Concept art straight into 3D meshes, the pipeline that only runs on serious VRAM.
Fastest card where every stage is a real measurement: NVIDIA H100 80GB HBM3, 2.2 min for the whole job.
| GPU | Total time | Compute | Energy | Basis |
|---|---|---|---|---|
| NVIDIA H100 80GB HBM3 | 2.2 min | 2 min | 12.69 Wh | all 2 stages measured |
| NVIDIA H200 | 2.4 min | 2 min | 13.15 Wh | all 2 stages measured |
| NVIDIA L40S | 2.9 min | 2.7 min | 13.29 Wh | all 2 stages measured |
| NVIDIA A100 80GB SXM4 | 4 min | 3.7 min | 16.75 Wh | all 2 stages measured |
| NVIDIA A10G | 7.6 min | 7.2 min | 21.16 Wh | all 2 stages measured |
| NVIDIA L4 | 8.3 min | 8.1 min | 9.43 Wh | all 2 stages measured |
Every one of these fails on the same kind of wall, a stage that will not fit in VRAM.
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.