NVIDIA Releases Open Synthetic Data Generation Pipeline for Training Large Language Models
NVIDIA Releases Open Synthetic Data Generation Pipeline for Training Large Language Models Nemotron-4 340B, a family of models optimized for NVIDIA NeMo and NVIDIA TensorRT-LLM, includes cutting-edge instruct and reward models, and a dataset for generative AI training. NVIDIA today announced Nemotron-4 340B, a family of open models that developers can use to generate synthetic data for training la
nvidia-dockerãŒçªç„¶GPUã‚’èªè˜ã—ãªããªã‚‹å•é¡Œ
以下ã®è¨˜äº‹ã§å¯¾å‡¦æ³•ã«ã¤ã„ã¦çŸ¥ã£ãŸã€‚ ãªãŠã€ã“ã®å•é¡Œã«ã¤ã„ã¦ã¯nvidia-dockerãŒISSUEã®åŽŸå› ã¨å¯¾å‡¦æ³•ã«ã¤ã„ã¦æ–‡æ›¸ã‚’公開ã—ã¦ã„る。 åŽŸå› ã¯ã€ç°¡å˜ã«ã„ã†ã¨ã€cgroupã®ç®¡ç†ãŒsystemctlã¨dockerã¨ã§ç«¶åˆã—ã¦ã„ã‚‹ã¨ã„ã†ã“ã¨ã‚‰ã—ã„。cgroupã®ç®¡ç†ã‚’systemctl以外ã«ã™ã‚‹ã“ã¨ãŒãƒ¯ãƒ¼ã‚¯ã‚¢ãƒ©ã‚¦ãƒ³ãƒ‰ã«ãªã‚‹ã€‚ nvidia-dockerã®æ¬¡ã®pactch releaseã§è§£æ±ºã™ã‚‹äºˆå®šã¨ã®ã“ã¨ã€‚ 2024/08/17 追記: NVIDIA/nvidia-container-toolkitã«issueãŒç§»ã•ã‚ŒãŸã‚ã¨é€²æ—ãªã—。 環境
ã•ãらインターãƒãƒƒãƒˆã€AIåŠå°Žä½“1万個調é”ã¸ã€€å¾“æ¥è¨ˆç”»ã®5å€ - 日本経済新èž
ã•ãらインターãƒãƒƒãƒˆã¯19æ—¥ã€ç±³ã‚¨ãƒŒãƒ“ディアãªã©ã‹ã‚‰ç”»åƒå‡¦ç†åŠå°Žä½“（GPU）8000å€‹ã‚’è¿½åŠ èª¿é”ã™ã‚‹ã¨ç™ºè¡¨ã—ãŸã€‚人工知能（AI）ã®é–‹ç™ºã«ä½¿ã†ã‚¹ãƒ¼ãƒ‘ーコンピューターã®æ•´å‚™ã«å‘ã‘ã€GPUを組ã¿è¾¼ã‚“ã 高性能サーãƒãƒ¼ã‚’増やã™ã€‚従æ¥è¨ˆç”»ã¨åˆã‚ã›ã¦2027年末ã¾ã§ã«åˆè¨ˆ1万個ã®GPUを購入ã™ã‚‹ã€‚ã•ãらãƒãƒƒãƒˆã¯23年以é™ã€ã‚¨ãƒŒãƒ“ディアãªã©ã‹ã‚‰2000個ã®GPUã®èª¿é”を進ã‚ã¦ã„る。エヌビディア製ã®GPUã®èª¿é”
デル・テクノãƒã‚¸ãƒ¼ã‚ºã€ã€ŽNVIDIA GTCã€ã§ã€ŒDell PowerEdgeã€ã‚’発表 ï¼ AIã®å°Žå…¥ãƒ»æ´»ç”¨ã‚’åŠ é€Ÿ | Dell Technologies
ブãƒã‚°åŽŸç¨¿ã€€https://www.dell.com/en-us/blog/dell-poweredge-xe9680-ai-acceleration-announcements-at-nvidia-gtc/ ç†è€…： Robert McNeal | 2024å¹´3月18æ—¥ 『NVIDIA GPU Technology Conference（GTC）ã€ã¯ã€AIイノベーターã€AIデベãƒãƒƒãƒ‘ーã€AIã«é«˜ã„関心を寄ã›ã¦ã„ã‚‹ãŠå®¢æ§˜ã‚’対象ã«ã€NVIDIAãŒé–‹å‚¬ã™ã‚‹ã‚¤ãƒ™ãƒ³ãƒˆã§ã™ã€‚デル・テクノãƒã‚¸ãƒ¼ã‚ºã¯ã€AIインフラストラクãƒãƒ£ãƒ¼åˆ†é‡Žã«ãŠã‘るリーダーã®1社ã§ã‚ã‚Šã€NVIDIAã¨ã®ãƒ†ã‚¯ãƒŽãƒã‚¸ãƒ¼ コラボレーションをベースã¨ã™ã‚‹æœ€æ–°ã®æˆæžœã‚’åŒã‚¤ãƒ™ãƒ³ãƒˆã§ç´¹ä»‹ã—ã¾ã—ãŸã€‚ 展示会ブースã¨ã‚ªãƒ³ãƒ©ã‚¤ãƒ³ã®ãƒãƒ¼ãƒãƒ£ãƒ« セッションã§ã€ã€ŒDell Generative AI Solutions with NVIDIAã€ã‚„ã€ãŠå®¢æ§˜ã®
AIãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã«ãŠã‘ã‚‹GPUã¨å…‰ãƒ¢ã‚¸ãƒ¥ãƒ¼ãƒ«ã®æ¯”率ã¨éœ€è¦ã®åˆ†æž - Qiita
å¸‚å ´ã§ã¯ã€å…‰ãƒ¢ã‚¸ãƒ¥ãƒ¼ãƒ«ã¨GPUã®æ¯”率を計算ã™ã‚‹ãŸã‚ã®è¤‡æ•°ã®ã‚¢ãƒ—ãƒãƒ¼ãƒãŒå˜åœ¨ã—ã€ä¸€è²«æ€§ã®ãªã„çµæžœã‚’ã‚‚ãŸã‚‰ã—ã¦ã„ã¾ã™ã€‚ã“れらã®é•ã„ã®ä¸»ãªåŽŸå› ã¯ã€ã•ã¾ã–ã¾ãªãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯æ§‹é€ ã«å®Ÿè£…ã•ã‚Œã‚‹å…‰ãƒ¢ã‚¸ãƒ¥ãƒ¼ãƒ«ã®æ•°ã®å¤‰å‹•ã«ç”±æ¥ã—ã¦ã„ã¾ã™ã€‚å¿…è¦ãªå…‰ãƒ¢ã‚¸ãƒ¥ãƒ¼ãƒ«ã®æ£ç¢ºãªæ•°é‡ã¯ã€ã„ãã¤ã‹ã®é‡è¦ãªè¦å› ã«ä¸»ã«ä¾å˜ã—ã¦ã„ã¾ã™ã€‚ ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã‚«ãƒ¼ãƒ‰ãƒ¢ãƒ‡ãƒ« 主ã«2ã¤ã®ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã‚«ãƒ¼ãƒ‰ãŒå«ã¾ã‚Œã¦ãŠã‚Šã€ConnectX-6（200Gb/sã€ä¸»ã«A100ã¨ä½¿ç”¨ã•ã‚Œã‚‹ï¼‰ã¨ConnectX-7（400Gb/sã€ä¸»ã«H100ã¨ä½¿ç”¨ã•ã‚Œã‚‹ï¼‰ã§ã™ã€‚ åŒæ™‚ã«ã€æ¬¡ä¸–代ã®ConnectX-8 800Gb/sãŒ2024å¹´ã«ãƒªãƒªãƒ¼ã‚¹ã•ã‚Œã‚‹äºˆå®šã§ã™ã€‚ スイッãƒãƒ¢ãƒ‡ãƒ« 主ã«2種類ã®ã‚¹ã‚¤ãƒƒãƒãŒå«ã¾ã‚Œã¦ãŠã‚Šã€QM 9700スイッãƒï¼ˆ32ãƒãƒ¼ãƒˆOSFP 2x400Gb/s）ãŒã‚ã‚Šã¾ã™ã€‚åˆè¨ˆ64ãƒãƒ£ãƒ³ãƒãƒ«ã®400Gb/sã®è»¢é€é€Ÿåº¦ã¨ã€åˆè¨ˆ51.2Tb/sã®ã‚¹ãƒ«ãƒ¼
NVIDIA Mellanox ConnectX-7 ã® lspci ã§ã®è¦‹ãˆæ–¹ - Vengineerã®å¦„想
ã¯ã˜ã‚ã« NVIDIA Mellanox ConnectX-7 ã‚’ lspci ã§è¦‹ã‚‹ã¨ã©ã‚“ãªæ„Ÿã˜ã«ãªã£ã¦ã„ã‚‹ã‹ã‚’知るã“ã¨ãŒã§ãã¾ã—ãŸã€‚ Multifunction ã«è¦‹ãˆã‚‹ Googleå›ã«èžã„ãŸã‚‰ã€ NVIDIA ConnectX-7 Adapter Cards User Manual ãŒè¦‹ã¤ã‹ã£ãŸã€‚47é ã«æ¬¡ã®ã‚ˆã†ãªè¨˜è¼‰ãŒã‚ã£ãŸã€‚ Single-port PCIe x16 Card # lspci | grep mellanox -ia 3:00.0 Infiniband controller: Mellanox Technologes TM2910 Family [ConnectX-7] Dual-port PCIe x16 Card # lspci | grep mellanox -ia 86:00.0 Infiniband controller: Mellanox Techno
NVIDIA A100/H100 ã®L2 Cacheã«ã¤ã„ã¦ã€å†åº¦èª¿ã¹ãŸ - Vengineerã®å¦„想
ã¯ã˜ã‚ã« NVIDIA A100ã«ã¦ã€L2 Cacheã®æ§‹æˆãŒå¤‰ã‚ã£ãŸã“ã¨ã¯ã€ä¸‹è¨˜ã®ãƒ–ãƒã‚°ã§æ›¸ãã¾ã—ãŸã€‚ vengineer.hatenablog.com 今回ã¯ã€L2 Cache ã®ã‚µã‚¤ã‚ºãŒã€P100ã®4MBã€V100ã®6MBã‹ã‚‰ A100 ã§ã¯ 40MB (48MB)ã€H100 ã§ã¯ 50MB (60MB) ã«ãªã£ã¦ã€ãã®åˆ©ç”¨ã«ã¤ã„ã¦èª¿ã¹ã¦ã¿ã¾ã—ãŸã€‚ NVIDIA GA100 ã® L2 Cache A100 ã® L2 Cache ã¯ã€40MB (GA100ã§ã¯ 48MB ã§ã™ãŒã€A100 ã¨ã—ã¦ã¯ 40MB ã—ã‹ä½¿ãˆã¾ã›ã‚“) ã¨ã€V100 ã® 6MB ã‹ã‚‰ 大ãã増ãˆã¾ã—ãŸã€‚ å‰å›žã®ãƒ–ãƒã‚°ã§æ›¸ã„ãŸã‚ˆã†ã«ã€GA100 ã® L2 Cacheã¯2ã¤ã®ãƒ–ãƒãƒƒã‚¯ã«åˆ†å‰²ã•ã‚Œã€å„ブãƒãƒƒã‚¯ã¯ 20MB。20MB ã¯ã€512KB x 40 個ã¨ã„ã†æ§‹æˆã«ãªã£ã¦ã„ã¾ã™ã€‚ GA100ã¯ã€6個ã®HBM2e ãŒ
