List your company

Training Throughput on Qwen3-30B-A3B workload

280,000Throughput (tokens/s)PithTrain
82,920134,085185,250236,415May 29, 2026

Evaluation results

MethodLinks
PithTrainHardware=4×8-H100, Parallelism=PP4,DP1,CP1,EP8, SeqLen=4096, Precision=BF162026.05280,000
Megatron-LMHardware=4×8-H100, Parallelism=PP4,DP1,CP1,EP8, SeqLen=4096, Precision=BF162026.05264,100
PithTrainHardware=1×8-B200, Parallelism=PP2,DP1,CP1,EP4, SeqLen=8192, Precision=FP82026.05134,500
Megatron-LMHardware=2×8-H100, Parallelism=PP2,DP1,CP1,EP8, SeqLen=2048, Precision=BF162026.05126,700
PithTrainHardware=2×8-H100, Parallelism=PP2,DP1,CP1,EP8, SeqLen=2048, Precision=BF162026.05124,900
Megatron-LMHardware=1×8-B200, Parallelism=PP2,DP1,CP1,EP4, SeqLen=8192, Precision=FP82026.05106,200
TorchTitanHardware=2×8-H100, Parallelism=PP2,DP1,CP1,EP8, SeqLen=2048, Precision=BF162026.0590,500