Expired50% confidenceBenchmarkExact time
DeepSeek reported selected dense GEMM results approaching the hardware upper limits they claimed, measured on Ascend 950D with CANN 9.20 using L2 profiling, but these are DeepSeek's own non-independently reproduced numbers.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/3/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 10/17/2026(expired)
Sources
Related Entities
Related Claims
UnverifiedDeepSeek and Huawei jointly optimized the Ascend 950 large-scale cluster solution, performing in-depth tuning for multi-card training and large model inference.78% similarUnverifiedThe six components have distinct roles: DeepGEMM accelerates general matrix operations, DeepEP handles large-scale cross-device communication, TileKernels provides common vector computation and memory access operators, FlashMLA provides sparse attention operators, and DeepSelect performs efficient data selection.75% similarUnverifiedThe interconnect bandwidth of DeepSeek operators on Huawei Ascend supernodes has approached the hardware limit.75% similarUnverifiedDeepSeek的编程智能体DeepSWE在SWE-bench基准测试中从预览版的12.8分提升至62.7分75% similarUnverifiedDeepSeek replicated top-tier model performance at extremely low compute costs.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/970601API
curl https://kongchang.com/api/v1/knowledge/claims/970601MCP
get_claim(id=970601)