Unverified50% confidenceBenchmarkExact time
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 TPS
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/28/2026
First Seen
Valid until: 12/27/2026
Sources
Related Entities
Related Claims
UnverifiedTPU v4单芯片峰值算力约275 TFLOPS(BF16精度),TPU v4-32 pod由32个芯片通过ICI高速互联组成,总算力约8800 TFLOPS71% similarUnverified华为昇腾910B在单卡算力上已接近A100水平,但在多卡互联效率上与NVIDIA存在显著差距71% similarUnverifiedDeepSeek V4 Flash的UD-Q8_K_XL量化版本文件体积为162GB,采用MXFP4+BF16混合精度,达到完全无损71% similarUnverified华为昇腾950超节点真机正式亮相,实现1024卡互联,算力达到百亿亿次量级70% similarUnverified在一台128G算力舱设备上,Qwen3.8 Flash的解码速度常态达241 TPS,峰值达274 TPS69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/956037API
curl https://kongchang.com/api/v1/knowledge/claims/956037MCP
get_claim(id=956037)