NVIDIA

Latest AI news, models and releases from NVIDIA. ['A100', 'AI Aerial', 'AI platform', 'Alpamayo', 'B200', 'B300', 'BioNeMo', 'BioNeMo Agent Toolkit', 'Blackwell', 'Blackwell NVL72', 'Blackwell Ultra', 'BlueField-4', 'BlueField-4 DPU', 'BlueField-4 STX', 'Boltz-2', 'Canary-Qwen-2.5B', 'CMX', 'ConnectX-9', 'ConnectX-9 SuperNIC', 'Cosmos', 'Cosmos 3 Edge', 'Cosmos 3 Edge Policy (DROID)', 'Cosmos 3 Nano', 'Cosmos 3 Super', 'Cosmos 3 Super 4-Step Distillation', 'Cosmos-Dreams', 'Cosmos-H Dreams', 'Cosmos-H-Dreams', 'Cosmos-H-Surgical-Simulator', 'Cosmos-Predict2.5-2B', 'Cosmos-Reason2-2B', 'CUDA', 'DeepSeek-R1', 'DeepSeek V4', 'DeepSeek V4 Pro', 'DGX GB300', 'DGX Spark', 'DGX SuperPOD', 'DGX Vera Rubin NVL72', 'DPU BlueField-4']

Official site ↗

Research 🇺🇸

New DeepSeek-V3 technical report: how hardware-software co-design enables low-cost training of large models

A new technical paper from the DeepSeek team, with CEO Wenfeng Liang as co-author, explores hardware-aware model co-design to reduce LLM training costs. Using DeepSeek-V3 trained on 2048 NVIDIA H800 GPUs as a case study, the paper details innovations in memory efficiency (MLA), sparse computation (DeepSeekMoE), FP8 training, and interconnect-aware routing.

DeepSeek

NVIDIA

Alibaba/Qwen

NVIDIA Blackwell Ultra: 15 Pflops NVFP4, 288 GB HBM3e, no FP64

NVIDIA unveiled the Blackwell Ultra accelerator, featuring 208 billion transistors, 15 Pflops in NVFP4, 288 GB HBM3e memory, and new fifth-gen Tensor Cores. The chip omits FP64 but introduces NVFP4 format for efficient AI inference.

NVIDIA

Taiwan Semiconductor Manufacturing Company

ServerNews27.07 · 17:04

Show more ↓