Optimization step takes 5+ minutes on H200 GPU

[devtony01]

The animation optimization step consistently takes 5+ minutes on an H200 GPU, even with reduced settings.

Environment

• GPU: H200
• Modal pipeline
• optimization.py

Settings

• imgSize: 720 (also tested 960)
• fps: 10
• optimizer.steps: 100
• optimizer.maxFrames: 60
• lossWeights: depth=0, opticalFlow=0

Observed
Optimization step still takes 5+ minutes per job.

Expected
Significantly faster runtime at these settings.

Logs

INFO:autorig.runner:run cmd=/usr/local/bin/python optimization.py
--save_path /tmp/autorig/local-test/results/animation --iter 100
--input_path /tmp/autorig/local-test/inputs --seq_name local-test --save_name
autorig --img_fps 10 --flow_wt 0.0 --depth_wt 0.0 --mask_wt 1.0
--proj_joint_wt 1.0 --proj_vert_wt 1.0 --smooth_weight 0.25 --device cuda:0
--img_size 720