Skip to content
DimensionLab

DimensionLab

Case studies

Measured results from physics models trained, evaluated and deployed with Frontier: the data, the method and the numbers, including where the models fall short.

  • Evaluating NVIDIA's pretrained GeoTransolver on all 48 held-out DrivAerML cars

    A Frontier study that ran NVIDIA's published GeoTransolver checkpoint against every held-out DrivAerML design, with hashes for the model, data and split, for about $13 of cloud GPU time. Drag R² 0.995, mean drag error 0.84%.

    held-out car designs evaluated
    48 / 48
    drag force R²
    0.995
    mean relative drag error
    0.84%
    single H100 pod, end to end
    3.3 h
  • Reading only what you need: bounded scientific data reads and 134-million-voxel volumes on an 8 GB GPU

    Frontier studies of its data layer: ArrayRoute's bounded reads on real ERA5 data, million-sample dataset previews in milliseconds, 75% fewer bytes for overlapping readers, and NanoVDB volumes at 10⁸ nodes in NVIDIA Omniverse.

    faster station read on real ERA5 (124.8 → 1.47 ms)
    85×
    preview of a 1,000,000-sample dataset
    3.7 ms
    fewer bytes for four overlapping readers
    75%
    to open a 512³ (134 M-node) volume
    1.1 s
  • From 21% to 4.9% surface-pressure error on AhmedML, for $15.54 of cloud GPU

    A Frontier study that ingested all 500 AhmedML simulations, trained reference surrogates on rented GPUs, and certified them against a benchmark with calibrated drag intervals. Test pressure error fell from 0.213 to 0.049 relative L2.

    test surface-pressure rel-L2 (was 0.213)
    0.049
    surface-pressure R² on 50 test designs
    0.996
    conformal Cd interval coverage (45/50)
    90%
    total RunPod spend
    $15.54
  • One verified package, every target: TensorRT at 1.8× PyTorch speed, a 1.7 MB edge model, and an FMI co-simulation unit

    A Frontier study that took trained CFD surrogates from checkpoint to TensorRT, ONNX, edge CPU, an HTTP service and an FMI 2.0 FMU, with each export's parity and accuracy measured before it was stored. Total cloud cost: $0.44.

    TensorRT FP32 vs PyTorch at 131k points
    1.8×
    TensorRT FP32 parity (relative L2)
    3e-7
    faster on one ARM core with dynamic INT8
    4.3×
    total RunPod spend
    $0.44

Earlier work

These projects predate Frontier. They are kept for reference and are not related to any product or service we currently offer.