Unverified50% confidenceBenchmarkExact time
On the Huawei 950DT chip, DeepGEMM Ascend edition achieves a hardware utilization rate of 99.8% for BF16 multiplied by BF16.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/3/2026
First Seen
Valid until: 1/1/2027
Sources
Related Entities
Related Claims
UnverifiedThe 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.71% similarUnverifiedThe DeepGEMM-Ascend library supports BF16 (16-bit), FP8 (8-bit), and FP4 (4-bit) numeric formats.70% similarVerifiedLiang Wenfeng said Huawei gave DeepSeek 16,000 950 chips, equivalent to 4,000 B-series cards; 4 Huawei 950s match 1 GB300, lagging two years behind in time and costing 50% to twice as much.69% similarUnverifiedHuawei's Ascend 950 super-node physical machine has officially made its debut, achieving 1,024-card interconnection with computing power reaching the exaflops scale.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/970641API
curl https://kongchang.com/api/v1/knowledge/claims/970641MCP
get_claim(id=970641)