Play new and move-first

  • 홈
  • 태그
  • 방명록

fluctuation 2

AI Datacenter 에 필요한 Load(부하) 특성 _ Part 6_(LLM Inference)

Part 4 LLM TrainingPart 5 FinetuningPart 6 LLM Inference Part 6에서는 LLM Inference 시의 부하(Load) 변동에 대해서 살펴보도록 하겠습니다.Case 3. LLM Inference Power Consumption: GPT-2 vs. nanoGPT대상 모델: GPT-2, nanoGPT실험 조건: GPU 기반 LLM 추론(inference) 과정의 전력 소모 패턴 분석전력 소모 패턴의 주요 특징1) 급격한 전력 변동공통점: 두 모델 모두 추론 과정에서 저전력(Idle) 상태와 고전력(Active) 상태를 빠르게 오가며, 전력 소모가 급격히 변화함.- 전력 피크: 약 300W까지 순간적으로 상승.- 피크 지속 시간: 25~50초, 표준편차 약 50..

Data Center/Data Center 2025.07.18

AI Datacenter 에 필요한 Load(부하) 특성 _ Part 3

Total Data Center Power Consumption(MIT Supercloud Dataset 예시)시스템 구성 및 Data 는 아래와 같이 구성되어 있으며 BERT*) 라는 모델을 사용하였을때 부하의 변동을 정리한 내용입니다. System Overview : TXGaia Cluster at MIT’s datacenter. 224 GPU-accelerated nodes, each with: - Two 20-core Intel Xeon Gold 6248 CPUs (384GB RAM). - Two NVIDIA Volta V100 GPUs (32GB RAM each). - Additional CPU-only nodes.Dataset Composition (~2.1 TB) Time series ..

Data Center/Data Center 2025.07.07
이전
1
다음
더보기
프로필사진

Play new and move-first

  • 분류 전체보기 (181)
    • Data Center (12)
      • Data Center (12)
    • 철학 (0)
      • 철학 (0)
    • AI Application (12)
      • AI 활용(생산성) (12)
    • 경제 (9)

Tag

chat, load, 티스토리챌린지, 데이터센터, AI, prompt, AWP, 오블완, 전력, transformer, data center, GPT, ChatGPT, Datacenter, POWER, 주식, construction, LLM, 프롬프트, 생산성,

최근글과 인기글

  • 최근글
  • 인기글

최근댓글

공지사항

페이스북 트위터 플러그인

  • Facebook
  • Twitter

Archives

Calendar

«   2026/01   »
일 월 화 수 목 금 토
1 2 3
4 5 6 7 8 9 10
11 12 13 14 15 16 17
18 19 20 21 22 23 24
25 26 27 28 29 30 31

방문자수Total

  • Today :
  • Yesterday :

Copyright © AXZ Corp. All rights reserved.

티스토리툴바