Test Page HTML

WSI AI Inference Race

H200 vs MI300X vs Gaudi 3 vs BEN-6 Plus + OPU

Model: Tokens:

BenCell Sparsity: 70% Tensor Rank: 25%

Ready — press RUN INFERENCE

H200 NVIDIA

— tok/s

MI300X AMD

— tok/s

Gaudi 3 INTEL

— tok/s

BEN-6 + OPU

— tok/s

Token output stream:

System Power (Watts) — Lower is Better

700W

H200

750W

MI300X

600W

Gaudi 3

115W

BEN-6+OPU

WSI uses 6.1x less power than H200 · 6.5x less than MI300X