22 results for “NVL72”
NVIDIA declared Scale-In the fifth pillar of AI networking infrastructure, powered by BlueField-4 DPUs, DOCA software, and Spectrum-X Ethernet. The architecture offloads security, storage, and telemetry from host CPUs at up to 800 Gb/s to keep pace with agentic AI factories.
NVIDIA detailed how Spectrum-X Ethernet uses hardware-accelerated adaptive routing, congestion control, and plane load balancing to solve traditional Ethernet's failures at giga-scale AI training. The platform delivers 1.6x performance over off-the-shelf Ethernet, near-perfect multi-tenant isolation, and 2.68 ms failover versus 1.08 seconds for standard Ethernet.
NVIDIA published a practical observability framework mapping telemetry tools to every layer of DGX and HGX AI factory infrastructure. The guide helps teams catch gray failures like InfiniBand bit-error drift before they waste multi-hour training runs.
Nvidia has formally opened the cuFile APIs and the vertical storage stack beneath them, moving the interface that lets GPUs read and write storage. Opening cuFile is a strategic departure; this layer has historically lived inside CUDA.
Nvidia has agreed to invest up to $3 billion in Lancium, the Texas-based power infrastructure developer behind the first operational site of the Stargate. Power has replaced silicon as the scarcest resource in frontier AI development.
NVIDIA's Groq 3 LPX accelerator hit 3,431 output tokens per second on the Artificial Analysis 100K context benchmark with Gemma 4 31B. The result enables responsive multi-agent systems that can process full session histories at long context lengths.
NVIDIA hand-delivered its first Vera CPUs to Anthropic, OpenAI, SpaceXAI, and Oracle Cloud Infrastructure in mid-May 2026, correcting earlier reports of late August shipments. The 88-core Olympus processor targets agentic AI workloads where CPU orchestration gates GPU utilization.
NVIDIA launched NVLink Fusion with NVHBM on August 26, 2026, giving hyperscalers a standardized path to deploy custom XPUs alongside NVIDIA GPUs. The co-designed memory and interconnect stack promises 30% higher bandwidth, 25% more compute area, and 15% lower HBM power versus standard HBM4e.
NVIDIA published Vera CPU benchmarks showing up to 3.67x faster CRC32C integrity checks and 3.26x faster Reed-Solomon recovery versus an unnamed x86 CPU. The gains target the storage processing bottleneck that can throttle agentic AI workloads.
NVIDIA committed to a 20-year land, power, and shell agreement at the PORTS-Pike campus in Ohio, securing 4.25 GW of AI factory capacity for OpenAI as anchor tenant. The deal frames physical infrastructure as a reusable asset across multiple GPU generations, unlocking an estimated $600 billion in NVIDIA compute revenue through 2030.