Intelligence

Compute efficiency by workload · MLPerf benchmarks × market pricing · 2026-09-01

New
GPU Performance Evolution
The same GPUs get faster over time from software alone — see it charted across every MLPerf round, in tok/s and in real dollars per token.
→
LLM Inference
Generative text inference — tokens produced per second per GPU.
6,939,963
tok/$ · best on-demand
Lambda Labs · NVIDIA H200-SXM-141GB
NLP
Extractive language understanding — queries answered per second per GPU.
14,500,415
samp/$ · best on-demand
Lambda Labs · NVIDIA H200-SXM-141GB
Vision
Image classification throughput — images classified per second per GPU.
150,963,210
samp/$ · best on-demand
Lambda Labs · NVIDIA H200-SXM-141GB
Recommendation
Recommendation inference — user-item scoring queries per second per GPU.
125,241,878
samp/$ · best on-demand
Lambda Labs · NVIDIA H200-SXM-141GB
Medical Imaging
3D medical image segmentation — volumes processed per second per GPU.
10,752
samp/$ · best on-demand
Lambda Labs · NVIDIA H200-SXM-141GB