Unverified50% confidenceFactExact time
张量并行（Tensor Parallelism）是将单个模型层的权重矩阵切分到多块 GPU 上并行计算，适用于模型参数超出单卡显存的场景
1
Sources
50%
Confidence
Long-term
Relevance
8/21/2026
First Seen
Sources
Related Claims
UnverifiedNVIDIA提出了非均匀张量并行（Nonuniform Tensor Parallelism）技术，用于在GPU故障时最大化LLM训练的有效产出（Goodput）73% similarUnverifiedAITER GEMM 和 FlyDSL FusedMoE 内核同时支持 Tensor Parallelism（TP）和 Data Parallelism + Expert Parallelism（DP+EP）两种并行策略71% similarUnverifiedNVIDIA's non-uniform tensor parallelism implementation introduces an Unbalanced GEMM Kernel, allowing different GPUs to perform matrix multiplications with different numbers of columns within the same layer.68% similarUnverifiedEfficient operation of Tensor Cores requires matrix dimensions to be aligned to multiples of 16 or 8.68% similarUnverifiedA mid-range GPU can achieve tens to hundreds of times the matrix computation throughput of a CPU68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/781305API
curl https://kongchang.com/api/v1/knowledge/claims/781305MCP
get_claim(id=781305)