ryan Smith

Published: 2026-11-17T15:45:19Z

Content type: news

Language: en

Sources: [ServeTheHome](<https://devfeed.tech/sources/servethehome.md>)

Topics: [High-Performance Computing](<https://devfeed.tech/topics/high-performance-computing.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [CUDA](<https://devfeed.tech/topics/cuda.md>), [Tensor Cores](<https://devfeed.tech/topics/tensor-cores.md>)

Tags: [accelerators](<https://devfeed.tech/tags/accelerators.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [hpc](<https://devfeed.tech/tags/hpc.md>), [news](<https://devfeed.tech/tags/news.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [rubin](<https://devfeed.tech/tags/rubin.md>), [rubin-hpc](<https://devfeed.tech/tags/rubin-hpc.md>), [sc26](<https://devfeed.tech/tags/sc26.md>), [science](<https://devfeed.tech/tags/science.md>), [server](<https://devfeed.tech/tags/server.md>)

### AI overview

NVIDIA says its Rubin HPC GPU, aimed at science and engineering workloads, is planned for release in 2028. The announced chip will have more than 50,000 CUDA cores--about twice as many as the current Rubin server GPU--reflecting an emphasis on vector processing that many traditional HPC workloads need. NVIDIA also cites a balanced approach to compute and data movement, NVLink 7, and "sparse data acceleration," but has not disclosed performance figures, memory details, or how the sparse-data feature works. Support across precisions from NVFP4 to FP64 is promised; actual FP64 performance will depend on the architecture, including how much hardware is allocated to FP64 and tensor operations. The article presents the announcement as NVIDIA's return to dedicated HPC after Blackwell, with AMD's MI430X expected in 2027 and Intel preparing a PCIe GPU D-Series.

### Source excerpt

Marking their return to the HPC market, at SC26 NVIDIA has announced the Rubin HPC chip. The science and engineering-focused GPU will have over 50,000 CUDA cores and will be released in 2028 The post NVIDIA Announces Rubin HPC: Science-Focused Server GPU to Arrive in 2028 appeared first on ServeTheHome.

## Ian Buck Kicks Off NVIDIA's Week at SC26 as Science Enters the Age of Agents

DevFeed: [Ian Buck Kicks Off NVIDIA's Week at SC26 as Science Enters the Age of Agents](<https://devfeed.tech/articles/ian-buck-kicks-off-nvidia-s-week-at-sc26-as-science-enters-the-age-of-agents-138205.md>)

Original publisher: [Read original article](<https://blogs.nvidia.com/blog/ian-buck-sc26/>)

Author: Brian Caulfield

Published: 2026-11-17T14:00:10Z

Content type: news

Language: en

Sources: [NVIDIA](<https://devfeed.tech/sources/nvidia-blog.md>)

Topics: [Nemotron](<https://devfeed.tech/topics/nemotron.md>), [High-Performance Computing](<https://devfeed.tech/topics/high-performance-computing.md>), [Local AI](<https://devfeed.tech/topics/local-ai.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>), [DGX Spark](<https://devfeed.tech/topics/dgx-spark.md>), [NVIDIA Vera](<https://devfeed.tech/topics/nvidia-vera.md>), [CUDA](<https://devfeed.tech/topics/cuda.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [code](<https://devfeed.tech/tags/code.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [events](<https://devfeed.tech/tags/events.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [high-performance](<https://devfeed.tech/tags/high-performance.md>), [hpc](<https://devfeed.tech/tags/hpc.md>), [ian-buck](<https://devfeed.tech/tags/ian-buck.md>), [nemotron](<https://devfeed.tech/tags/nemotron.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [rubin](<https://devfeed.tech/tags/rubin.md>), [sc26](<https://devfeed.tech/tags/sc26.md>), [science](<https://devfeed.tech/tags/science.md>), [supercomputing](<https://devfeed.tech/tags/supercomputing.md>)

### AI overview

NVIDIA's SC26 announcements combine locally runnable science agents with new HPC hardware and quantum-computing work. Nemotron for Science pairs the Nemotron 3.5 Super VL open model with an open chemistry dataset and post-training recipes; researchers can run it on DGX Spark, a workstation or their own supercomputer. Buck cited an Argonne agent that planned and tested hypotheses, selected simulations and identified two binding sites, plus examples for GROMACS repair, telescope data plotting and GPU porting. The Nsight AI Porting Agent reportedly sped up a weather-model routine by more than 5x. NVIDIA also announced Rubin HPC, supporting FP64 to FP4 and expected in 2028, and Vera CPUs selected for a German climate supercomputer expected in early 2028. For quantum research, NVIDIA, IBM and Fermilab plan a co-design platform; Buck cited a CUDA-Q decoding algorithm that reduced a research computation from 12 weeks to 45 minutes. These are reported announcements and results from the excerpt.

### Source excerpt

NVIDIA's Ian Buck kicked off the company's week at SC26 in Chicago on Monday with new tools for AI agents, software that can plan experiments, run code and check results on researchers' own systems. In NVIDIA's special address, Buck announced Nemotron for Science, the Rubin HPC GPU and an expanded Vera CPU platform for high-performance [...]

## CoreWeave Becomes the First Cloud Provider to Deliver Multi-Rack NVIDIA Vera Rubin Results in MLPerf® Training v6.1

DevFeed: [CoreWeave Becomes the First Cloud Provider to Deliver Multi-Rack NVIDIA Vera Rubin Results in MLPerf® Training v6.1](<https://devfeed.tech/articles/coreweave-becomes-the-first-cloud-provider-to-deliver-multi-rack-nvidia-vera-rubin-results-in-mlperf-training-v6-1-136727.md>)

Original publisher: [Read original article](<https://wf.coreweave.com/blog/coreweave-becomes-the-first-cloud-provider-to-deliver-multi-rack-nvidia-vera-rubin-results-in-mlperf-r-training-v6-1>)

Author: CoreWeave

Published: 2026-11-16T19:52:27Z

Content type: news

Language: en

Sources: [Coreweave Blog Posts](<https://devfeed.tech/sources/coreweave-blog-posts.md>)

Topics: [NVIDIA Vera Rubin](<https://devfeed.tech/topics/nvidia-vera-rubin.md>), [Vera Rubin NVL72](<https://devfeed.tech/topics/vera-rubin-nvl72.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [distributed-training](<https://devfeed.tech/topics/distributed-training.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>), [NVIDIA Spectrum-X Ethernet](<https://devfeed.tech/topics/nvidia-spectrum-x-ethernet.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [GB200](<https://devfeed.tech/topics/gb200.md>), [NVFP4](<https://devfeed.tech/topics/nvfp4.md>), [TensorRT-LLM](<https://devfeed.tech/topics/tensorrt-llm.md>)

Tags: [blog](<https://devfeed.tech/tags/blog.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cloud-provider](<https://devfeed.tech/tags/cloud-provider.md>), [coreweave](<https://devfeed.tech/tags/coreweave.md>), [delivered](<https://devfeed.tech/tags/delivered.md>), [first](<https://devfeed.tech/tags/first.md>), [full](<https://devfeed.tech/tags/full.md>), [full-stack](<https://devfeed.tech/tags/full-stack.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [leading](<https://devfeed.tech/tags/leading.md>), [mlperf](<https://devfeed.tech/tags/mlperf.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-vera](<https://devfeed.tech/tags/nvidia-vera.md>), [nvidia-vera-rubin](<https://devfeed.tech/tags/nvidia-vera-rubin.md>), [rack](<https://devfeed.tech/tags/rack.md>), [results](<https://devfeed.tech/tags/results.md>), [rubin](<https://devfeed.tech/tags/rubin.md>), [training](<https://devfeed.tech/tags/training.md>), [vera](<https://devfeed.tech/tags/vera.md>), [vera-rubin](<https://devfeed.tech/tags/vera-rubin.md>)

### AI overview

In MLPerf Training v6.1, CoreWeave reports training DeepSeek-V3 671B to target quality in 17.65 minutes on 256 Vera Rubin GPUs across four racks, versus 48.40 minutes on the same number of GB200 GPUs (2.7x faster, per CoreWeave's calculation). It also reports the round's fastest 72-GPU Llama 3.1 8B submission: 7.16 minutes on one Vera Rubin rack, compared with 13.70 minutes on GB200. The article attributes multi-rack performance to an integrated stack: Spectrum-X RoCE networking and ConnectX-9 SuperNICs, topology-aware placement and orchestration, optimized software, and fleet monitoring and recovery. On GB200, it reports Qwen3.5 397B reinforcement learning in 22.78 minutes across four racks, split evenly between training and generation using NeMo and TensorRT-LLM. These are CoreWeave's descriptions of published results; its speedup and fastest-submission claims are its own analyses and were not verified by MLCommons.

