Cisco is championing an 'Ethernet-first' strategy for AI infrastructure, arguing that Ethernet is becoming the predominant networking technology for AI supercomputing due to its open standards, scalab…
A new arXiv paper (arXiv:2607.28633v1) proposes a topology-aware transfer orchestrator for disaggregated GPU inference, claiming existing systems like DistServe, Splitwise, and Mooncake ignore that ba…
NVIDIA CTO Michael Kagan said at WeAreDevelopers World Congress in Berlin that modern AI is no longer just about chips but about software, networking, and power, describing AI infrastructure as an "AI…
Data center switch spending for AI back-end networks is projected to reach $1 trillion by 2030, with scale-up interconnectivity accounting for more than half the total market, according to a report fr…
Cisco's new white paper, 'A Day in the Life of a Prompt,' warns that AI inference is becoming a networking issue as data movement emerges as the primary bottleneck for GPU performance. The paper forec…
Nvidia's NVLink bridge boosts prompt processing by 30% and FSDP training by nearly 3x on dual RTX 3090s, but token generation and DDP training show no benefit, according to tests by a developer runnin…
NVIDIA GB300 NVL72 set a world record for pre-training DeepSeek-V3 671B at 1,648 TFLOPs per GPU, demonstrating how advances across the AI platform—from silicon to networking to software—push training …
NVIDIA's NVLink and NVSwitch technologies form a scale-up fabric that connects GPUs tightly enough to behave as a single machine, contrasting with scale-out fabrics like InfiniBand and RoCE that link …
NVIDIA NVLink, the sixth-generation scale-up networking fabric for AI factories, delivers the highest GPU-to-GPU bandwidth and lowest latency all-to-all topology to enable accelerators to work as a si…
Nvidia Corp. networking chief Gilad Shainer argues that agentic inference transforms the network into part of the computer, making it a first-class participant in the computational process. In a speci…
Nvidia Corp. networking chief Gilad Shainer argues that in the era of agentic AI, the network has become a first-class participant in computation, transforming AI infrastructure from a server farm int…
NVIDIA researchers introduced Nonuniform Tensor Parallelism (NTP), a framework that dynamically adjusts tensor parallelism to maintain high Goodput during large-scale LLM training despite GPU interrup…
NVIDIA announced that its Confidential Computing technology for Blackwell GPUs achieves up to 98% of the inference performance of non-secure solutions, enabling hardware-rooted AI security without sig…
NVIDIA announced multi-device inference support in TensorRT 11.0, enabling native high-performance multi-GPU inference for generative AI workloads. The feature integrates with NCCL for distributed col…
Enterprise leaders are realizing that AI efficiency depends not only on model optimization but also on co-design of hardware, compilers, runtime, and cloud architecture. Inefficient AI can cost billio…
Microsoft announced a broad set of enhancements to Azure Kubernetes Service at Build 2026, including bare metal support, fleet management, and AI infrastructure features, aiming to make Kubernetes a f…
Frontier LLMs including GPT-5.5, Gemini 3 Pro, and Opus 4.7 failed to correctly solve over two-thirds of problems in ParallelKernelBench, a new benchmark for multi-GPU kernel generation. The benchmark…
NVIDIA swept MLPerf Training v6.0 benchmarks, achieving the fastest training times at scale and highest per-accelerator performance across all tests, including new DeepSeek-V3 and GPT-OSS-20B workload…
NVIDIA's Blackwell platform swept MLPerf Training 6.0, achieving the fastest training times across all seven benchmarks, scaling to 8,192 GPUs on DeepSeek-V3 671B, and delivering up to 1.6x performanc…
A developer found that 4 to 8 dedicated GPUs, such as the H200 NVLink, are sufficient for most production inference workloads on 70B to 200B parameter models, debunking the need for multi-node cluster…