NVIDIA’s New Ethernet Networking Platform for AI Available Soon From Dell Technologies, Hewlett Packard Enterprise, Lenovo
NVIDIA’s New Ethernet Networking Platform for AI Available Soon From Dell Technologies, Hewlett Packard Enterprise, Lenovo End-to-End Platform Features Latest NVIDIA Spectrum-X Networking, Provides Foundation for Customers to Transform Business With AI NVIDIA today announced that Dell Technologies, Hewlett Packard Enterprise and Lenovo will be the first to integrate NVIDIA Spectrum-X™ Ethernet net
GitHub - NVIDIA/TensorRT-LLM: TensorRT-LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and build TensorRT engines that contain state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorR
TensorRT-LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and build TensorRT engines that contain state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT-LLM also contains components to create Python and C++ runtimes that execute those TensorRT engines. It also includes a backend for integration with the NVIDIA Triton Inferen
リリースã€éšœå®³æƒ…å ±ãªã©ã®ã‚µãƒ¼ãƒ“スã®ãŠçŸ¥ã‚‰ã›
最新ã®äººæ°—エントリーã®é…ä¿¡
j次ã®ãƒ–ックマーク
kå‰ã®ãƒ–ックマーク
lã‚ã¨ã§èªã‚€
eコメント一覧を開ã
oページを開ã
Data Center Best Practices with DGX B200
GitHub - NVIDIA/cuda-samples: Samples for CUDA Developers which demonstrates features in CUDA Toolkit
https://www.t4.gsic.titech.ac.jp/sites/default/files/2024-06/TITECH_DLHackathon_NV_202406.pdf
NVIDIA Performance Libraries (NVPL)
GitHub - NVIDIA/cuda-checkpoint: CUDA checkpoint and restore utility
Nvidia-smi
Nvidia’s Blackwell Offers FP4, Second-Gen Transformer Engine - EE Times
nvidia-blackwell-architecture-technical-brief.pdf
GitHub - NVIDIA/NeMo-Curator: Scalable data pre processing and curation toolkit for LLMs
A Look at the Liquid Cooled Supermicro SYS-821GE-TNHR 8x NVIDIA H100 AI Server
https://s201.q4cdn.com/141608511/files/doc_presentations/2023/Oct/NVIDIA-Oct-2023-Investor-Presentation.pdf
{{#tags}}- {{label}}
{{/tags}}