### Source excerpt

See how CoreWeave's full-stack optimizations delivered leading NVIDIA Vera Rubin performance in MLPerf® Training v6.1, from a single 72-GPU rack to multi-rack scale at 256 GPUs

## NVIDIA Announces Rubin HPC -- A New GPU to Accelerate Agents for Science, Engineering and Analytics

DevFeed: [NVIDIA Announces Rubin HPC -- A New GPU to Accelerate Agents for Science, Engineering and Analytics](<https://devfeed.tech/articles/nvidia-announces-rubin-hpc-a-new-gpu-to-accelerate-agents-for-science-engineering-and-analytics-136625.md>)

Original publisher: [Read original article](<https://blogs.nvidia.com/blog/rubin-hpc/>)

Author: Dan Ernst

Published: 2026-11-16T16:00:01Z

Content type: release

Language: en

Sources: [NVIDIA](<https://devfeed.tech/sources/nvidia-blog.md>)

Topics: [GPU](<https://devfeed.tech/topics/gpu.md>), [High-Performance Computing](<https://devfeed.tech/topics/high-performance-computing.md>), [CUDA-X](<https://devfeed.tech/topics/cuda-x.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [AI for science](<https://devfeed.tech/topics/ai-for-science.md>), [NVIDIA Vera Rubin](<https://devfeed.tech/topics/nvidia-vera-rubin.md>), [Simulation and Design](<https://devfeed.tech/topics/simulation-and-design.md>)

Tags: [accelerate](<https://devfeed.tech/tags/accelerate.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [analytics](<https://devfeed.tech/tags/analytics.md>), [cuda-x](<https://devfeed.tech/tags/cuda-x.md>), [data](<https://devfeed.tech/tags/data.md>), [data-processing](<https://devfeed.tech/tags/data-processing.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [hpc](<https://devfeed.tech/tags/hpc.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-rubin](<https://devfeed.tech/tags/nvidia-rubin.md>), [nvidia-vera-rubin](<https://devfeed.tech/tags/nvidia-vera-rubin.md>), [nvlink](<https://devfeed.tech/tags/nvlink.md>), [rubin](<https://devfeed.tech/tags/rubin.md>), [rubin-hpc](<https://devfeed.tech/tags/rubin-hpc.md>), [sc26](<https://devfeed.tech/tags/sc26.md>), [science](<https://devfeed.tech/tags/science.md>), [simulation](<https://devfeed.tech/tags/simulation.md>), [simulations](<https://devfeed.tech/tags/simulations.md>), [supercomputing](<https://devfeed.tech/tags/supercomputing.md>), [test](<https://devfeed.tech/tags/test.md>), [vera-rubin](<https://devfeed.tech/tags/vera-rubin.md>), [workflows](<https://devfeed.tech/tags/workflows.md>)

### AI overview

NVIDIA says Rubin HPC, a GPU for the Vera Rubin platform announced at SC26, is designed to speed simulations and data processing used in science, engineering and analytics. It combines more than 50,000 CUDA cores with sparse data acceleration to keep cores supplied with data, plus seventh-generation NVLink for GPU-to-GPU connectivity. The design targets recurring workloads such as sparse matrix operations, stencil computations, sorting and reductions, balancing data movement with compute. NVIDIA CUDA-X libraries provide optimized algorithms intended to work with the GPU. Faster simulations could give agents and researchers quicker feedback as they test and refine candidate ideas. Rubin HPC is expected to be available in 2028.

### Source excerpt

Announced today at SC26, NVIDIA Rubin HPC is a new GPU in the NVIDIA Vera Rubin platform designed to accelerate simulation and data processing for science, engineering and analytics workflows. As agents help scientists and engineers explore more candidate materials, molecules and designs, they rely on simulations to test those ideas and guide their next [...]

## NVIDIA Vera Rubin NVL72 Leads Agentic Inference Performance on MLPerf Endpoints v1.0

DevFeed: [NVIDIA Vera Rubin NVL72 Leads Agentic Inference Performance on MLPerf Endpoints v1.0](<https://devfeed.tech/articles/nvidia-vera-rubin-nvl72-leads-agentic-inference-performance-on-mlperf-endpoints-v1-0-136678.md>)

Original publisher: [Read original article](<https://developer.nvidia.com/blog/nvidia-vera-rubin-nvl72-leads-agentic-inference-performance-on-mlperf-endpoints-v1-0/>)

Author: Zhihan Jiang

Published: 2026-11-16T16:00:00Z

Content type: article

Language: en

Sources: [NVIDIA Developer](<https://devfeed.tech/sources/nvidia-developer.md>), [NVIDIA Technical Blog](<https://devfeed.tech/sources/nvidia-technical-blog.md>)

Topics: [Vera Rubin NVL72](<https://devfeed.tech/topics/vera-rubin-nvl72.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [Agentic AI](<https://devfeed.tech/topics/what-is-agentic-ai.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-ai-generative-ai](<https://devfeed.tech/tags/agentic-ai-generative-ai.md>), [agentic-tasks](<https://devfeed.tech/tags/agentic-tasks.md>), [agentic-workloads](<https://devfeed.tech/tags/agentic-workloads.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [api](<https://devfeed.tech/tags/api.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [blackwell-ultra](<https://devfeed.tech/tags/blackwell-ultra.md>), [co-design](<https://devfeed.tech/tags/co-design.md>), [coding](<https://devfeed.tech/tags/coding.md>), [context](<https://devfeed.tech/tags/context.md>), [data-center-cloud](<https://devfeed.tech/tags/data-center-cloud.md>), [dynamo](<https://devfeed.tech/tags/dynamo.md>), [gb300](<https://devfeed.tech/tags/gb300.md>), [groq-3-lpx](<https://devfeed.tech/tags/groq-3-lpx.md>), [inference](<https://devfeed.tech/tags/inference.md>), [inference-performance](<https://devfeed.tech/tags/inference-performance.md>), [infiniband](<https://devfeed.tech/tags/infiniband.md>), [mlperf](<https://devfeed.tech/tags/mlperf.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-vera-rubin](<https://devfeed.tech/tags/nvidia-vera-rubin.md>), [nvl72](<https://devfeed.tech/tags/nvl72.md>), [tensorrt-llm](<https://devfeed.tech/tags/tensorrt-llm.md>), [tool](<https://devfeed.tech/tags/tool.md>), [top-stories](<https://devfeed.tech/tags/top-stories.md>), [turns](<https://devfeed.tech/tags/turns.md>), [vera-rubin-nvl72](<https://devfeed.tech/tags/vera-rubin-nvl72.md>)

### AI overview

Provisional MLPerf Endpoints v1.0 results report that NVIDIA Vera Rubin NVL72 delivered over 10x the throughput per kilowatt of GB300 NVL72 at a shared minimum average interactivity of 50 tokens per second per user. The comparison uses different configurations and concurrency: Vera Rubin's 32 accelerators served 1,024 queries at 50.4 tokens per second per user, while GB300's 48 accelerators served 48 queries at 56.2; the Vera Rubin results are pending MLCommons verification and peer review. The benchmark checks generated tool calls and workflow intent, with separate SWE-bench Verified evaluations for end-to-end task solving. The excerpt attributes performance to hardware and serving co-design, including HBM4, Tensor Cores, NVLink, prefix caching, state reuse, Dynamo routing, disaggregated prefill and decode, and model-specific kernels. It describes comparisons across throughput and responsiveness without interpolating between measured points.

### Source excerpt

Agentic workloads are demanding more from inference systems. AI agents carry growing conversation histories across turns and generate a mix of reasoning, tool...

