NVLink
PulseAugur coverage of NVLink — every cluster mentioning NVLink across labs, papers, and developer communities, ranked by signal.
6 day(s) with sentiment data
-
New research optimizes disaggregated LLM inference with topology-aware data movement
A new research paper introduces a topology-aware data movement system designed to optimize disaggregated LLM inference. The system addresses the challenge of transferring KV caches between separate GPU pools by discover…
-
Nvidia shows Vera Rubin NVL72 racks running OpenAI workloads
Nvidia has provided an exclusive look inside its Engineering SuperLab, showcasing the Vera Rubin NVL72 rack system in operation. This facility, located near Nvidia's headquarters, is designed for engineers to test and d…
-
NVIDIA Vera Rubin platform expands globally, boosting AI factory efficiency · 4 sources tracked
NVIDIA has announced the global expansion and ramp-up of its Vera Rubin platform, designed for large-scale AI factories. This system, built with extreme co-design across seven chips and five rack trays, aims to deliver …
-
NVIDIA NVLink targets AI factories with scale-up networking
NVIDIA's NVLink technology is being positioned as a crucial networking solution for AI factories. It is designed to handle the increasing demands of large-scale AI workloads and complex model training. The technology ai…
-
Nvidia CEO denies Vera Rubin AI platform delays, cites 'giant amounts' in production
Nvidia CEO Jensen Huang has refuted claims of production delays for the company's upcoming Vera Rubin AI platform, stating that it is already in production and will be delivered in 'giant amounts.' While Huang confirmed…
-
AI model serving speedups decomposed: runtime gains dominate, quantization offers minor boost · 2 sources tracked
A new research paper details a method for decomposing speedups in AI model serving, separating gains from runtime, kernel optimizations, and quantization. The study, conducted on four NVIDIA RTX A5000 GPUs, found that r…
-
LocalLLaMA users discuss SXM2 GPU costs for LLM tasks
A Reddit user is inquiring about the costs and feasibility of using two NVIDIA SXM2 GPUs, potentially V100s or A100s, for local large language model (LLM) tasks. The user is specifically interested in whether NVLink can…
-
Frontier LLMs struggle with multi-GPU kernel generation, new benchmark reveals
A new benchmark called ParallelKernelBench (PKB) has been developed to evaluate the ability of frontier large language models to generate efficient multi-GPU kernels. Testing models like GPT-5.5, Gemini 3 Pro, and Opus …
-
Nvidia launches Vera Rubin supercomputing platform for AI and HPC
Nvidia has launched its Vera Rubin supercomputing platform, designed for demanding HPC and AI workloads like climate modeling and energy exploration. The platform integrates Rubin GPUs and Vera CPUs, featuring NVLink, I…
-
NVIDIA Vera Rubin NVL4 system to launch Q4 2026 with 7 exaflops AI power · 2 sources tracked
NVIDIA has announced its Vera Rubin NVL4 system, designed for demanding HPC and AI workloads like climate modeling and fluid dynamics. This platform integrates Rubin GPUs with Vera CPUs, featuring NVLink, InfiniBand, an…
-
Upscale AI raises $190M at $2B valuation for AI networking fabric
Upscale AI, a startup focused on building networking infrastructure for AI data centers, has secured $190 million in Series A-1 funding, valuing the company at $2 billion. This latest round, led by Premji Invest, brings…
-
NVIDIA Blackwell platform dominates MLPerf Training 6.0 benchmarks
NVIDIA's Blackwell platform has set new records in the MLPerf Training 6.0 benchmarks, achieving the fastest times across all seven tests. The platform demonstrated strong scaling, with clusters of up to 8,192 GPUs show…
-
Nvidia H100 GPU Pricing and Alternatives in 2026
In 2026, the Nvidia H100 GPU remains a critical component for AI infrastructure, with purchase prices ranging from $30,000 to over $40,000. Cloud rental costs vary significantly, with specialized GPU clouds offering low…
-
NVIDIA Blackwell platform dominates MLPerf Training 6.0 benchmarks · 4 sources tracked
NVIDIA's Blackwell platform has achieved top performance across all seven benchmarks in the MLPerf Training 6.0 industry standard tests. The platform demonstrated the fastest training times and enabled the largest-scale…
-
RTX Spark's memory bandwidth corrected to below 600GB/s
Reports circulating about the RTX Spark's memory bandwidth have been corrected, with new information indicating it does not possess 600GB/s of bandwidth. This figure actually refers to the NvLink speed, not the memory b…
-
LLMs and new frameworks boost GPU kernel optimization
Researchers are exploring novel ways to optimize GPU kernel performance for large language models. One approach uses language models as surrogates to predict kernel performance, significantly increasing the number of ca…
-
NVENC Bridge Enables Multi-GPU AI Over Ethernet, Bypassing NVLink
Researchers have developed a custom NVENC encoder bridge that allows multiple GPUs to be linked over standard Ethernet or Wi-Fi, bypassing the need for expensive NVLink hardware. This innovation enables the splitting of…
-
Astera Labs launches Scorpio X, an open AI fabric switch challenging Nvidia's NVSwitch
Astera Labs has introduced the Scorpio X, a new AI fabric switch designed as an alternative to Nvidia's NVSwitch. This ASIC offers 320 lanes of PCIe 6.0 connectivity and 5.12 TB/s of bidirectional bandwidth, aiming to p…
-
Multi-node training enables scaling foundation models across GPU clusters
Training large foundation models necessitates distributing the workload across numerous GPUs housed in multiple interconnected machines, a process known as multi-node training. This approach is essential for handling mo…