[KongchangAI]
Unverified50% confidenceReleaseExact time

NVIDIA在Hugging Face发布Qwen3 235B的NVFP4量化模型，将模型从BF16的约470GB显存压缩至约156GB，可运行在8张H100单节点集群上

1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/19/2026
First Seen
Valid until: 10/17/2026

Sources

Anthropic封号真相曝光：客户端检测逻辑与多款AI模型集中发布

Related Claims

UnverifiedNVIDIA H100单卡显存80GB，8卡节点640GB仍不足以加载Qwen3.8全量权重75% similarUnverifiedNVIDIA A100的显存带宽为2TB/s，加载280GB的FP32模型需要约140ms，仅权重加载就决定了每秒最多生成约7个token73% similarVerifiedNVIDIA A100单卡拥有80GB HBM2e显存和312 TFLOPS的FP16算力72% similar

Cite This Claim

Stable URI
https://kongchang.com/claim/561864
API
curl https://kongchang.com/api/v1/knowledge/claims/561864
MCP
get_claim(id=561864)