## NVIDIA Vera Rubin NVL72 Achieves up to 2x Blackwell Ultra Performance in MLPerf Training v6.1

DevFeed: [NVIDIA Vera Rubin NVL72 Achieves up to 2x Blackwell Ultra Performance in MLPerf Training v6.1](<https://devfeed.tech/articles/nvidia-vera-rubin-nvl72-achieves-up-to-2x-blackwell-ultra-performance-in-mlperf-training-v6-1-136677.md>)

Original publisher: [Read original article](<https://developer.nvidia.com/blog/nvidia-vera-rubin-nvl72-achieves-up-to-2x-blackwell-ultra-performance-in-mlperf-training-v6-1/>)

Author: Farshad Ghodsian

Published: 2026-11-16T16:00:00Z

Content type: news

Language: en

Sources: [NVIDIA Developer](<https://devfeed.tech/sources/nvidia-developer.md>), [NVIDIA Technical Blog](<https://devfeed.tech/sources/nvidia-technical-blog.md>)

Topics: [Vera Rubin NVL72](<https://devfeed.tech/topics/vera-rubin-nvl72.md>), [NVIDIA Vera Rubin](<https://devfeed.tech/topics/nvidia-vera-rubin.md>), [grpo](<https://devfeed.tech/topics/grpo.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [GB300 NVL72](<https://devfeed.tech/topics/gb300-nvl72.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [NVIDIA Vera](<https://devfeed.tech/topics/nvidia-vera.md>), [NeMo](<https://devfeed.tech/topics/nemo.md>), [agentic workflows](<https://devfeed.tech/topics/agentic-workflows.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [NVFP4](<https://devfeed.tech/topics/nvfp4.md>), [NVIDIA Spectrum-X Ethernet](<https://devfeed.tech/topics/nvidia-spectrum-x-ethernet.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [a-team](<https://devfeed.tech/tags/a-team.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-ai-generative-ai](<https://devfeed.tech/tags/agentic-ai-generative-ai.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [blackwell](<https://devfeed.tech/tags/blackwell.md>), [blackwell-ultra](<https://devfeed.tech/tags/blackwell-ultra.md>), [cloud-services](<https://devfeed.tech/tags/cloud-services.md>), [data-center-cloud](<https://devfeed.tech/tags/data-center-cloud.md>), [featured](<https://devfeed.tech/tags/featured.md>), [llm-benchmarking](<https://devfeed.tech/tags/llm-benchmarking.md>), [mlperf](<https://devfeed.tech/tags/mlperf.md>), [models](<https://devfeed.tech/tags/models.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-vera](<https://devfeed.tech/tags/nvidia-vera.md>), [nvidia-vera-rubin](<https://devfeed.tech/tags/nvidia-vera-rubin.md>), [nvl72](<https://devfeed.tech/tags/nvl72.md>), [performance](<https://devfeed.tech/tags/performance.md>), [rubin](<https://devfeed.tech/tags/rubin.md>), [software-defined-data-center](<https://devfeed.tech/tags/software-defined-data-center.md>), [time](<https://devfeed.tech/tags/time.md>), [top-stories](<https://devfeed.tech/tags/top-stories.md>), [train](<https://devfeed.tech/tags/train.md>), [training](<https://devfeed.tech/tags/training.md>), [vera-rubin](<https://devfeed.tech/tags/vera-rubin.md>), [workflows](<https://devfeed.tech/tags/workflows.md>)

### AI overview

In MLPerf Training v6.1, NVIDIA reports that Vera Rubin NVL72 had the fastest time on six of seven benchmarks and delivered 1.4-1.6x the performance of GB300 NVL72 on most workloads. On DeepSeek-V3 671B at 1,024 GPUs, Vera Rubin finished in 4.14 minutes versus GB300's 8.03; GB300 recorded the round's fastest absolute time by scaling to 10,240 GPUs. Vera Rubin also led the new Qwen 3.5 397B GRPO benchmark (10.44 versus 13.78 minutes at 1,024 GPUs), with results using both NeMo RL with TensorRT-LLM and vLLM. The article attributes performance and scaling to the CUDA software stack, rack-level NVLink 6 and cluster networking, and Vera CPUs supporting simulated RL environments. It describes subsequent runs on newer software as unverified and outside MLCommons review.

### Source excerpt

As AI agents tackle increasingly complex workflows, reducing time-to-train is critical for rapidly iterating on both models and the agents they power. MLPerf...

## Zero-SM NCCL: A Systematic Guide to Designing and Deploying SM-Free GPU Communication

DevFeed: [Zero-SM NCCL: A Systematic Guide to Designing and Deploying SM-Free GPU Communication](<https://devfeed.tech/articles/zero-sm-nccl-a-systematic-guide-to-designing-and-deploying-sm-free-gpu-communication-134894.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.18075>)

Author: Zhenhao He, Siyuan Shen, Ke Wen, Prajwal Singhania, Shiqing Fan, Youngeun Kwon, Sangkug Lym, Vasudevan Rengasamy, Vishnuswaroop Ramesh, Yicheng Qiang, Craig Belusar, Jeff Hammond, Torsten Hoefler

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>)

Topics: [NCCL](<https://devfeed.tech/topics/nccl.md>), [rdma](<https://devfeed.tech/topics/rdma.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>), [Distributed Systems & Parallel Computing](<https://devfeed.tech/topics/distributed-systems-parallel-computing.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [GPU](<https://devfeed.tech/topics/gpu.md>)

Tags: [accelerate](<https://devfeed.tech/tags/accelerate.md>), [apis](<https://devfeed.tech/tags/apis.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [building](<https://devfeed.tech/tags/building.md>), [communication](<https://devfeed.tech/tags/communication.md>), [copy](<https://devfeed.tech/tags/copy.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cuda](<https://devfeed.tech/tags/cuda.md>), [data](<https://devfeed.tech/tags/data.md>), [data-movement](<https://devfeed.tech/tags/data-movement.md>), [deep](<https://devfeed.tech/tags/deep.md>), [designing](<https://devfeed.tech/tags/designing.md>), [domain](<https://devfeed.tech/tags/domain.md>), [domains](<https://devfeed.tech/tags/domains.md>), [free](<https://devfeed.tech/tags/free.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [guide](<https://devfeed.tech/tags/guide.md>), [nccl](<https://devfeed.tech/tags/nccl.md>), [sm](<https://devfeed.tech/tags/sm.md>), [zero](<https://devfeed.tech/tags/zero.md>)

### AI overview

Zero-SM NCCL avoids SM kernels by synchronizing with CUDA stream memory operations, using copy engines for transfers within a scale-up domain, and driving network RDMA through a stream-ordered CPU proxy. It supports AllGather and AlltoAll, plus one-sided ncclPutSignal and ncclWaitSignal APIs. On GB200 NVL72 and B200 NVL8, collectives achieved up to 1.36x NCCL's legacy AllGather bandwidth and reduced median overlapped GEMM slowdown from 33% to 4%, with higher per-operation latency. In Megatron-LM, FSDP improved by up to 3.5% and pipeline parallelism by up to 15% with one-sided puts. The gains depend on workload: the approach benefits large transfers in GPU-bound workloads and deep pipelines with large bubbles, but loses on small messages and host-bound workloads.

### Source excerpt

arXiv:2611.18075v1 Announce Type: new Abstract: GPU communication libraries such as NCCL use Streaming Multiprocessors (SMs) for data movement and synchronization, slowing overlapped computation at a cost that grows with interconnect bandwidth. We present Zero-SM NCCL, which communicates within and across scale-up domains without launching any SM kernel. It synchronizes through CUDA stream memory operations, moves data asynchronously with copy engines within a scale-up domain, and drives network RDMA from a stream-ordered CPU proxy. It supports data-movement collectives such as AllGather and AlltoAll, plus one-sided ncclPutSignal and ncclWaitSignal APIs. On GB200 NVL72 and B200 NVL8 systems, Zero-SM collectives reach up to 1.36x higher AllGather bandwidth than NCCL's legacy SM-based kernels and cut the median slowdown of overlapped GEMMs from 33% to 4%, at a higher latency per operation. In Megatron-LM training, they accelerate FSDP by up to 3.5%, and one-sided puts accelerate pipeline parallelism by up to 15%. Zero-SM communication is not a universal win, however. Through an end-to-end study across models and parallelism strategies, we show that it pays off for large transfers in GPU-bound workloads and for deep pipelines with large bubbles, but loses on small messages and host-bound workloads. We distill these findings into guidelines for enabling zero-SM communication in NCCL and building it into other libraries.

