Skip to content

[Perf] -- RTX PRO 6000 performance table + --stats-path gap for interactive-drive OmniDreams #592

Description

@jmataiNV
  1. Add a documented, reproducible RTX PRO 6000 single-view interactive-drive performance table

docs/source/models/omnidreams.rst currently has a GB300-only performance table with no documented measurement methodology — no PyTorch/CUDA/driver version, no sample count, no warmup/steady-state split (unlike, e.g., the Waypoint model page's table, which documents all of that). Worth adding an equivalent, properly-documented table for RTX PRO 6000, run through flashdreams-benchmark for real environment capture and multiple samples. For example, we have GB300 numbers from here: https://flashdreams.org/main/models/omnidreams.html#performance-table

Image
  1. --stats-path produces empty samples (sample_count: 0)

Running interactive-drive-omnidreams-optimized-rtx-pro-6000 with --stats-path and --world-model-profile together writes correct frame_count per step but sample_count: 0 for every step — no timing data is actually captured there, even though --world-model-profile clearly does measure and log per-stage timing separately. The two mechanisms appear disconnected: the structured --stats-path JSON isn't being fed from the profiling data.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions