Nvidia's GB300 GPU shows 2.7x faster inference than GB200

Nvidia's GB300 ultra NVL72 has demonstrated a 2.7x speed advantage over the GB200 NVL72 in inference tasks using the vLLM project's engine. This performance leap exceeds theoretical expectations based on the GB300's specifications, which include a 1.5x increase in NVFP4 FLOPs and HBM capacity, alongside identical HBM bandwidth compared to the GB200. AI

影响 This hardware advancement could accelerate AI model training and inference, potentially lowering costs and enabling more complex models.

排序理由 Announcement of a new hardware product (GB300 ultra NVL72) with significant performance improvements over its predecessor. [lever_c_demoted from significant: ic=1 ai=0.7]

在 X — SemiAnalysis 阅读 →

• GB200
• GB300
• Nvidia
• NVL72
• vLLM

• 基础设施
• 产品

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →