50-image depth pass

Read 3D structure out of flat photos, the first step in relighting, parallax and 3D conversion work.

The pipeline

  1. Estimate depth on 50 images, depth-anything-v2-l

Fastest card where every stage is a real measurement: NVIDIA RTX PRO 6000 Blackwell Workstation Edition, 4 s for the whole job.

Every GPU, slowest job to fastest

GPUTotal timeComputeEnergyBasis
NVIDIA RTX PRO 6000 Blackwell Workstation Edition4 s2 s0.06 Whall 1 stages measured
NVIDIA B3005 s3 s0.18 Whall 1 stages measured
NVIDIA A100 80GB SXM45 s3 s0.08 Whall 1 stages measured
NVIDIA H100 80GB HBM35 s3 s0.11 Whall 1 stages measured
NVIDIA B2006 s3 s0.15 Whall 1 stages measured
NVIDIA L40S6 s3 s0.07 Whall 1 stages measured
NVIDIA H2006 s3 s0.1 Whall 1 stages measured
NVIDIA A10G7 s4 s0.06 Whall 1 stages measured
NVIDIA A100 40GB SXM410 s6 s0.12 Whall 1 stages measured
NVIDIA T413 s8 s0.1 Whall 1 stages measured
NVIDIA L417 s4 s0.03 Whall 1 stages measured

How these numbers are built

Each stage time is the quantity of work divided by that card's measured throughput for that model, from our own bench. The pipeline is assumed to run batched, every image, then every clip, so each model loads once. Model load time is added where we recorded it; our text-generation runs don't carry a load measurement yet, so pipelines with a language-model stage are slightly optimistic. Nothing here is a single timed run of the whole pipeline, and we don't present it as one.