2025-08-20_h100-vs-gb200-nvl72-training-benchmarks

Canonical Target Counts

TargetOccurrences
04_knowledge_base/GB200 NVL7261
04_knowledge_base/GPU55
04_knowledge_base/H10049
02_companies/NVDA27
04_knowledge_base/Llama19
04_knowledge_base/Backplane19
04_knowledge_base/Mixture of Experts14
04_knowledge_base/BF1614
04_knowledge_base/GB20012
04_knowledge_base/FP811
04_knowledge_base/GPT-311
02_companies/Switch9
04_knowledge_base/Paladin connector9
02_companies/DeepSeek8
04_knowledge_base/Hyperscaler8
04_knowledge_base/Compute Tray8
03_industries/Neocloud產業7
04_knowledge_base/PyTorch7
04_knowledge_base/NVLink6
04_knowledge_base/Batch size5
04_knowledge_base/Sequence length5
04_knowledge_base/Hopper4
04_knowledge_base/Copper4
04_knowledge_base/Weak scaling4
04_knowledge_base/NVSwitch3
04_knowledge_base/TPU3
04_knowledge_base/Expert Parallelism2
04_knowledge_base/TorchTitan2
02_companies/META2
04_knowledge_base/H2002
04_knowledge_base/HGX2
04_knowledge_base/GCP2
04_knowledge_base/NVIDIA Collective Communications Library (NCCL)2
04_knowledge_base/Burn-in2
04_knowledge_base/Trainium2
04_knowledge_base/Pre-training2
04_knowledge_base/Data Parallelism2
04_knowledge_base/CUDA2
04_knowledge_base/CI-CD2
04_knowledge_base/NVIDIA B2002
04_knowledge_base/FSDP22
04_knowledge_base/Strong Scaling2
04_knowledge_base/Model FLOPS Utilization2
04_knowledge_base/Power Usage Effectiveness2
02_companies/GOOG1
04_knowledge_base/Arithmetic intensity1
04_knowledge_base/ASIC1
02_companies/AMD1
02_companies/APH1
04_knowledge_base/PCB1
04_knowledge_base/Mid-training1
04_knowledge_base/Blackwell1
04_knowledge_base/Scale-out networking1
04_knowledge_base/InfiniBand1
04_knowledge_base/液冷散熱1
04_knowledge_base/Tensor Parallelism1
04_knowledge_base/Flyover cable1
04_knowledge_base/Pipeline parallelism1
04_knowledge_base/Slurm1
04_knowledge_base/ClusterMAX1
04_knowledge_base/Reinforcement learning1
04_knowledge_base/CPU1
04_knowledge_base/Software Stack1
04_knowledge_base/Bianca Board1
04_knowledge_base/Chiller1
04_knowledge_base/DGX1