## CoreWeave Unleashes the Power of the NVIDIA GB200 NVL72: A Glimpse into the Future of AI

DevFeed: [CoreWeave Unleashes the Power of the NVIDIA GB200 NVL72: A Glimpse into the Future of AI](<https://devfeed.tech/articles/coreweave-unleashes-the-power-of-the-nvidia-gb200-nvl72-a-glimpse-into-the-future-of-ai-133531.md>)

Original publisher: [Read original article](<https://wf.coreweave.com/blog/coreweave-unleashes-the-power-of-the-nvidia-gb200-nvl72-a-glimpse-into-the-future-of-ai>)

Author: CoreWeave

Published: 2026-11-13T20:57:46Z

Content type: article

Language: en

Sources: [Coreweave Blog Posts](<https://devfeed.tech/sources/coreweave-blog-posts.md>)

Topics: [GB200](<https://devfeed.tech/topics/gb200.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [NCCL](<https://devfeed.tech/topics/nccl.md>), [Distributed Systems & Parallel Computing](<https://devfeed.tech/topics/distributed-systems-parallel-computing.md>), [observability](<https://devfeed.tech/topics/observability.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [blog](<https://devfeed.tech/tags/blog.md>), [cooling](<https://devfeed.tech/tags/cooling.md>), [coreweave](<https://devfeed.tech/tags/coreweave.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [energy](<https://devfeed.tech/tags/energy.md>), [energy-efficiency](<https://devfeed.tech/tags/energy-efficiency.md>), [future](<https://devfeed.tech/tags/future.md>), [gb200](<https://devfeed.tech/tags/gb200.md>), [infiniband](<https://devfeed.tech/tags/infiniband.md>), [innovation](<https://devfeed.tech/tags/innovation.md>), [kubernetes](<https://devfeed.tech/tags/kubernetes.md>), [networking](<https://devfeed.tech/tags/networking.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-gb200](<https://devfeed.tech/tags/nvidia-gb200.md>), [nvl72](<https://devfeed.tech/tags/nvl72.md>), [observability](<https://devfeed.tech/tags/observability.md>), [performance](<https://devfeed.tech/tags/performance.md>), [power](<https://devfeed.tech/tags/power.md>), [quantum](<https://devfeed.tech/tags/quantum.md>), [support](<https://devfeed.tech/tags/support.md>)

### AI overview

CoreWeave demonstrated an NVIDIA GB200 NVL72 rack with distributed GPU benchmarks and a Megatron training run using CoreWeave Kubernetes Service (CKS) and Slurm on Kubernetes (SUNK). The company says CKS provides hypervisor-free Kubernetes tailored to AI workloads, while SUNK adds burst and batch job scheduling. CoreWeave Observability Services tracked GPU use, memory, power, and temperatures; dashboards also showed cooling output responding to workload and energy use scaling with performance. The rack uses Quantum-2 InfiniBand with 400 Gb/s per GPU and SHARP in-network collective communications. CoreWeave says it plans to deploy Quantum-X800 at 800 Gb/s per GPU. The text describes a live demonstration but provides no benchmark results or quantified efficiency comparisons.

### Source excerpt

We showcased NVIDIA GB200 NVL72 AI performance, energy efficiency, and cooling innovation, backed by Kubernetes support, observability tools, and Quantum-2 InfiniBand networking.

## DeepSeek-V4.1-Flash on Vera Rubin NVL72: 58x Throughput vs. GB300 NVL72 at 387 tokens/s

DevFeed: [DeepSeek-V4.1-Flash on Vera Rubin NVL72: 58x Throughput vs. GB300 NVL72 at 387 tokens/s](<https://devfeed.tech/articles/deepseek-v4-1-flash-on-vera-rubin-nvl72-58x-throughput-vs-gb300-nvl72-at-387-tokens-s-132530.md>)

Original publisher: [Read original article](<https://vllm.ai/blog/2026-11-12-deepseek-v41-flash-vera-rubin>)

Author: Inferact and the vLLM Team

Published: 2026-11-12T00:00:00Z

Content type: news

Language: en

Sources: [vLLM Blog](<https://devfeed.tech/sources/vllm-blog.md>)

Topics: [Vera Rubin NVL72](<https://devfeed.tech/topics/vera-rubin-nvl72.md>), [performance-engineering](<https://devfeed.tech/topics/performance-engineering.md>), [vllm](<https://devfeed.tech/topics/vllm.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [ai-engineering](<https://devfeed.tech/tags/ai-engineering.md>), [deepseek](<https://devfeed.tech/tags/deepseek.md>), [deepseek-v4-1-flash](<https://devfeed.tech/tags/deepseek-v4-1-flash.md>), [disaggregation](<https://devfeed.tech/tags/disaggregation.md>), [gb300](<https://devfeed.tech/tags/gb300.md>), [gb300-nvl72](<https://devfeed.tech/tags/gb300-nvl72.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [large-scale-serving](<https://devfeed.tech/tags/large-scale-serving.md>), [llm-inference](<https://devfeed.tech/tags/llm-inference.md>), [model-serving](<https://devfeed.tech/tags/model-serving.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-vera](<https://devfeed.tech/tags/nvidia-vera.md>), [nvidia-vera-rubin](<https://devfeed.tech/tags/nvidia-vera-rubin.md>), [nvl72](<https://devfeed.tech/tags/nvl72.md>), [open-source-ai](<https://devfeed.tech/tags/open-source-ai.md>), [performance](<https://devfeed.tech/tags/performance.md>), [rack](<https://devfeed.tech/tags/rack.md>), [rubin](<https://devfeed.tech/tags/rubin.md>), [semianalysis](<https://devfeed.tech/tags/semianalysis.md>), [throughput](<https://devfeed.tech/tags/throughput.md>), [v4-1-flash](<https://devfeed.tech/tags/v4-1-flash.md>), [vera](<https://devfeed.tech/tags/vera.md>), [vera-rubin](<https://devfeed.tech/tags/vera-rubin.md>), [vera-rubin-nvl72](<https://devfeed.tech/tags/vera-rubin-nvl72.md>), [vllm](<https://devfeed.tech/tags/vllm.md>), [vllm-blog](<https://devfeed.tech/tags/vllm-blog.md>)

### AI overview

On a Vera Rubin NVL72 rack, vLLM reports 1.7M tokens/s per GPU on DeepSeek-V4.1-Flash at 387 tokens/s per user, described as 58x GB300 NVL72 at the same interactivity. The gains come from kernels adapted to Rubin's separate memory domains and wider FP4-capable MMA instructions, a specialized TP8 decode path, and shared rack memory pools. The excerpt reports roughly 28% faster DEP2 prefill steps and 25% faster DEP32 decode steps; TP8 reaches 1,810 tokens/s at batch size 1 in the TL;DR. Mooncake pools KV data across DRAM and spare HBM, while Engram tables are split across GPU HBM and accessed over NVLink. At concurrency 4,800, pools plus workload-aware routing on three DEP2 instances serve 68% more requests, reduce TTFT P50 from 29.8 to 4.2 seconds, and raise prefix hit rate from 91.8% to 96.1%. The HBM pool is described as experimental.

### Source excerpt

On one NVIDIA Vera Rubin NVL72 rack, vLLM serves DeepSeek-V4.1-Flash on SemiAnalysis AgentX at 1.7M tokens/s per GPU at a P90 of 387 tokens/s per user, 58x GB300 NVL72 at the same interactivity.

## Decode Token Parallelism: Memory-Efficient Distributed Attention Parallelism for Scalable LLM Decoding

DevFeed: [Decode Token Parallelism: Memory-Efficient Distributed Attention Parallelism for Scalable LLM Decoding](<https://devfeed.tech/articles/decode-token-parallelism-memory-efficient-distributed-attention-parallelism-for-scalable-llm-decoding-127801.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.13293>)

Author: Susav Shrestha, Brad Settlemyer, Narasimha Reddy

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [sharding](<https://devfeed.tech/topics/sharding.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>), [vllm](<https://devfeed.tech/topics/vllm.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [long-context](<https://devfeed.tech/topics/long-context.md>), [Nemotron](<https://devfeed.tech/topics/nemotron.md>)

Tags: [1-7](<https://devfeed.tech/tags/1-7.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [aggregation](<https://devfeed.tech/tags/aggregation.md>), [all](<https://devfeed.tech/tags/all.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [attention](<https://devfeed.tech/tags/attention.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [batch](<https://devfeed.tech/tags/batch.md>), [block](<https://devfeed.tech/tags/block.md>), [blocks](<https://devfeed.tech/tags/blocks.md>), [but](<https://devfeed.tech/tags/but.md>), [cache](<https://devfeed.tech/tags/cache.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [decode](<https://devfeed.tech/tags/decode.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [llm](<https://devfeed.tech/tags/llm.md>), [memory](<https://devfeed.tech/tags/memory.md>), [parallelism](<https://devfeed.tech/tags/parallelism.md>), [token](<https://devfeed.tech/tags/token.md>)

### AI overview

TKNP separates model-weight placement from attention ownership to scale long-context decoding without replicating weights. It distributes one model copy across nodes in balanced layer blocks: the node holding each block runs projections and MLPs for the full batch with intra-node tensor parallelism, while every node attends to its requests using locally resident KV cache. Only current-token activations cross nodes, dense all-reduces stay within NVLink, and attention needs no cross-node aggregation. The abstract contrasts this with data parallelism, which limits KV capacity through weight replication, and DCP, which shards sequences but aggregates partial attention results across nodes. In vLLM evaluations on Llama 3.1 405B and Nemotron-4 340B using up to 64 H100 GPUs, TKNP reported up to 1.7x higher completion throughput than DCP and 2.06x than data parallelism at matched GPU count for long-context, high-concurrency workloads.

