24-frame storyboard

Break a script into beats and draw the whole board, fast turbo diffusion, so it lands in minutes.

The pipeline

Break the script into 24 beats, qwen3-32b
Draw 24 storyboard frames, z-image-turbo

Fastest card where every stage is a real measurement: NVIDIA B200, 89 s for the whole job.

Every GPU, slowest job to fastest

GPU	Total time	Compute	Energy	Basis
NVIDIA B100	74 s	74 s	n/a	anchored estimate (0/2 stages measured)
NVIDIA GH200 Grace Hopper	80 s	80 s	n/a	anchored estimate (0/2 stages measured)
NVIDIA H100 NVL	81 s	81 s	n/a	anchored estimate (0/2 stages measured)
NVIDIA H800 80GB	84 s	84 s	n/a	anchored estimate (0/2 stages measured)
NVIDIA B200	89 s	62 s	12.85 Wh	all 2 stages measured
NVIDIA B300	1.5 min	53 s	11 Wh	all 2 stages measured
NVIDIA H100 PCIe	1.8 min	1.8 min	n/a	anchored estimate (0/2 stages measured)
NVIDIA H100 80GB HBM3	1.8 min	84 s	13.08 Wh	all 2 stages measured
NVIDIA RTX PRO 6000 Blackwell Server Edition	2.3 min	2 min	15.7 Wh	all 2 stages measured
NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition	2.3 min	2.3 min	n/a	anchored estimate (0/2 stages measured)
NVIDIA A800 80GB	2.9 min	2.9 min	n/a	anchored estimate (0/2 stages measured)
NVIDIA A100 40GB SXM4	3 min	2.9 min	2.08 Wh	anchored estimate (1/2 stages measured)
NVIDIA A100 40GB PCIe	3.2 min	3.2 min	n/a	anchored estimate (0/2 stages measured)
NVIDIA RTX PRO 5000 Blackwell	3.4 min	3.1 min	14.39 Wh	all 2 stages measured
NVIDIA RTX PRO 6000 Blackwell Workstation Edition	3.6 min	1.9 min	16.08 Wh	all 2 stages measured
NVIDIA A100 80GB PCIe	3.7 min	3.1 min	13.93 Wh	all 2 stages measured
NVIDIA A100 80GB SXM4	3.7 min	2.9 min	16.68 Wh	all 2 stages measured
NVIDIA H200	3.9 min	80 s	12.08 Wh	all 2 stages measured
NVIDIA L40S	4.1 min	3.6 min	18.66 Wh	all 2 stages measured
NVIDIA GeForce RTX 4090	5 min	3.8 min	24.32 Wh	all 2 stages measured
NVIDIA RTX 5000 Ada Generation	5.1 min	5 min	18.41 Wh	all 2 stages measured
NVIDIA RTX A5500	5.3 min	5.3 min	n/a	anchored estimate (0/2 stages measured)
NVIDIA RTX 6000 Ada Generation	5.5 min	5.3 min	25.57 Wh	all 2 stages measured
NVIDIA RTX 5880 Ada Generation	6.2 min	6.2 min	n/a	anchored estimate (0/2 stages measured)
NVIDIA RTX PRO 4500 Blackwell	6.2 min	4.3 min	13.21 Wh	all 2 stages measured
NVIDIA GeForce RTX 3090 Ti	6.3 min	6.1 min	37.71 Wh	all 2 stages measured
NVIDIA RTX PRO 4000 Blackwell	6.4 min	6.3 min	14.71 Wh	all 2 stages measured
NVIDIA RTX A5000	7 min	6.7 min	24.57 Wh	all 2 stages measured
NVIDIA L40	7 min	5 min	23.31 Wh	all 2 stages measured
NVIDIA RTX A6000	7.1 min	5.2 min	23.42 Wh	all 2 stages measured
NVIDIA GeForce RTX 3090	8.3 min	6.9 min	37.34 Wh	all 2 stages measured
NVIDIA A10G	9 min	8 min	19.54 Wh	all 2 stages measured
AMD Radeon Pro W7900	10.1 min	10.1 min	n/a	anchored estimate (0/2 stages measured)
NVIDIA A40	10.2 min	8.3 min	39.46 Wh	all 2 stages measured
AMD Radeon RX 7900 XTX	10.3 min	10.3 min	n/a	anchored estimate (0/2 stages measured)
AMD Radeon RX 7900 XT	10.6 min	10.6 min	n/a	anchored estimate (0/2 stages measured)
NVIDIA L4	12.2 min	11.6 min	13.69 Wh	all 2 stages measured
NVIDIA GeForce RTX 5090	12.5 min	2.6 min	21.46 Wh	all 2 stages measured
AMD Radeon Pro W7800	13.5 min	13.5 min	n/a	anchored estimate (0/2 stages measured)
NVIDIA RTX 4500 Ada Generation	15.2 min	15.2 min	n/a	anchored estimate (0/2 stages measured)
AMD Radeon Pro W6800	20.1 min	20.1 min	n/a	anchored estimate (0/2 stages measured)
NVIDIA Quadro RTX 8000	54.9 min	54.4 min	223 Wh	all 2 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.

