NVIDIA Vera Rubin NVL72

Headlines · 3

1. 4d
   Foxconn shows NVIDIA Vera Rubin rack platform at OCP 2026

   At the 2026 OCP Global Summit, Foxconn (Hon Hai) showcased a new generation of AI Rack technology, with its subsidiary Ingrasys displaying the NVIDIA Vera Rubin NVL72 platform alongside NVIDIA Groq 3 LPX Rack and Compute Tray, NVIDIA Vera CPU Rack Compute Tray and the NVIDIA HGX Rubin NVL8 platform, plus what it describes as full product support for NVIDIA Vera Rubin on the NVIDIA MGX architecture. Foxconn said Vera Rubin NVL72 targets frontier model training and large-scale AI inference and can reach up to 10x inference performance per watt versus NVIDIA's Blackwell platform through co-design, while the liquid-cooled Vera CPU Rack fits up to 256 Vera CPUs per rack, supports 22,500 sandboxes and delivers up to 2x performance for agentic AI, reinforcement learning and data processing workloads. FIT also showed 1.6T OSFP and 12.8T XPO optical interconnect platforms, a liquid-cooled busbar system rated up to 480kW and 400A high-current AC whips, and Foxconn said it is working with Intel on Intel Xeon-based rack-scale agentic AI solutions.

   科技新报 · 🔥 0
2. 4d
   Pegatron shows Vera Rubin NVL72 rack-scale AI platforms at OCP Global Summit 2026

   Pegatron said it will showcase next-generation AI infrastructure at OCP Global Summit 2026 under the theme "Scaling Innovation for the AI Era," stressing the path from system design and liquid cooling to volume manufacturing and rack-level validation. Its centerpiece is the RA4803-72N3 rack-scale platform, described as an NVIDIA Vera Rubin NVL72 system integrating 72 NVIDIA Rubin GPUs, 36 NVIDIA Vera CPUs and sixth-generation NVIDIA NVLink in a fully liquid-cooled rack, alongside a 2U liquid-cooled AS210-2T1-8H3 (NVIDIA HGX Rubin NVL8) and 48U rack configurations for up to 64 or 72 GPUs. Pegatron also showed an NVIDIA Vera CPU liquid-cooling system and a solution it calls NVIDIA Groq 3 LPX, plus AMD EPYC 9006-based AI-ready compute and storage servers, with workflows built on the NVIDIA DSX AI factory platform linking digital-twin simulation to factory deployment, according to the company.

   科技新报 · 🔥 0
3. Oct 9
   vLLM adds Vera Rubin NVL72 support, citing 7.8x throughput over GB200 NVL72

   The vLLM team, working with Inferact, Red Hat and NVIDIA, says vLLM now runs on NVIDIA Vera Rubin NVL72 with daily container builds (vllm/vllm-openai:cu134-nightly on CUDA 13.4 and PyTorch 2.15) and can serve models including DeepSeek, Kimi, GLM and MiniMax. Because Rubin continues the Blackwell architecture family (sm100f kernels can run unmodified), early results claim up to 7.84x throughput per GPU versus GB200 NVL72 on SemiAnalysis AgentX with MiniMax M3 at matched interactivity, and up to 3.7x VLM throughput versus GB300 NVL72 in MLPerf Inference v6.1 using Qwen3-VL-235B-A22B. Rubin NVL72 is described as offering 5x NVFP4 FLOPS, about 2.4x HBM bandwidth and 1.7x sixth-generation NVLink bandwidth over GB200 NVL72, while vLLM uses FlashInfer 0.7.0 Rubin-tuned kernels and CUDA 13.4 locality domains to speed up MoE decode by roughly 1.2x in preliminary tests. All figures are early, self-reported results from the collaborating teams.

   vLLM Blog · 🔥 0

Experience and discussion from the community

Share my NVIDIA Vera Rubin NVL72 experienceAsk about NVIDIA Vera Rubin NVL72