### Source excerpt

arXiv:2611.13293v1 Announce Type: new Abstract: Serving large language models at high concurrency with long contexts requires KV-cache capacity and attention bandwidth that grow with the decode pool. Existing parallelization strategies trade one against the other: data parallelism replicates model weights, limiting KV capacity; tensor parallelism stops scaling attention at the model's KV-head count and pushes dense collectives across nodes; and decode context parallelism (DCP) recovers capacity by sharding the sequence but adds cross-node aggregation of partial attention results. We observe that during decode, dense layers and attention scale differently: dense layers amortize weight reads across the batch and run efficiently within a single NVLink domain, whereas attention cost grows with every active request's KV cache. We present Decode Token Parallelism (TKNP), a decode-pool architecture that decouples weight placement from attention ownership. TKNP distributes a single model copy across nodes in balanced layer blocks. The active node holding a block executes its projections and MLP for the full batch using intra-node tensor parallelism, while every node computes attention for its own requests against locally resident KV cache. This extends request-local attention across nodes without replicating weights: only current-token activations cross the network, dense all-reduces remain within NVLink, and attention requires no cross-node merging of partial results. Implemented in vLLM and evaluated on Llama 3.1 405B and Nemotron-4 340B on up to 64 H100 GPUs, TKNP improves completion throughput for long-context, high-concurrency workloads by up to 1.7$\times$ over DCP and by up to 2.06$\times$ over data parallelism at matched GPU count.

## One-Sided Expert-Parallel All-to-All for Mixture-of-Experts Inference over NVLink

DevFeed: [One-Sided Expert-Parallel All-to-All for Mixture-of-Experts Inference over NVLink](<https://devfeed.tech/articles/one-sided-expert-parallel-all-to-all-for-mixture-of-experts-inference-over-nvlink-127800.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.13188>)

Author: Bo Li, Chulian Zhang, Or Biri, Anthony Chang, Anton Korzh, Noam Duvdevani, XingFei Xi, Daniel Stokes, Zhenhuan Chen, Ludwig Schneider, Artem Y. Polyakov, Julien Demouth, Chandler Zhou

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [GPU optimization](<https://devfeed.tech/topics/gpu-optimization.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>)

Tags: [activation](<https://devfeed.tech/tags/activation.md>), [backend](<https://devfeed.tech/tags/backend.md>), [batch](<https://devfeed.tech/tags/batch.md>), [blackwell](<https://devfeed.tech/tags/blackwell.md>), [communication](<https://devfeed.tech/tags/communication.md>), [communications](<https://devfeed.tech/tags/communications.md>), [compute](<https://devfeed.tech/tags/compute.md>), [copy](<https://devfeed.tech/tags/copy.md>), [costs](<https://devfeed.tech/tags/costs.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-pf](<https://devfeed.tech/tags/cs-pf.md>), [data-movement](<https://devfeed.tech/tags/data-movement.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [deepep](<https://devfeed.tech/tags/deepep.md>), [deepgemm](<https://devfeed.tech/tags/deepgemm.md>), [dispatch](<https://devfeed.tech/tags/dispatch.md>), [expert-parallelism](<https://devfeed.tech/tags/expert-parallelism.md>), [experts](<https://devfeed.tech/tags/experts.md>), [frameworks](<https://devfeed.tech/tags/frameworks.md>), [inference](<https://devfeed.tech/tags/inference.md>), [mixture-of-experts](<https://devfeed.tech/tags/mixture-of-experts.md>), [nvlink](<https://devfeed.tech/tags/nvlink.md>)

### AI overview

NVLinkOneSided reduces fixed communication overhead in small-batch MoE inference by having each source GPU push tokens directly into rank-major buffers on target GPUs, without a prior metadata exchange; grouped GEMMs read the data in place rather than from per-expert copies. Its Fence protocol targets high throughput, while Sentinel targets low latency. On Blackwell GPUs, dispatch and combine are 1.2-3.0x faster than DeepEP, with the largest gains at small batch sizes; the resulting MoE layer is up to 2.4x faster than DeepEP with DeepGEMM. It also outperforms fused DeepGEMM MegaMoE at up to 128 tokens per GPU. The method is limited to GPUs within one NVLink domain, and its modular interface is described as independent of activation quantization format and MoE compute backend.

### Source excerpt

arXiv:2611.13188v1 Announce Type: new Abstract: Expert parallelism (EP) shards the experts of a Mixture-of-Experts (MoE) layer across GPUs, so every MoE layer requires two all-to-all communications, namely dispatch and combine. Compared with training, inference processes relatively few tokens per GPU at a time, and at small batch sizes fixed costs that do not grow with the number of tokens, rather than data movement, dominate the latency. We present NVLinkOneSided, a one-sided EP all-to-all for GPUs within a single NVLink domain. Each source rank pushes its tokens, without first exchanging metadata, into rank-major buffers on the target GPUs, and the grouped GEMMs read them in place instead of from a per-expert copy. We provide two protocols for the synchronization that the communication requires, namely Fence and Sentinel, for high-throughput and low-latency scenarios respectively. On Blackwell GPUs, NVLinkOneSided completes dispatch and combine 1.2-3.0$\times$ faster than DeepEP, with the largest speedups at small batch sizes. The resulting MoE layer is up to 2.4$\times$ faster than DeepEP with DeepGEMM. It even outperforms the fused DeepGEMM MegaMoE at up to 128 tokens per GPU, making it still competitive for decoding. We expose NVLinkOneSided through a modular interface, so that it is agnostic to the activation quantization format, decoupled from the MoE compute backend and easy to integrate into inference frameworks.

## Rebellions R200 NPU Racks Get NVLink Fusion, With NVIDIA GPUs on Prefill and NPUs on Decode, Deploying in 2028

DevFeed: [Rebellions R200 NPU Racks Get NVLink Fusion, With NVIDIA GPUs on Prefill and NPUs on Decode, Deploying in 2028](<https://devfeed.tech/articles/rebellions-r200-npu-racks-get-nvlink-fusion-with-nvidia-gpus-on-prefill-and-npus-on-decode-deploying-in-2028-124327.md>)

Original publisher: [Read original article](<https://www.storagereview.com/news/rebellions-r200-nvidia-nvlink-fusion-prefill-decode-2028>)

Author: Harold Fritts

Published: 2026-11-09T16:05:26Z

Content type: news

Language: en

Sources: [StorageReview.com](<https://devfeed.tech/sources/storagereview-com.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-data](<https://devfeed.tech/tags/ai-data.md>), [ai-data-center](<https://devfeed.tech/tags/ai-data-center.md>), [ai-inference](<https://devfeed.tech/tags/ai-inference.md>), [ai-inference-infrastructure](<https://devfeed.tech/tags/ai-inference-infrastructure.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [build](<https://devfeed.tech/tags/build.md>), [data](<https://devfeed.tech/tags/data.md>), [data-center](<https://devfeed.tech/tags/data-center.md>), [decode](<https://devfeed.tech/tags/decode.md>), [deploy](<https://devfeed.tech/tags/deploy.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [fusion](<https://devfeed.tech/tags/fusion.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [npu](<https://devfeed.tech/tags/npu.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-gpus](<https://devfeed.tech/tags/nvidia-gpus.md>), [nvlink](<https://devfeed.tech/tags/nvlink.md>), [nvlink-fusion](<https://devfeed.tech/tags/nvlink-fusion.md>), [prefill](<https://devfeed.tech/tags/prefill.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

NVIDIA, Rebellions, and SK Telecom plan to deploy R200 NPU systems in South Korea in 2028, using NVIDIA's MGX rack architecture and NVLink Fusion to connect the processors. Their proposed inference split assigns prompt prefill to NVIDIA GPUs and token-by-token decode to Rebellions NPUs, aiming for higher throughput and better power efficiency. The companies say the NPU is designed for low-latency, energy-efficient inference on frontier models, including multitrillion-parameter mixture-of-experts models. SK Telecom plans to run its models, including A.X K2, and services on the systems. These benefits remain design goals: the excerpt reports no performance results, R200 specifications, per-NPU NVLink bandwidth, sampling schedule, or pricing.

### Source excerpt

NVIDIA, Rebellions, and SK Telecom announced a plan to build sovereign AI inference infrastructure in South Korea around Rebellions' next-generation R200 neural processing unit, connected inside the rack with NVIDIA NVLink Fusion and built on the NVIDIA MGX rack architecture. SK Telecom will deploy the systems in its AI data center alongside its models and The post Rebellions R200 NPU Racks Get NVLink Fusion, With NVIDIA GPUs on Prefill and NPUs on Decode, Deploying in 2028 appeared first on StorageReview.com.

## From Chips to Skills, NVIDIA and Korea Expand Sovereign AI Collaboration

DevFeed: [From Chips to Skills, NVIDIA and Korea Expand Sovereign AI Collaboration](<https://devfeed.tech/articles/from-chips-to-skills-nvidia-and-korea-expand-sovereign-ai-collaboration-123631.md>)

Original publisher: [Read original article](<https://blogs.nvidia.com/blog/korea-sovereign-ai-collaboration/>)

Author: Jesse Clayton

Published: 2026-11-09T05:00:04Z

Content type: news

Language: en

Sources: [NVIDIA](<https://devfeed.tech/sources/nvidia-blog.md>)

Topics: [Sovereign AI](<https://devfeed.tech/topics/sovereign-ai.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [Learning](<https://devfeed.tech/topics/learning.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>), [NeMo](<https://devfeed.tech/topics/nemo.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [chips](<https://devfeed.tech/tags/chips.md>), [collaboration](<https://devfeed.tech/tags/collaboration.md>), [data-center](<https://devfeed.tech/tags/data-center.md>), [data-center-infrastructure](<https://devfeed.tech/tags/data-center-infrastructure.md>), [dynamo](<https://devfeed.tech/tags/dynamo.md>), [ecosystem](<https://devfeed.tech/tags/ecosystem.md>), [fusion](<https://devfeed.tech/tags/fusion.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [korea](<https://devfeed.tech/tags/korea.md>), [korean](<https://devfeed.tech/tags/korean.md>), [models](<https://devfeed.tech/tags/models.md>), [npu](<https://devfeed.tech/tags/npu.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvlink](<https://devfeed.tech/tags/nvlink.md>), [nvlink-fusion](<https://devfeed.tech/tags/nvlink-fusion.md>), [security](<https://devfeed.tech/tags/security.md>), [semiconductors](<https://devfeed.tech/tags/semiconductors.md>), [skills](<https://devfeed.tech/tags/skills.md>), [sovereign-ai](<https://devfeed.tech/tags/sovereign-ai.md>)

### AI overview

NVIDIA and Korea's MSIT are extending sovereign AI collaboration across semiconductors, infrastructure, talent, startups, models, agents and security. The partnership lays groundwork for Korean NPU and XPU companies to explore joining the NVIDIA NVLink Fusion ecosystem, alongside NVIDIA training, certification, hackathons and technical seminars aimed at strengthening GPU utilization. Through Korea's AI for All initiative, NVIDIA will work with three selected consortia and four foundation model builders, using NVIDIA Dynamo to optimize and scale inference. A separate collaboration with SK Telecom and Rebellions, supported by MSIT, will advance sovereign infrastructure for Korean models and workloads with NVLink Fusion.

### Source excerpt

NVIDIA and Korea's Ministry of Science and ICT (MSIT) are expanding their collaboration across the country's sovereign AI ecosystem, from AI semiconductors and data center infrastructure to talent, startups, models, agents and security. The collaboration lays the groundwork for Korean NPU and XPU companies to explore participation in the NVIDIA NVLink Fusion ecosystem, and includes [...]

## ISER-MoE: Dynamic In-Switch Replication of Hot Experts for Load Balancing in Expert-Parallel

DevFeed: [ISER-MoE: Dynamic In-Switch Replication of Hot Experts for Load Balancing in Expert-Parallel](<https://devfeed.tech/articles/iser-moe-dynamic-in-switch-replication-of-hot-experts-for-load-balancing-in-expert-parallel-123683.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.09008>)

Author: Jintao Peng, Ethan Zhang, Kefeng Duan, Bangyu Sheng, Min Li, Itamar Rosenfeld Rauch, Jack Yang, June Yang

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>)

Topics: [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>), [GPU optimization](<https://devfeed.tech/topics/gpu-optimization.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>), [GPU](<https://devfeed.tech/topics/gpu.md>), [TensorRT-LLM](<https://devfeed.tech/topics/tensorrt-llm.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [balancing](<https://devfeed.tech/tags/balancing.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [broadcast](<https://devfeed.tech/tags/broadcast.md>), [broadcasting](<https://devfeed.tech/tags/broadcasting.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [deepseek](<https://devfeed.tech/tags/deepseek.md>), [experts](<https://devfeed.tech/tags/experts.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [groups](<https://devfeed.tech/tags/groups.md>), [latency](<https://devfeed.tech/tags/latency.md>), [layer](<https://devfeed.tech/tags/layer.md>), [llm](<https://devfeed.tech/tags/llm.md>), [load](<https://devfeed.tech/tags/load.md>), [load-balancing](<https://devfeed.tech/tags/load-balancing.md>), [moe](<https://devfeed.tech/tags/moe.md>), [parallel](<https://devfeed.tech/tags/parallel.md>), [replication](<https://devfeed.tech/tags/replication.md>), [switch](<https://devfeed.tech/tags/switch.md>)

### AI overview

ISER-MoE addresses hot-expert imbalance in expert-parallel MoE by multicasting expert weights through the NVLink switch to fixed GPU groups, so each weight byte leaves its owner once. A GPU scheduler assigns hot experts to groups per layer, while multi-port broadcasts spread data across group ports; both operations overlap with the shared expert. The approach is implemented in TensorRT-LLM. In trace replay, scheduling removed 98-99% of excess load. In DeepSeek-V4-Pro prefill serving, throughput increased 17.4% on a math benchmark and 7.2% on agentic prompts.

### Source excerpt

arXiv:2611.09008v1 Announce Type: new Abstract: Expert-parallel mixture-of-experts (MoE) layers run at the pace of the most loaded GPU, and a few hot experts carry most of the excess load. Replicating them at every MoE layer balances the load only if their weights arrive within the layer, and broadcasting from the owner is the bottleneck: point-to-point copies overload it, and baseline broadcasts pay latency or bandwidth overheads. We present ISER-MoE (In-Switch Expert Replication for MoE), built in TensorRT-LLM, which multicasts hot experts through the NVLink switch to fixed GPU groups, so each byte leaves its owner once. A GPU scheduler assigns hot experts to groups per layer, and a multi-port broadcast spreads the injection over the group's ports, both overlapped with the shared expert. In trace replay, the scheduler removed 98-99% of the excess load; in prefill serving of DeepSeek-V4-Pro, throughput rose 17.4% on a math benchmark and 7.2% on agentic prompts.

## FastAFD: When Does Attention-FFN Disaggregation Pay on Rack-Scale NVLink?

DevFeed: [FastAFD: When Does Attention-FFN Disaggregation Pay on Rack-Scale NVLink?](<https://devfeed.tech/articles/fastafd-when-does-attention-ffn-disaggregation-pay-on-rack-scale-nvlink-120564.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.07149>)

Author: Yichao Fu, Yuxuan Zhang, Minshen Zhang, Ruitian Wang, Junda Chen, Hao Zhang

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>), [vllm](<https://devfeed.tech/topics/vllm.md>), [GB200](<https://devfeed.tech/topics/gb200.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [groq](<https://devfeed.tech/topics/groq.md>)

Tags: [accelerators](<https://devfeed.tech/tags/accelerators.md>), [attention](<https://devfeed.tech/tags/attention.md>), [batch](<https://devfeed.tech/tags/batch.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [decode](<https://devfeed.tech/tags/decode.md>), [disaggregation](<https://devfeed.tech/tags/disaggregation.md>), [engine](<https://devfeed.tech/tags/engine.md>), [expert-parallelism](<https://devfeed.tech/tags/expert-parallelism.md>), [experts](<https://devfeed.tech/tags/experts.md>), [gb200](<https://devfeed.tech/tags/gb200.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [groq](<https://devfeed.tech/tags/groq.md>), [lpx](<https://devfeed.tech/tags/lpx.md>), [mixture-of-experts](<https://devfeed.tech/tags/mixture-of-experts.md>), [mixture-of-experts-moe](<https://devfeed.tech/tags/mixture-of-experts-moe.md>), [moe](<https://devfeed.tech/tags/moe.md>), [nvl72](<https://devfeed.tech/tags/nvl72.md>), [nvlink](<https://devfeed.tech/tags/nvlink.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [rack-scale](<https://devfeed.tech/tags/rack-scale.md>)

### AI overview

FastAFD separates attention from MoE experts on GB200 NVL72, overlapping one microbatch's MoE work with another's attention, and compares against vLLM's best wide expert-parallel setting. It reports up to 1.17x throughput on Qwen3, 1.16x on Qwen3.5, and parity on MiniMax-M2.5. Its model predicts AFD throughput within 3% from colocated vLLM profiles. The model attributes gains to larger batches and overlap, weighed against the GPUs allocated to experts; AFD pays off when the gains exceed that cost. Speculative decoding particularly helps light-expert models, adding 8% to AFD's advantage over wide EP. The abstract projects an ideal 2.48x speedup with speculative decoding if experts use on-chip SRAM accelerators, which would remove most disaggregation cost.

