100-photo restoration batch

A shoebox of old family photos, repaired in one unattended run.

The pipeline

  1. Restore and retouch 100 photos, flux-1-kontext-dev

Fastest card where every stage is a real measurement: NVIDIA B300, 10.2 min for the whole job.

Every GPU, slowest job to fastest

GPUTotal timeComputeEnergyBasis
NVIDIA B30010.2 min9.7 min166.12 Whall 1 stages measured
NVIDIA B20017.8 min17.5 min257.65 Whall 1 stages measured
NVIDIA B10021.8 min21.8 minn/aanchored estimate (0/1 stages measured)
NVIDIA GH200 Grace Hopper22.8 min22.8 minn/aanchored estimate (0/1 stages measured)
NVIDIA H20023.3 min22.8 min261.51 Whall 1 stages measured
NVIDIA H100 NVL23.5 min23.5 minn/aanchored estimate (0/1 stages measured)
NVIDIA H800 80GB23.5 min23.5 minn/aanchored estimate (0/1 stages measured)
NVIDIA H100 80GB HBM323.8 min23.5 min267 Whall 1 stages measured
NVIDIA H100 PCIe27.1 min27.1 minn/aanchored estimate (0/1 stages measured)
NVIDIA RTX PRO 6000 Blackwell Server Edition33.4 min33.1 min301.72 Whall 1 stages measured
NVIDIA RTX PRO 6000 Blackwell Workstation Edition33.4 min32.4 min322.75 Whall 1 stages measured
NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition41.7 min41.7 minn/aanchored estimate (0/1 stages measured)
NVIDIA A100 40GB SXM450.2 min50.2 minn/aanchored estimate (0/1 stages measured)
NVIDIA A800 80GB50.2 min50.2 minn/aanchored estimate (0/1 stages measured)
NVIDIA A100 80GB SXM450.8 min50.2 min330.66 Whall 1 stages measured
NVIDIA A100 40GB PCIe54.3 min54.3 minn/aanchored estimate (0/1 stages measured)
NVIDIA A100 80GB PCIe54.7 min54.3 min268.4 Whall 1 stages measured
NVIDIA RTX PRO 5000 Blackwell55.9 min55.6 min277.59 Whall 1 stages measured
NVIDIA L40S59.6 min59.1 min334.51 Whall 1 stages measured
NVIDIA L401 h 30 min89.8 min448.23 Whall 1 stages measured
NVIDIA RTX A60001 h 37 min1 h 35 min470.32 Whall 1 stages measured
NVIDIA RTX 6000 Ada Generation1 h 37 min1 h 37 min482.35 Whall 1 stages measured
NVIDIA RTX PRO 4500 Blackwell1 h 54 min1 h 54 min286.12 Whall 1 stages measured
NVIDIA RTX 5000 Ada Generation2 h 6 min2 h 6 min374.53 Whall 1 stages measured
NVIDIA RTX 5880 Ada Generation2 h 26 min2 h 26 minn/aanchored estimate (0/1 stages measured)
AMD Radeon Pro W79003 h 15 min3 h 15 minn/aanchored estimate (0/1 stages measured)
AMD Radeon Pro W78004 h 52 min4 h 52 minn/aanchored estimate (0/1 stages measured)
AMD Radeon Pro W68005 h 12 min5 h 12 minn/aanchored estimate (0/1 stages measured)

Cards that can't run this pipeline

Every one of these fails on the same kind of wall, a stage that will not fit in VRAM.

…and 31 more.

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.