딥시크 추론 프레임워크 완전 오픈소스화, 엔비디아·화웨이 아센드 호환, 일일 155T 토큰 처리량, 카드당 시간당 38달러 수익

36Kr · 2026.10.20 22:43 · 원문 사이트

🇰🇷 한글 번역

DeepSeek, 추론 프레임워크를 완전한 오픈소스로 공개: NVIDIA 및 화웨이 어센드 호환, 단일 클러스터 일일 처리량 155T 토큰, 카드당 시간당 수익 38달러 36크루

📄 원문 (English)

DeepSeek Fully Open-Sources Inference Framework: NVIDIA & Huawei Ascend Compatible, 155T Tokens Daily Single-Cluster Throughput, $38 Per Card Per Hour Revenue    36Kr

▶ 칩 라이브 24 홈으로 — chiplive24.com

다른 반도체 뉴스

Micron Is Quietly Building a Financial Fortress. Here's Why Investors Should Care. · Yahoo Finance Chinese Tech Giants Order $35 Billion In Compliant Nvidia Chips · Yahoo Finance Nvidia Stock Is Near Its High, but the Market Is Wobbling. Should Investors Sell? · Yahoo Finance AMD Stock Price Prediction: Is the AI Chip Opportunity Bigger Than Wall Street Thinks? · Yahoo Finance AMD’s 12-Core Ryzen CPU With A Single 12-Core “Zen 6” CCD Could Match The Ryzen 9 9950X3D2 “Zen 5” Chip In Multi-Threaded Performance, Says Hydra Tool Creator · Wccftech Mysterious AMD Zen 6 APU with 24GB of LPDDR5 memory appears in benchmarks — chip may be for next-gen Steam Deck or gaming handhelds · Tom's Hardware Mysterious AMD Zen 6 APU with 24GB of LPDDR5 memory appears in benchmarks — chip may be for next-gen Steam Deck or gaming handhelds · Tom's Hardware Stocks making big moves yesterday: Datadog, SolarEdge, AMD, Western Digital, and Amazon · Yahoo Finance