### Source excerpt

arXiv:2611.07149v1 Announce Type: cross Abstract: Attention-FFN disaggregation (AFD) separates attention from the experts of a mixture-of-experts (MoE) model and has been reported to raise decode throughput over colocated serving. Whether this advantage holds on rack-scale NVLink systems such as GB200 NVL72 remains unclear: prior results were obtained on other hardware, against narrower colocated baselines, or with analytical models, whereas production engines serve large MoE models on these systems with wide expert parallelism (wide EP). We build and tune FastAFD, an AFD engine for GB200 NVL72 that overlaps the MoE of one microbatch with the attention of the other, and compare it against the best vLLM wide EP setting. Experiments show that AFD has a throughput advantage of up to 1.17x over wide EP on Qwen3 and 1.16x on Qwen3.5, and matches it on MiniMax-M2.5. We derive an analytical model that predicts AFD throughput within 3% using only profiles of colocated vLLM, telling whether switching to AFD would pay off before deploying it. The model yields a concise insight: AFD's speedup factors into a batch gain (larger batches, since attention GPUs hold no expert weights), an overlap gain (shorter steps, since the MoE runs on the expert GPUs behind attention), and a disaggregation cost (the GPUs given to experts). AFD has the advantage when the two gains outweigh the cost. This view explains which models benefit from AFD, and shows that speculative decoding helps light-expert models most, where the enlarged overlap gain outweighs the larger disaggregation cost and adds a further 8% to AFD's advantage over wide EP. Looking ahead, serving the experts on accelerators that keep their weights in on-chip SRAM, such as the Groq LPX accelerators paired with Vera Rubin NVL72, would remove most of the disaggregation cost, raising the ideal speedup to 2.48x with speculative decoding. FastAFD is open source at https://github.com/hao-ai-lab/FastAFD.

## Five Ways NVIDIA Co-Design Maximizes the Value of Every Byte in the AI Factory

DevFeed: [Five Ways NVIDIA Co-Design Maximizes the Value of Every Byte in the AI Factory](<https://devfeed.tech/articles/five-ways-nvidia-co-design-maximizes-the-value-of-every-byte-in-the-ai-factory-119477.md>)

Original publisher: [Read original article](<https://developer.nvidia.com/blog/five-ways-nvidia-co-design-maximizes-the-value-of-every-byte-in-the-ai-factory/>)

Author: Shara Tibken

Published: 2026-11-05T16:00:00Z

Content type: article

Language: en

Sources: [NVIDIA Developer](<https://devfeed.tech/sources/nvidia-developer.md>), [NVIDIA Technical Blog](<https://devfeed.tech/sources/nvidia-technical-blog.md>)

Topics: [AI Factory](<https://devfeed.tech/topics/ai-factory.md>), [AI Architecture](<https://devfeed.tech/topics/ai-architecture.md>), [NVFP4](<https://devfeed.tech/topics/nvfp4.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>), [vllm](<https://devfeed.tech/topics/vllm.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-ai-generative-ai](<https://devfeed.tech/tags/agentic-ai-generative-ai.md>), [agentic-workloads](<https://devfeed.tech/tags/agentic-workloads.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai-factory](<https://devfeed.tech/tags/ai-factory.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [arithmetic](<https://devfeed.tech/tags/arithmetic.md>), [attention](<https://devfeed.tech/tags/attention.md>), [batch](<https://devfeed.tech/tags/batch.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [better](<https://devfeed.tech/tags/better.md>), [bits](<https://devfeed.tech/tags/bits.md>), [block](<https://devfeed.tech/tags/block.md>), [bluefield-dpu](<https://devfeed.tech/tags/bluefield-dpu.md>), [cache](<https://devfeed.tech/tags/cache.md>), [cached-context](<https://devfeed.tech/tags/cached-context.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [co-design](<https://devfeed.tech/tags/co-design.md>), [data-center-cloud](<https://devfeed.tech/tags/data-center-cloud.md>), [doca](<https://devfeed.tech/tags/doca.md>), [dynamo](<https://devfeed.tech/tags/dynamo.md>), [ethernet](<https://devfeed.tech/tags/ethernet.md>), [featured](<https://devfeed.tech/tags/featured.md>), [gb200](<https://devfeed.tech/tags/gb200.md>), [gb300](<https://devfeed.tech/tags/gb300.md>), [gb300-nvl72](<https://devfeed.tech/tags/gb300-nvl72.md>), [gpudirect](<https://devfeed.tech/tags/gpudirect.md>), [h100](<https://devfeed.tech/tags/h100.md>), [hbm](<https://devfeed.tech/tags/hbm.md>), [high-bandwidth-memory](<https://devfeed.tech/tags/high-bandwidth-memory.md>), [memory](<https://devfeed.tech/tags/memory.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvlink](<https://devfeed.tech/tags/nvlink.md>), [rubin-gpu](<https://devfeed.tech/tags/rubin-gpu.md>), [spectrum-x](<https://devfeed.tech/tags/spectrum-x.md>), [tensorrt-llm](<https://devfeed.tech/tags/tensorrt-llm.md>), [vera-cpu](<https://devfeed.tech/tags/vera-cpu.md>), [vera-rubin](<https://devfeed.tech/tags/vera-rubin.md>), [vera-rubin-nvl72](<https://devfeed.tech/tags/vera-rubin-nvl72.md>)

### AI overview

The excerpt describes five ways NVIDIA's AI factory design aims to increase useful work per memory byte: NVFP4 shrinks model weights and KV cache; TensorRT-LLM Wide-EP spreads MoE experts across GPUs to free per-GPU memory and improve weight reuse; Dynamo routes requests to cached prefixes and manages KV cache across GPU, CPU, and storage; Vera CPU memory provides a place to stage context during tool-call gaps; and CMX offers pod-level shared flash for context reuse across nodes. The article reports benchmark-specific gains, including up to 4x throughput over an FP8 baseline for NVFP4 and up to 30x throughput per megawatt for Vera Rubin NVL72 versus GB300 NVL72 at the same 288 GB per GPU. It also notes that KV offloading helps when retrieval is faster than recomputation; storage performance examples are vendor validation tests.

### Source excerpt

In an AI factory, high bandwidth memory (HBM)is the fastest, most expensive, and most contested memory tier. Every byte spent on a redundant weight copy, an...

