Unverified50% confidenceBenchmarkExact time
FlashMLA's prefill throughput for the sparse attention operator on the Ascend 950 reaches 410 TFLOPS (95% of the hardware's theoretical peak), and decode reaches 360 TFLOPS (83%), both figures as published by DeepSeek itself with no third-party measurements.
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 VerifiedOn a dual-Spark system, DeepSeek V4 Flash (FP8 quantization) peaks at approximately 53 tokens per second.72% similarUnverifiedOn a task with one million input tokens, thirty thousand output tokens, and a 99% cache hit rate, MiMo-V2.6 Flash costs less than half of Qwen 3.8 Flash.71% similarUnverifiedDeepSeek published a sparse prefill figure of 1350 TFLOPS (absolute value) for the NVIDIA B200, which uses a different metric from the 95% peak utilization figure cited for Ascend.71% similarUnverifiedThe Huawei Ascend 950PR successfully ran a pruned version of the DeepSeek V4-Flash model in a single-card configuration, achieving a peak decode speed of 208 TPS.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/963640API
curl https://kongchang.com/api/v1/knowledge/claims/963640MCP
get_claim(id=963640)