Unverified50% confidenceBenchmarkExact time
FlashMLA在昇腾950上稀疏注意力算子的预填充吞吐做到410 TFLOPS（硬件理论峰值的95%），解码360 TFLOPS（占83%），均为DeepSeek自己公布，无第三方实测
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/1/2026
First Seen
Valid until: 12/30/2026
Sources
Related Entities
Related Claims
Unverified盘古2.0 Flash在LiveCodeBench算法推理测试中得分85.1，仅次于DeepSeek V4 Flash的91.6，超过千问3.6 27B的83.974% similarPartially Verified在双Spark系统上，DeepSeek V4 Flash（FP8量化）峰值约每秒53个Token72% similarUnverified在百万输入、三万输出、99%缓存命中的任务下，MiMo-V2.6 Flash 花费不到千问 3.8 Flash 的一半71% similarUnverifiedDeepSeek公布英伟达B200稀疏预填充数字为1350 TFLOPS（绝对值），与昇腾的95%峰值比例口径不同71% similarUnverified华为昇腾950PR在单卡配置下成功运行了裁剪版的DeepSeek V4-Flash模型，解码峰值达到208 TPS70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/963640API
curl https://kongchang.com/api/v1/knowledge/claims/963640MCP
get_claim(id=963640)