AMD Radeon RX 6700, blocked at "Break the script into 24 beats" (Qwen3 32B)
AMD Radeon RX 6800, blocked at "Break the script into 24 beats" (Qwen3 32B)
AMD Radeon RX 6800 XT, blocked at "Break the script into 24 beats" (Qwen3 32B)
AMD Radeon RX 6900 XT, blocked at "Break the script into 24 beats" (Qwen3 32B)
AMD Radeon RX 6950 XT, blocked at "Break the script into 24 beats" (Qwen3 32B)
AMD Radeon RX 7600, blocked at "Break the script into 24 beats" (Qwen3 32B)
AMD Radeon RX 7700 XT, blocked at "Break the script into 24 beats" (Qwen3 32B)
AMD Radeon RX 7800 XT, blocked at "Break the script into 24 beats" (Qwen3 32B)
AMD Radeon RX 9070, blocked at "Break the script into 24 beats" (Qwen3 32B)
AMD Radeon RX 9070 XT, blocked at "Break the script into 24 beats" (Qwen3 32B)
GeForce GTX 1080 Ti, blocked at "Break the script into 24 beats" (Qwen3 32B)
GeForce RTX 4060, blocked at "Break the script into 24 beats" (Qwen3 32B)
GeForce RTX 4080 Super, blocked at "Break the script into 24 beats" (Qwen3 32B)
GeForce RTX 5060 Ti, blocked at "Break the script into 24 beats" (Qwen3 32B)
GeForce RTX 5070, blocked at "Break the script into 24 beats" (Qwen3 32B)
GeForce RTX 5070 Ti, blocked at "Break the script into 24 beats" (Qwen3 32B)
GeForce RTX 5080, blocked at "Break the script into 24 beats" (Qwen3 32B)
Intel Arc A750, blocked at "Break the script into 24 beats" (Qwen3 32B)
Intel Arc A770 Limited Edition, blocked at "Break the script into 24 beats" (Qwen3 32B)
Intel Arc B580, blocked at "Break the script into 24 beats" (Qwen3 32B)
Intel Arc Pro A60, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce GTX 1070 Ti, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce GTX 1080, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce GTX 1660, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce GTX 1660 Super, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce GTX 1660 Ti, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 2060, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 2060 Super, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 2070, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 2070 SUPER, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 2080 Founders Edition, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 2080 Super, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 2080 Ti Founders Edition, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 3050, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 3060, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 3060 Ti, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 3070 Founders Edition, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 3070 Ti, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 3080, blocked at "Break the script into 24 beats" (Qwen3 32B)
NVIDIA GeForce RTX 3080 Ti, blocked at "Break the script into 24 beats" (Qwen3 32B)

…and 18 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.