## NVIDIA CUDA 13.5 Brings NVIDIA Vera Rubin GA Support

DevFeed: [NVIDIA CUDA 13.5 Brings NVIDIA Vera Rubin GA Support](<https://devfeed.tech/articles/nvidia-cuda-13-5-brings-nvidia-vera-rubin-ga-support-117642.md>)

Original publisher: [Read original article](<https://developer.nvidia.com/blog/nvidia-cuda-13-5-brings-nvidia-vera-rubin-ga-support/>)

Author: Shara Tibken

Published: 2026-11-04T18:46:07Z

Content type: release

Language: en

Sources: [NVIDIA Developer](<https://devfeed.tech/sources/nvidia-developer.md>), [NVIDIA Technical Blog](<https://devfeed.tech/sources/nvidia-technical-blog.md>)

Topics: [CUDA](<https://devfeed.tech/topics/cuda.md>), [CUDA Tile](<https://devfeed.tech/topics/cuda-tile.md>), [NVIDIA Vera Rubin](<https://devfeed.tech/topics/nvidia-vera-rubin.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>), [CUDA Graphs](<https://devfeed.tech/topics/cuda-graphs.md>), [Tool](<https://devfeed.tech/topics/tool.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [64-bit](<https://devfeed.tech/tags/64-bit.md>), [accelerated-computing](<https://devfeed.tech/tags/accelerated-computing.md>), [acceleration](<https://devfeed.tech/tags/acceleration.md>), [accounting](<https://devfeed.tech/tags/accounting.md>), [agent-skill](<https://devfeed.tech/tags/agent-skill.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-coding-agents](<https://devfeed.tech/tags/ai-coding-agents.md>), [ai-workloads](<https://devfeed.tech/tags/ai-workloads.md>), [arm](<https://devfeed.tech/tags/arm.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [attention](<https://devfeed.tech/tags/attention.md>), [automatic](<https://devfeed.tech/tags/automatic.md>), [backend](<https://devfeed.tech/tags/backend.md>), [cuda](<https://devfeed.tech/tags/cuda.md>), [cuda-tile](<https://devfeed.tech/tags/cuda-tile.md>), [cutile](<https://devfeed.tech/tags/cutile.md>), [developer-tools-techniques](<https://devfeed.tech/tags/developer-tools-techniques.md>), [featured](<https://devfeed.tech/tags/featured.md>), [machine-learning-artificial-intelligence](<https://devfeed.tech/tags/machine-learning-artificial-intelligence.md>), [nsight-tools-compute](<https://devfeed.tech/tags/nsight-tools-compute.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-vera-rubin](<https://devfeed.tech/tags/nvidia-vera-rubin.md>), [python](<https://devfeed.tech/tags/python.md>), [release](<https://devfeed.tech/tags/release.md>), [rubin](<https://devfeed.tech/tags/rubin.md>), [vera-rubin](<https://devfeed.tech/tags/vera-rubin.md>)

### AI overview

CUDA Toolkit 13.5 makes NVIDIA Vera Rubin support generally available and adds platform and workload controls across CUDA. CUDA 13.5 can partition all SMs across locality domains, with discovery mode reporting potentially unequal partition sizes. MPS v3 adds weighted timeslicing, borrowable SM capacity, cgroup routing, namespace assignments, and polling as a memory-accounting fallback; polling adds work during eviction decisions. CUDA Graphs can capture batched 1D copies, and CUDA Tile adds multi-GPU communication plus preview support for calling selected SIMT library functions; custom device-function calls remain planned. CUDA Python and CCCL add batch copies, graph updates, public CUB tuning, batched Top-K, reproducible scans, and reductions that can consume GPU-resident counts. The release also updates profiling and debugging tools, math libraries, compiler controls, PTX, and binary inspection. On Rubin, WDDM and MCDM are supported; TCC is not. The excerpt describes many enhancements but directs readers to release notes for full compatibility details.

### Source excerpt

As artificial intelligence and accelerated computing continue to push the boundaries of what's possible, the latest NVIDIA CUDA release arrives to unlock...

## Convergent Science Advances Engineering Simulation on CoreWeave

DevFeed: [Convergent Science Advances Engineering Simulation on CoreWeave](<https://devfeed.tech/articles/convergent-science-advances-engineering-simulation-on-coreweave-114287.md>)

Original publisher: [Read original article](<https://wf.coreweave.com/blog/convergent-science-advances-engineering-simulation-on-coreweave>)

Author: CoreWeave

Published: 2026-11-03T15:33:33Z

Content type: article

Language: en

Sources: [Coreweave Blog Posts](<https://devfeed.tech/sources/coreweave-blog-posts.md>)

Topics: [Computational Fluid Dynamics (CFD)](<https://devfeed.tech/topics/cfd.md>), [Simulation](<https://devfeed.tech/topics/simulation.md>), [gpudirect rdma](<https://devfeed.tech/topics/gpudirect-rdma.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>)

Tags: [blog](<https://devfeed.tech/tags/blog.md>), [cfd](<https://devfeed.tech/tags/cfd.md>), [coreweave](<https://devfeed.tech/tags/coreweave.md>), [dns](<https://devfeed.tech/tags/dns.md>), [engineering-simulation](<https://devfeed.tech/tags/engineering-simulation.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [scale](<https://devfeed.tech/tags/scale.md>), [simulations](<https://devfeed.tech/tags/simulations.md>), [size](<https://devfeed.tech/tags/size.md>)

### AI overview

Convergent Science ran a 536-million-cell CONVERGE CFD direct numerical simulation to completion in 3.62 hours on two NVIDIA HGX B200 instances (16 Blackwell GPUs). In its scaling study, it also compared NVIDIA RTX PRO 6000 Blackwell Server Edition and HGX B200 using a validated turbulent-flow case. The excerpt distinguishes weak scaling (larger meshes as GPU resources grow) from strong scaling (same mesh completed faster), and explains that memory determines whether a job fits while communication affects speed as GPUs are added. RTX PRO systems target jobs within one server, using PCIe between GPUs; HGX B200 supports larger, multi-node cases with more memory per GPU, NVLink within nodes, and 400 Gb/s InfiniBand between them. Results depend on workload and topology. The tested DNS case was lean; production simulations with additional physics may require more memory at the same cell count, so its figures may underestimate those needs.

### Source excerpt

How far can GPU CFD scale? Convergent Science ran 536 million cells on two NVIDIA HGX B200 instances in 3.62 hours. Here is how to size and scale your own simulations.

## Diagnose Distributed AI Workload Failures with NVIDIA Attribution

DevFeed: [Diagnose Distributed AI Workload Failures with NVIDIA Attribution](<https://devfeed.tech/articles/diagnose-distributed-ai-workload-failures-with-nvidia-attribution-113259.md>)

Original publisher: [Read original article](<https://developer.nvidia.com/blog/diagnose-distributed-ai-workload-failures-with-nvidia-attribution/>)

Author: Elizabeth Goodman

Published: 2026-11-02T23:39:34Z

Content type: tutorial

Language: en

Sources: [NVIDIA Developer](<https://devfeed.tech/sources/nvidia-developer.md>), [NVIDIA Technical Blog](<https://devfeed.tech/sources/nvidia-technical-blog.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [distributed-training](<https://devfeed.tech/topics/distributed-training.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [Monitoring](<https://devfeed.tech/topics/monitoring.md>), [NCCL](<https://devfeed.tech/topics/nccl.md>)

Tags: [accelerated-computing-libraries](<https://devfeed.tech/tags/accelerated-computing-libraries.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-workloads](<https://devfeed.tech/tags/ai-workloads.md>), [data-center-cloud](<https://devfeed.tech/tags/data-center-cloud.md>), [developer-tools-techniques](<https://devfeed.tech/tags/developer-tools-techniques.md>), [diagnose](<https://devfeed.tech/tags/diagnose.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [error](<https://devfeed.tech/tags/error.md>), [errors](<https://devfeed.tech/tags/errors.md>), [fabric](<https://devfeed.tech/tags/fabric.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [mlops](<https://devfeed.tech/tags/mlops.md>), [nccl](<https://devfeed.tech/tags/nccl.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvl72](<https://devfeed.tech/tags/nvl72.md>), [scheduler](<https://devfeed.tech/tags/scheduler.md>)

### AI overview

NVIDIA Attribution diagnoses distributed AI workload failures by correlating workload-scoped logs, metrics, telemetry, and scheduler state with deterministic rules. The Blueprint scopes a workload, collects observations, and calls the Attribution Service, which extracts and correlates events; the Blueprint returns a structured verdict with outcome, attributed domain, affected nodes, symptoms, and collector coverage. In the example, XID 149 on node-02 is identified as the root cause of a signal-integrity failure, while downstream NCCL errors are retained as evidence. When evidence supports multiple causes, the result can mark the sub-domain ambiguous; diagnosis quality depends on telemetry completeness. The self-hosted HTTP Service runs the core engine, while the customizable Python Blueprint integrates schedulers and data sources. Its quickstart uses Docker and uv with sample evidence. Rules ship with the Service and are not uploaded dynamically through its API.

### Source excerpt

In distributed AI workloads, a single error can produce a cascade of symptoms. A GPU or fabric event may surface as NCCL errors on other ranks, scheduler state...

[Next page](<https://devfeed.tech/topics/nvlink.md?cursor=WyIyMDI2LTExLTAyVDIzOjM5OjM0KzAwOjAwIiwgIjllOTlmMTIxLTIzN2MtNDk4Zi05NDU4LWFjOTdmMzU5NjUwYiJd>)